<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Evan Zamir on Medium]]></title>
        <description><![CDATA[Stories by Evan Zamir on Medium]]></description>
        <link>https://medium.com/@thecity2?source=rss-949f03393eeb------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*omTVgt60nt5_3Sl3GUTfyA.jpeg</url>
            <title>Stories by Evan Zamir on Medium</title>
            <link>https://medium.com/@thecity2?source=rss-949f03393eeb------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 21:12:17 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@thecity2/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Bodybuilding Mnemonic]]></title>
            <link>https://thecity2.medium.com/bodybuilding-mnemonic-186f6a7f09fe?source=rss-949f03393eeb------2</link>
            <guid isPermaLink="false">https://medium.com/p/186f6a7f09fe</guid>
            <category><![CDATA[fitness]]></category>
            <category><![CDATA[weight-loss]]></category>
            <category><![CDATA[bodybuilding]]></category>
            <category><![CDATA[gym]]></category>
            <dc:creator><![CDATA[Evan Zamir]]></dc:creator>
            <pubDate>Sat, 15 Jun 2024 13:17:25 GMT</pubDate>
            <atom:updated>2024-06-15T15:16:42.939Z</atom:updated>
            <content:encoded><![CDATA[<p>Remember It</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qGbQVKuYa9R5qFPbjn1_ng.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@scottwebb?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Scott Webb</a> on <a href="https://unsplash.com/photos/woman-wearing-black-top-top-holding-black-dumbbells-standing-in-front-of-mirror-Vn39uEkX00s?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Unsplash</a></figcaption></figure><blockquote>Aesthetics is Genetics</blockquote><blockquote>We Strive for Size</blockquote><blockquote>Nutrition is the Condition</blockquote><h3><strong>Aesthetics is Genetics</strong></h3><p>For the most part your aesthetics is determined by genetics. You can’t change your hips, your insertions, your clavicle length, etc. These are all genetically determined and to a very large extent determine your ultimate aesthetic potential. This is what separates 99.9% of good looking gym bros from IFBB pro bodybuilders. You should give zero fucks worrying about your aesthetics because it is out of your control. You can only be the best version of you. To be that person focus on the next two things…</p><h3>We Strive for Size</h3><p>One of the things you can exert control over is how big your muscles get. You can’t change their shape, but you can change their size. That is why you go into the gym and lift heavy shit up and down, say it with me fellas, PROGRESSIVE OVERLOAD. I am characterizing this exertion of effort “STRIVE”. You cannot grow muscle without the Strive.</p><h3>Nutrition is the Condition</h3><p>When I say “Condition” I’m talking about your body fat level or leanness plus your overall health. This you can control through your diet and nutrition. This is where you decide to go into a deficit to get lean, go on a lean bulk to slowly add muscle without getting fat, or perhaps, do a dirty bulk when your 16 and skinny fat and it doesn’t matter how you eat you just want to get jacked as fast as possible and you’re not concerned about getting fat because you don’t even know how.</p><p>Now, of course, these aren’t exactly mutually exclusive categories. For example, if you eat like shit and get fat your aesthetics goes to shit. Same for lack of effort. Conversely if you Strive and adhere to a great diet you can improve your aesthetics. But by and large I’d say that 99% of the concerns and rules of thumb we come across in bodybuilding are summarized by this pithy little poem I wrote. So when you ask a question about your clavicles or whether you should cut or bulk or what should I work on or what your body fat is…remember it!</p><blockquote>Aesthetics is Genetics</blockquote><blockquote>We Strive for Size</blockquote><blockquote>Nutrition is Condition</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=186f6a7f09fe" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Still tracking calories and it’s still working]]></title>
            <link>https://thecity2.medium.com/still-tracking-calories-and-its-still-working-27162355ab6a?source=rss-949f03393eeb------2</link>
            <guid isPermaLink="false">https://medium.com/p/27162355ab6a</guid>
            <category><![CDATA[bodybuilding]]></category>
            <category><![CDATA[gym]]></category>
            <category><![CDATA[weight-loss]]></category>
            <category><![CDATA[fitness]]></category>
            <category><![CDATA[weight-loss-tips]]></category>
            <dc:creator><![CDATA[Evan Zamir]]></dc:creator>
            <pubDate>Mon, 25 Mar 2024 21:22:50 GMT</pubDate>
            <atom:updated>2026-06-27T23:57:09.186Z</atom:updated>
            <content:encoded><![CDATA[<p>When I previously left you I was only a couple months into my <a href="https://medium.com/@thecity2/the-best-kept-secret-for-a-successful-diet-plan-counting-calories-actually-works-40053ec84375">weight loss “journey”</a> (as the kids call it these days). At that point I was down about 14 pounds and in a January update at the end of that piece I said:</p><blockquote>I did hit my original target weight but decided to keep going and am now currently around 165 lbs. My new goal is 155 lbs, which I haven’t hit since way back in 2002…If all goes as planned I should hit my goal some time in March. To be continued…</blockquote><p>Well here we are nearing the end of March and I am right at 155 lbs. If I recall correctly when I started this diet (ok, I should call it a <em>cut</em> by the modern social media parlance) my target weight was 175 (a nice 20 pound chunk of weight loss), and I thought that was being optimistic. I have shifted my goal downwards multiple times in the last several months because I’ve realized the cut wasn’t nearly as difficult as I had anticipated and that obtaining “shredded” status is possible for me even as I’m nearing 50 (and when I say nearing — I’m 48, there I said it). I haven’t quite got there yet. You can see from the picture that my upper abs are now quite visible but the lower abs are still hidden by fat. I think another 10 lbs of cutting down to 145 (the lightest by far I will have ever been in my adult life) might get me that mythical 6-pack right in time for summer. By my calculations (based on various internet calculators) I should be roughly 12–14% body fat at 145 lbs. I’m probably somewhere around 17–19% currently.</p><p>Of course this is purely an aesthetic pursuit the rest of the way. After recently having a physical for the first time in half a decade, all my labs came back entirely in the normal range. So I know I’m physically healthy. I don’t need to keep cutting “just to see my abs”. I could just go back to maintenance mode. Eventually I will obviously have to do that. But pushing myself physically like this has actually been a blast. So putting aside motivation for now, if you’ve read this far, I have a hunch you’re interested enough to know how I did it because you are either thinking about starting on a major cutting phase or are in the midst of one already. So let’s get into it. First, here’s a look at my journey so far via a daily chart of my weight loss:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E-IioldJEr-8CXlqY03Hmw.png" /><figcaption>Daily weight chart</figcaption></figure><p>This chart is actually super important for several reasons. First, I want to emphasize that, at least for me, it is critical that I track my weight on a daily basis. To minimize day-to-day fluctuation (which you can see is still significant and much larger than any <em>actual</em> weight that is gained or lost in a day), I always weigh myself first thing in the morning (nude!) after I pee. You don’t have to do this, but I would encourage it as it really is likely to be the most consistent measurement. Alternatively, you can weight yourself (nude!) right before you go to sleep. If you have never done either consistently, <a href="https://www.npr.org/sections/krulwich/2013/06/19/193556929/every-night-you-lose-more-than-a-pound-while-youre-asleep-for-the-oddest-reason">one fun fact I have learned is that we humans lose about 1–2 lbs during sleep</a>. Apparently, this is mostly lost through water vapor in your breath. Or something like that. In addition to minimizing variance, or actually related to that, getting used to weighing yourself is important for maintaining a sense of control over your weight. Some people are actually afraid of weighing themselves and seeing these fluctuations. Or they feel guilty. Or whatever. Again, at least for me, I found that I need to confront my weight head on. I need to be realistic about it, and I need to be accountable for it to myself. Now having seen months and months of weight fluctuation, it doesn’t scare me. I know if my weight randomly goes up by a pound one morning, it’s not because my diet failed. It’s almost certainly just water retention, or the fact that I ate a ton of broccolli the night before. Or maybe I had a couple drinks (we’ll get back to this point later). Whatever the reason may be, I know that I haven’t actually ever gained or lost more than a pound of fat (and/or muscle) in one day. At most, you gain or lose a few ounces.</p><p>How do I know this? Well now that gets to the meat of this article. I track my calories and I control my diet — and more specifically, I control my calorie deficit to be around 500–600 calories per day. Since we know that (roughly speaking) losing a pound of fat requires burning 3500 calories, this amount of energy deficit results in roughly 1 pound of weight loss per week. There are (obviously) 16 ounces in a pound, so that’s roughly 2.3 ounces per day. Therefore, whatever day-to-day fluctuations you see in your weight will pretty much always swamp out the <em>actual</em> weight loss (or gain) in one day. You truly have to just trust the process. It also helps — and I would advise strongly — to track your median weekly weight loss. For me that chart looks like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BiZlNcwQDCMMcBMagKSsuQ.png" /><figcaption>Weekly median weight chart</figcaption></figure><p>From the weekly chart you see the week-to-week variance is much smaller than the daily chart and in the vast majority of weeks the weight is indeed decreasing by about a pound, if not slightly more. Some weeks are flat and a few times the weight goes up, but those weeks are pretty rare and by now I’m used to it. In fact I almost have come to expect it by now that after a few weeks of consistent median weight loss, I might encounter a week or two of what I would call “mini-plateaus”. These are not true plateaus because the long term trend is still highly linear. In other words, I have not seen that my rate of weight loss has slowed down much over time. The curve is not “bottoming” out. And as long as the “macro” trend is so linear, that means what I am doing is working and working really freaking well. So I keep doing it. The more I do it the more confidence I have that the process works, and the more I trust it.</p><p>So the long term strategy is to create a calorie deficit, measure your results, and stick to the plan. But to pursue this strategy requires individualized tactics. In other words, you have to figure out how the heck to actually do this and do it consistently over time (whether it’s a couple months or much longer that) to achieve your target weight. I can give you some tips along these lines, but keep in mind, what works for me may not work for you. The most important thing is that you have to figure out what works for you.</p><h3>Diet plans</h3><p>Use whatever diet works for you best. If it’s keto, that’s neato. If it’s intermittent fasting, then I’m not even asking. Just do it. For me it’s just tracking calories and getting 3 balanced meals per day and usually an extra mini-meal in the form of a protein shake or protein bar. My only diet constraints I place on myself is to eat roughly 1 g of protein per pound of body weight each day. Furthermore, I try to make sure every meal has a good amount of protein. I don’t even actively track or restrict my other macros (carbs and fat) although I notice they are usually roughly equal to each other over the long term.</p><p>The high protein intake — in my way of thinking — does two things for me. 1) Protein is highly satiating compared to fats and carbs. If I spread out my protein intake over 3–4 meals per day it keeps me from getting hungry. For example, having eggs for breakfast as opposed to cereal, or having a turkey sandwich for lunch as opposed to a bowl of ramen helps me from getting hungry in between meals even on my high activity days. During this entire cut, I really haven’t ever felt “hungry” except after very intense bouts of activity like taking a 4-hour hike or something along those lines where I might burn 1000 calories or more. 2) High protein helps me maintain muscle during a cutting phase, but more on this later.</p><p>Again, if keto or intermittent fasting or high carb/low fat or carnivore or whatever the heck it may be — if it works for you that’s what matters most. The diet you can stick to is the best one long term. I would humbly suggest, though, start with something similar to mine, because it’s pretty simple and it may just work fine. If it doesn’t work, then it’s time to experiment with alternative diets.</p><p>One thing you should keep in mind is that there are a lot of charlatans out there selling these diets. And most of them make money by convincing you “there is only one way”. That is just not true.</p><h3>Be Rigorous and Anal (at least in the beginning)</h3><p>If you’ve never tracked your calories rigorously, and you are overweight, you probably do not realize how much you’re actually eating. To become 30, 50, 100, 200 pounds overweight you have to be regularly eating well above your maintenance calories for months, years or even decades. To begin to unwind these eating habits, you have to be accountable to yourself. And that literally means counting the calories in your food, but also being strategic about it so you can be set up for success.</p><p>Before you begin the calorie deficit phase of your diet, take at least a few days, ideally more like a week, and count your calories. Eat at home and prepare your own meals to make this easier. Eating out makes it vastly harder to get an accurate calorie total for that day, let alone if you eat out multiple times per day per week. If you eat out this much, believe me, it’s no wonder you are at the size you are now (I speak from my own lived experience!).</p><p>During this pre-diet phase you want to be very careful about counting. Count literally everything you can, even the teaspoon of half-and-half you put in your coffee (and let’s be honest I bet it’s a lot more than a teaspoon!). If you don’t already have one, but a food scale that is accurate to 1 g. As much as possible don’t measure volumes, measure by weight (mass) in grams. Let me give you an example that will drive home the point. Every morning I make two eggs on English muffin. The eggs are 70 calories each. The English muffin (specifically the buttermilk version) is 130 calories. To the eggs I add one serving of egg whites, which is roughly 45 g. I weight it out. I even weigh out the “pat” of butter that I put in the pan. It’s usually about 10–12 g of butter which is like 60–70 calories. You might not think that matters, but I promise you every calorie counts. So you should count every calorie!</p><p>Whatever you *think* you have been eating in terms of calories, I guarantee you it’s a lot more than that. You are probably eating 500 or even 1000 calories more than you think every day. I certainly was! And what’s crazy is that I have successfully dieted in the past, so I should have “known better”. But I let old habits gradually creep into my diet and I had never really tracked my calories before. My “old” way of dieting was based on caloric restriction or exclusion, where I just said “I’m not going to eat red meat” or “I’m not going to eat carbs”. Those diets worked because at the end of the day I was still creating a caloric deficit. But they weren’t fun diets by any means because excluding foods you love entirely isn’t fun. You shouldn’t have to do that. If you track your calories, you can still enjoy pretty much everything you love, you just have to pick and choose when and where it’s appropriate.</p><p>Now that you’ve counted the calories in your current diet, you are ready to construct your meal plan. This will probably take you a few days to a couple weeks to see what you can live with and without. I’ll use myself as a case study here. I was eating around 2800–3000 calories per day (easily). I calculated that I needed to get down to around 2000 calories to be in a 500 calorie deficit (it turned out to be somewhat higher than this but that’s besides the point for now). So I had to figure out how to pare down roughly 1000 calories from my diet. This was a serious food audit.</p><p>Let me give you some examples of how I did it:</p><ul><li>A typical breakfast in my old diet might have been a bagel with two eggs and an avocado. That might sound relatively harmless — and maybe even healthy! — but it’s easily 600, if not 700 calories (depending on bagel and size of avocado). I had to cut a few hundred calories here and still be satiated. So out goes the bagel, in comes the English muffin (saving ~100–200 calories). Out goes the avocado. In comes an additional serving of egg whites. That’s a savings of 200+ calories. Is it quite as satisfying or satiating? Maybe not, but it’s good enough to keep me going until my next meal, even including the 2-hour walk I take my dog on every day.</li><li>I added a 200 calorie protein smoothie between my breakfast and lunch meals. The smoothie is simply 30 grams of whey protein (130 cals), 145 g of frozen strawberries (45 cals), and 1 cup of unsweetened almond milk (30 cals). I actually look forward to this every day, even though it’s not the absolute sweetest thing ever, it’s very filling and the satiation lasts well into the lunchtime hours.</li><li>For lunch, I removed the 1000-calorie sandwiches, burritos etc, and I keep my lunches in the 400–600 calorie range each and every day. Big lunches were a big no no for me in terms of calories. I just make sure to get 30-ish grams of protein in my lunch and I’m good to go.</li><li>For dinner I was eating A LOT of steaks, beef stews, and basically didn’t realize how many calories were in those meat dishes. A grilled rib steak could easily be 1000 calories on its own. Don’t get me wrong, it’s delicious. But that can’t be an every night or really even a once a week thing for me. Those kinds of steaks have to be a special night out while I’m on the diet. Also out were huge portions pasta. The steak got mostly replaced with lean chicken (chicken breast and boneless/skinless chicken thighs) and occasionally salmon (even salmon is quite caloric and is more like a once-per-week thing for me). The carbs are almost always limited to around 200 calories for dinner, unless I have extra room in my budget that day, where I can take them up to 300 or so. To put it in perspective, 200 calories of rice is 1/3 cup raw which amounts to around 1 cup of cooked rice more or less. I also added in a fibrous vegetable to every dinner, such as broccoli, Brussels sprouts or cauliflower. These low calorie density vegetables help fill you without breaking the calorie bank. You still should track them though! I easily eat 50–100 calories in vegetables at dinner every night.</li></ul><p>In the beginning of the diet phase — at least the first few weeks — try to make your routine as consistent as possible. This means — again, at least for a few weeks — not making big changes to your diet and avoiding going out. Try to prepare as much as you can at home so you can measure and make sure you are sticking to your plan. That way you can see if your plan actually works. If you are aiming for a 500 calorie deficit — 1 pound-per-week — and at the end of 4 weeks you haven’t lost any weight, or maybe just 1 or 2 pounds, then it means you haven’t been executing the plan correctly. Now you have to diagnose the problem. If your tracking is off, it will make troubleshooting even more difficult. Conversely, for example, in my case, as soon as I started my plan and tracking pretty much religiously, the weight just started coming off. And each week that I shed a pound I had more confidence that I knew it was simply a matter of creating that calorie deficit. This meant I could start going a little bit “off plan” and being a little bit more creative with my diet as long as I could still track it. This is when you really hit your stride in the diet, when you realize you actually do have more freedom to be creative with your diet than you thought. But that doesn’t come without earning it. You really can only earn that freedom by mastering the calorie accounting principles first. Like I said, that will hopefully just take you a few weeks, and they probably won’t be easy. But if it were easy everyone would be doing it, right?</p><h3>Alcohol</h3><p>There are some other things I had to mostly cut out entirely from my daily calorie budget, and one of the most notable was alcoholic beverages. I was on average drinking roughly 1 drink per day, usually in the form of a cocktail or a glass of wine. It was probably 200 calories on average. It doesn’t sound like a lot, but when you are constructing a budget of around 2000 calories as opposed to eating at 3000 calories, you really have to pick your battles. Instead of that drink I decided I’d rather have a “healthy” dessert every night that is roughly the same amount of calories. My dessert is usually in the form of a parfait consisting of 1 cup of strawberries, 1/3 cup of granola, and about 100 g of nonfat Greek yogurt. This is certainly healthier than whatever alcohol I was consuming, definitely more satiating, and doesn’t make me as prone to late night snacking as the alcohol tended to be. The other thing about drinking is that it was probably worse for my sleep and workouts the next day. I haven’t gone completely sober — I still will have a couple drinks from time to time when I meet up with friends or family — but I have realized I was probably drinking a bit too much, even aside from the dietary concerns. So all things considered, while I am not telling you what to do about alcohol, I am telling you it’s definitely a thing you will want to consider dealing with in your diet. If you can keep drinking and meet your weight loss goals more power to you! You’ll get no moral superiority takes from me on that decision.</p><h3>Resistance Training</h3><p>One “advanced” dieting concept I learned about fairly recently is the “p-ratio”. This is basically the ratio of fat mass to total weight lost in a cut or gained during a bulking phase. When we talk about losing “weight”, we all think that weight=fat, but part of the weight lost is almost certainly going to be muscle, it’s just a matter of how much. One of the most important things to prevent losing excess muscle during a “fat loss” phase is to engage in resistance training, ie weight lifting. This isn’t the article to tell you exactly how to do that, but I will say that I try to get in 3 full-body weight lifting workouts per week and I keep track of my p-ratio (using various formulas I’ve come across) and it’s somewhere between 0.8–0.85 which seems fairly reasonable. That means for every 10 pounds I am losing, it’s roughly 1–1.5 pounds of muscle. And that’s with my eating a very high amount of protein and lifting quite regularly. So that’s just something to think about. If you are obese or very overweight the p-ratio tends to be naturally higher. It’s even possible to put on muscle mass and lose fat mass simultaneously if you are new to lifting and overweight.</p><p>One “internet myth” or even meme at this point is the idea that gaining muscle will drive your metabolism so high that you become a fat burning machine and will be able to eat as much as you want. This just isn’t the case. It’s been shown that each additional pound of muscle a person adds increases their basal metabolic rate (BMR) by roughly 8–10 calories. So even if you were to add 30 pounds of lean muscle to your body, that’s 240–300 calories. Well, I guess that means you will have earned that one drink per night! But yeah, you have to add 30 pounds of muscle. Not a bad goal, but it’s not going to make you a “fat burning machine”. The reasons you should lift are simply for overall health as you age — did you know falling is one of the leading killers of people over the age of 75? — and to prevent losing what muscle you already have while you lose fat. Oh, and you want to look good too right?</p><h3>It’s OK to say No</h3><p>One of the things you have to get used to doing during a diet to make it successful is to avoid all the sirens out there trying to bait you into breaking your diet. I think it’s a subconscious thing, but it feels like people are testing your will power to see how strong you really are. Just know that whatever they are offering you, a piece of cake, a slice of pie, a beer, you don’t actually need it. You don’t even really want it. A lot of what we eat in excess is just based on habit. Somebody offers you something for free, you kindly take it. Well, just know that you can simply break this habit. You won’t hurt their feelings. The cake won’t go to waste. And so what if it does. It’s not your job to be a biological garbage disposal. Instead of that cake, have a cup of coffee or a bottle of sparkling water.</p><h3>It’s OK to say Yes…If you earned it</h3><p>Of course, sometimes you know you’re going to want to eat cake or pie or ice cream or beer or fried chicken or steak. My main suggestion here is simply to plan ahead for those times by accounting for them in your budget *ahead of time*, instead of thinking you will make up for it later, which is much harder psychologically for most people. At least, it is for me. Planning ahead makes the cake a reward, as opposed to a punishment when you have to cut calories or do more cardio further after. One of the reasons I like doing a lot of cardio on my weekends is because it allows me to raise my calorie budget. The cake that you earn will taste that much better.</p><h3>The Diet Phase Does Not Last Forever</h3><p>There’s this idea out there that people can’t stick to diets because they can’t do it for the rest of their lives. Well, duh. If you tried to maintain a calorie deficit indefinitely you would eventually die. That’s called starvation. It’s amazing that so many people seemingly don’t understand how this is supposed to work. A weight loss phase is implementing a calorie deficit over a finite amount of time until a desired weight target is reached. The target isn’t 0. Once you reach the target (or before that, if you can’t get quite get there), you go back to maintenance, which means neither a calorie deficit or surplus — of course, you may decide to go into a surprlus, for example, to add muscle to your now lean body *wink*. So just to be clear, if your maintenance calorie budget is 2500 and you are dieting at 2000 calories, once you reach your target weight, you simply go back to 2500 calories. Now, if you were actually eating at 3000 calories for many years, even that 2500 might feel like a diet, but if you cut to 2000 calories for several months or even a year or more, then 2500 calories will probably feel absolutely amazing to you. With the tracking pyschology in place, I think it becomes much harder to revert to those truly bad habits, but of course, it’s possible you’ll just gain the weight back. Maybe at that point you should really consider medical alternatives (ie Ozempic and friends). Just a thought.</p><p>With that I wish you good luck in your journey, and I will check back in when I get back to my maintenance phase. :)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=27162355ab6a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The best kept secret for a successful diet plan…counting calories actually works!]]></title>
            <link>https://thecity2.medium.com/the-best-kept-secret-for-a-successful-diet-plan-counting-calories-actually-works-40053ec84375?source=rss-949f03393eeb------2</link>
            <guid isPermaLink="false">https://medium.com/p/40053ec84375</guid>
            <category><![CDATA[weight-loss]]></category>
            <category><![CDATA[diet]]></category>
            <category><![CDATA[weight-loss-tips]]></category>
            <category><![CDATA[nutrition]]></category>
            <dc:creator><![CDATA[Evan Zamir]]></dc:creator>
            <pubDate>Sun, 24 Sep 2023 21:18:45 GMT</pubDate>
            <atom:updated>2024-01-19T21:21:40.174Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tZMa3-KOewO2zvVqhH0akw.jpeg" /><figcaption>Photo by Towfiqu barbhuiya: <a href="https://www.pexels.com/photo/a-person-holding-his-belly-fat-9927899/">https://www.pexels.com/photo/a-person-holding-his-belly-fat-9927899/</a></figcaption></figure><p>My weight had crept up during COVID…well, crept would be kind. It was more like a jump scare. I am a 5&#39;6&quot; 47-year old male who has struggled with weight since childhood. At my all time “stockiness” I was in my mid-20s in grad school and weighed around 215 pounds. Here’s a pic of me circa Y2K at a conference giving a poster presentation (the short fat one in the middle, if you hadn’t guessed):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/876/1*Uot2Y7x0Hv_vgUYeLK9Y_Q.jpeg" /></figure><p>Within a year or so I lost ~50 pounds and got down to around 160 lbs. Here I am traveling in Israel in 2001:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*uX6WGMSbJIlrJgqE_D3rHw.jpeg" /><figcaption>Yep that’s me. Same guy as in the first pic after losing 50+ lbs.</figcaption></figure><p>I don’t remember much about how I lost the weight to be honest, other than the following: I started running a shit ton (like 6 miles every day), going to the gym and lifting weights almost every day, stopped eating red meat, and cut way back on carbs. Carbs were always my downfall. At this point I wasn’t really counting calories. It wasn’t as easy to do in 2000 as it is now without having a super computer in your pocket connected to the greatest wealth of knowledge human history has ever had. I guess I was doing some combination of “intuitive eating” and “Atkins”? I didn’t have a name for it, I just figured I had to do more activity and less eating. That much was clear.</p><p>Well fast forward to 2020…twenty years later and now in my mid-40s, I had let go of a lot of my good eating habits (started eating red meat again in earnest after a trip to Korea lol). I was still very active physically, but clearly my diet had gone off the rails. Living in San Francisco it was way too easy to go from “foodie” to “GLUTTON” and I’d say that’s essentially what had happened. My weight was back up around 185 and then COVID hit…and without the gym and with my eating habits somehow getting worse (at some point I convinced myself eating a donut for breakfast almost every day along with a cocktail in the evening was fine), my weight got up at one point around 205 or maybe even 210. Towards the end of 2022 or maybe the beginning of 2023 I knew I had to get control of my health back, especially approaching the “scary” age of 50. So I added an extra gym day and cut out the donuts…I thought maybe that would be enough. Well my weight did start slowly coming down but it stagnated around 195.</p><p>At that point I was almost resigned to it. I didn’t think I could do much to my diet to change the matter. And then something small at the time but massive in retrospect would happen. I went to a Giants game with my sister and nephew. My sister had actually been losing a lot of weight this past year and she casually mentioned a calorie tracking app that she was using. I won’t name the app here, as I don’t want anyone to think this is a paid commercial. It isn’t. There are tons of calorie tracking apps. It turns out in 2023 it’s really easy to track calories. So I started tracking my calories…</p><p>It literally only took a couple of days of calorie tracking when a lightning bolt went off in my head. My diet was out of control. It was clear why I had not been able to shed more pounds even though I was burning roughly 1000 calories “actively” every day (according to my Apple Watch) in addition to my basal metabolic activity. I was eating way way way too much. Probably around 2800–3000 calories every day. BTW if you don’t believe I was active take a look at my average steps over the past year:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*k7d1Cn030gtEHFt7TtJB5w.jpeg" /></figure><p>I got a pandemic pup in November of 2020 and since he was old enough to walk, we’ve literally done 5–6 mile walks or hikes every day for the past 2 years or so. Add that to the 3 days I go to the gym to lift or do a HIIT class, my problem was not cardio. My problem was a bad diet, that was not only high in calories but likely too low in nutritive value.</p><p>Once I got my hands on the calorie tracking app and after going down the YouTube fitness rabbit hole for a week or so, I renovated my diet. I came up with just a few guardrails that I wanted to adhere to for my self-made nutrition plan. Incredibly, all you had to do to get these tips *FOR FREE* was read this far :)</p><ul><li>Create a calorie deficit around 500 calories per day</li><li>Make sure I eat protein with every meal and over the course of a day eat around 150 g protein which is roughly 1 g per pound based on my goal weight of 160 lbs.</li></ul><p>You thought there would be more? Nope, that’s essentially it. I calculated my basal metabolic rate to be around 1800 calories per day using one of those online calculators and then add in the active calories according to my Apple Watch which is around 1000 calories per day. To create the 500 calorie deficit that means I need to eat no more than 2300 calories per day. Why 500 calories? Well, at that rate your calorie deficit in a week would be 3500 calories which is roughly one pound. “They” say a healthy rate of weight loss where you don’t risk losing too much muscle is around 0.5–1% of your body weight per week. Since I started the diet around 195 lbs that equates to 1–2 lbs per week.</p><p>Since I started the diet in earnest I have been weighing myself every morning right after I wake up and pee. It still jumps around more than I wish, but it’s way better than weighing yourself at completely random times throughout the day. Once I get my weight reading for the day I enter it into a spreadsheet and track my “performance” with the following two graphs. The first is a chart that shows my weekly median weight:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xudVQ9qg-XXS4OcrPTco8A.png" /><figcaption>Weekly median weight chart</figcaption></figure><p>The second is the daily weight chart which shows all the fluctuations, but I actually find it useful to see the “noise”, so you learn to get used to it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*I4CwuSSrjir67kabpLubrQ.png" /><figcaption>The daily weight chart. The red line is a reference hypothetical trend line which demarcates losing weight at a rate of 1 lb per week. (Didn’t weigh myself for a week in September when I vacationed in Paris. c’est la vie!<strong>)</strong></figcaption></figure><p>What I have found super useful in charting my weight is fitting a linear trend line to it. The slope of this trend indicates how much weight on average I am losing per day. You can see in the charts the number to the left of the “x” in the equations is roughly 0.19, which means I am losing 0.19 pounds per day. So how does that line up with my calculations? Well, 0.19 lbs per day equates to 7*0.19=1.33 lbs per week. That’s a bit higher than the 1 pound per week I was aiming for but it makes sense because, indeed, my “average” daily calorie deficit is probably closer to 700 calories than 500 calories. I could eat a bit more and slow down the weight loss, but I don’t really feel unsatiated enough to bother. I’m pretty happy with what I’m eating on a daily basis, and I figure this way, if I “cheat” a few times a month, it won’t hurt my progress. The tracking and the charting keeps me headed in the right direction. It makes me accountable for all my choices, and that’s really what you need in a diet.</p><p>You can see from the charts I’m almost back down to 180 pounds and assuming nothing crazy in my life happens I should get down to my goal weight at some point during the holiday season, at which point, I will go back to “maintenance” mode, meaning I’ll increase my calories back to where I’m taking in what I’m expending in energy on a daily basis.</p><p>You might be wondering what I eat in a typical day? Well, that’s easy for me to show you because I am staying pretty regimented. My “prototype” meal plan virtually every day looks like this:</p><h4>Breakfast</h4><ul><li>Two eggs scrambled on a whole wheat English muffin (~300 cals)</li><li>6 oz latte with 2% milk (~90 cals)</li></ul><h4>Mid-morning Protein Shake (~ 200 cals)</h4><ul><li>1 cup unsweetened almond milk</li><li>30 g whey protein isolate</li><li>handful of frozen strawberries or a small frozen banana</li></ul><h4>Lunch (400–600 calories)</h4><ul><li>On most days I’m eating a prepared meal from one of those apps (won’t name the one I use here, again so as not to make this seem like a paid commercial, but there are plenty of them to choose from). My only requirement is that the meal has 30+ g of protein.</li></ul><h4>Dinner (600–800 calories depending on my activity level that day)</h4><ul><li>1–2 days per week I’ll have one of the higher calorie meals from that same online service</li><li>More typically I will have the classic “chicken, broccolli, and rice” dinner that bodybuilders swear by and for good reason. It’s high protein, low calorie, and for the most part satiating, at least, it is for me.</li><li>Maybe once a week I’ll substitute salmon for the chicken. Or I’ll go out for dinner or there will be some other wildcard. But I try not to go crazy here. And if I go out, I try to apply the same guardrails. High protein, less carbs, etc.</li></ul><h4>Dessert (~230 calories)</h4><ul><li>My goto is a parfait consisting of some frozen strawberries, blackberries, a couple tablespoons of nonfat Greek yogurt, and a few tablespoons of granola. It is very easy to go overboard with all the ingredients and make this dessert 500 calories. A kitchen scale is your friend here!</li></ul><p>One of the keys to developing a meal plan for me was really tracking calories as carefully as possible in the beginning so I could really understand where I was going wrong. The main culprit was portion control, or more precisely, lack thereof. You don’t realize how much you’re really eating until you weigh it. Of course, it’s not easy to weigh everything, so my advice is to keep your meals simple and repeatable especially when you are starting down this path. What you really want to do is change your mindset and build good habits. Is it boring eating chicken, rice and broccoli every night or most every night? I mean, yeah, of course. It’s not super exciting food. But I will say that’s one of the reasons I like the idea of eating a lot of protein. Protein is way more satiating than carbs or fat. After my chicken, rice and broccoli dinners, I may not have a huge smile on my face, but I also feel perfectly satiated with just enough room left to enjoy my parfait. I don’t need to be excited by my food every night. If you want that, I’m not saying it isn’t possible. But it will require more effort on your part to craft those meals while sticking to the nutrition plan. It just will.</p><p>Now to the real reason I wrote this little diet manifesto. ALL THIS WORKS.</p><p>Calories In Minus Calories Out works.</p><p>A lot of people seem to not want you to believe that. They will tell you that it doesn’t work, because as you cut calories your metabolism will crash or something. They will tell you that you can’t track calories because it’s too hard. They will list off dozens of excuses why “calories don’t actually matter”.</p><p>I’m here to tell you it’s all bullshit they made up to sell you books or meal plans. That is the honest truth. They are literally telling you that the laws of thermodynamics don’t work because it doesn’t fit their business model.</p><p>They will tell you that you “have to do keto” or you “have to do intermittent fasting”…bullshit and more bullshit. It’s all bullshit. You literally don’t “have to do anything” except this one thing…</p><p>Make sure the calories you take in minus the calories you expend put you in a deficit. Not a huge deficit, but a real one. 500 calories. Maybe 700 calories like the mine currently. Probably not much more than that unless you don’t care about losing muscle. But that’s literally all it takes to lose weight. All this other stuff is complete bullshit pedaling from charlatans, many of whom have enough education that they should know better, and even worse, probably do know better but don’t care. They just want to take your money.</p><p>Well, I’m here to tell you that you don’t have to give anyone your money to lose weight successfully. You don’t need anything fancy. Just willpower and sweat. Go do it.</p><h3>UPDATE Jan 2024</h3><p>I did hit my original target weight but decided to keep going and am now currently around 165 lbs. My new goal is 155 lbs, which I haven’t hit since way back in 2002. Here’s the updated version of the daily weight chart (still linear!).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VMtJRo9GoDWm-blGIjUo3g.png" /></figure><p>Here’s the weekly median weigh-ins:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tidh7RFlEPAo69I4Lpql_w.png" /></figure><p>If all goes as planned I should hit my goal some time in March. To be continued…</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=40053ec84375" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Zen of Data]]></title>
            <link>https://thecity2.medium.com/the-zen-of-data-534bc9a4bb97?source=rss-949f03393eeb------2</link>
            <guid isPermaLink="false">https://medium.com/p/534bc9a4bb97</guid>
            <category><![CDATA[python]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[data]]></category>
            <dc:creator><![CDATA[Evan Zamir]]></dc:creator>
            <pubDate>Sun, 25 Apr 2021 17:37:10 GMT</pubDate>
            <atom:updated>2021-04-29T17:47:23.898Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FWDuHypQPgE9fknxwNK7pw.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@simonrae?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Simon Rae</a> on <a href="https://unsplash.com/s/photos/zen?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>Inspired by <a href="https://www.python.org/dev/peps/pep-0020/">The Zen of Python</a></p><ul><li>More data is better than less data.</li><li>Correct data is better than incorrect data.</li><li>Ethically gathered data is better than unethically gathered data.</li><li>Explicit assumptions are better than implicit ones.</li><li>Explaining predictions is desirable whenever possible.</li><li>Correlation does not imply causation. Really.</li><li>Randomized controlled experiment (RCE) is the gold standard for proving causality.</li><li>In some observational datasets causal inference *can* still be achieved even without RCE.</li><li>All else being equal the simpler model is better.</li><li>A good model using bad data is a bad model.</li><li>False Positives and False Negatives do not always have the same weight.</li><li>Visualizations should not sacrifice information content and readability for cleverness or novelty.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=534bc9a4bb97" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Bayesian Inference of Mock NBA Draft Order]]></title>
            <link>https://medium.com/data-science/bayesian-inference-of-mock-nba-draft-order-f4fa4a26be17?source=rss-949f03393eeb------2</link>
            <guid isPermaLink="false">https://medium.com/p/f4fa4a26be17</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[bayesian-statistics]]></category>
            <category><![CDATA[pytorch]]></category>
            <category><![CDATA[nba]]></category>
            <category><![CDATA[nba-draft]]></category>
            <dc:creator><![CDATA[Evan Zamir]]></dc:creator>
            <pubDate>Mon, 09 Nov 2020 22:35:00 GMT</pubDate>
            <atom:updated>2020-11-12T22:02:52.521Z</atom:updated>
            <content:encoded><![CDATA[<h4>How to use Uber’s probabilistic programming library Pyro to model the NBA Draft order</h4><p>Many of us who follow the NBA Draft closely have a keen desire to know the precise ordering of the Draft, eg where each player will be taken (“where” having a dual meaning here, both the team and draft slot). To this end, it is common behavior to ingest the information we get from the dozen or more “prominent” mock drafts available leading up to the Draft. There are different ways analysts and media try to aggregate this information into what I like to call “Meta Mocks”. Typically this just involves very simple operations like averaging ranks or looking at the minimum and maximum draft position. Chris Feller just the other day shared a very cool approach using survival analysis. I haven’t dug into it yet, but it will be interesting to compare to what I’m doing here.</p><h3></h3><p>In preparation for the NBA Draft, I&#39;ve compiled 20 mock drafts from various draft analysts to create a survival analysis which calculates the probability of a prospect being available at each draft slot (1/n)Dashboard Link: https://t.co/f6bdHv5gtu pic.twitter.com/yXrxJqrymD</p><p>What I wanted to do that is a bit different from using survival analysis, but can yield similar insights as I’ll show later, is to create a generative meta-model for mock drafts. In theory this enables one to infer some measure of “value” from rank orderings (no mock draft that I could find assigns numerical ratings to players) and then generate random mock draft orders from the posterior distribution. In other words, much like Nate Silver runs simulations to predict who will win the election, we can create a simulation for mock drafts and do all kinds of neat things therein. Let me show you how!</p><p>The first thing we need is a probability distribution that makes sense for rank data. Say hello to the <a href="https://cran.rstudio.com/web/packages/PlackettLuce/vignettes/Overview.html">Plackett-Luce Distribution</a>. The basic idea is that if you have some set <em>S</em> of items each with some “worth” <em>w_i </em>you can generate a probability distribution on the rank of any arbitrary permutation those items in some vector. I won’t show you the equations (see the reference), but I can show you a very simple simulation to get the idea across quickly. First we need to import some packages and define the PL Distribution which I found someone had implemented on github <a href="https://github.com/agadetsky/pytorch-pl-variance-reduction">here</a>. The work herein is all done using the <a href="http://pyro.ai/">Pyro probabilistic programming language</a> that is made open-source by Uber and built upon <a href="https://pytorch.org/">Facebook’s PyTorch</a> deep learning library.</p><pre><strong>import</strong> <strong>random</strong><br><br><strong>import</strong> <strong>pandas</strong> <strong>as</strong> <strong>pd</strong><br><strong>import</strong> <strong>numpy</strong> <strong>as</strong> <strong>np</strong><br><br><strong>import</strong> <strong>torch</strong><br><strong>from</strong> <strong>pyro.infer.autoguide</strong> <strong>import</strong> AutoDiagonalNormal <br><strong>from</strong> <strong>pyro.infer</strong> <strong>import</strong> SVI, Trace_ELBO, Predictive<br><strong>from</strong> <strong>pyro.distributions.distribution</strong> <strong>import</strong> Distribution<br><strong>from</strong> <strong>torch.distributions</strong> <strong>import</strong> constraints<br><strong>from</strong> <strong>pyro.distributions.torch_distribution</strong> <strong>import</strong> TorchDistributionMixin, TorchDistribution<br><strong>import</strong> <strong>pyro.distributions</strong> <strong>as</strong> <strong>dist</strong><br><strong>import</strong> <strong>pyro</strong><br><strong>import</strong> <strong>seaborn</strong> <strong>as</strong> <strong>sns</strong></pre><pre><em># some utilities defined in https://github.com/agadetsky/pytorch-pl-variance-reduction/blob/master/utils.py</em><br><br><strong>def</strong> logcumsumexp(x, dim):<br>    <em># slow implementation, but ok for now</em><br>    <strong>if</strong> (dim != -1) <strong>or</strong> (dim != x.ndimension() - 1):<br>        x = x.transpose(dim, -1)<br><br>    out = []<br>    <strong>for</strong> i <strong>in</strong> range(1, x.size(-1) + 1):<br>        out.append(torch.logsumexp(x[..., :i], dim=-1, keepdim=<strong>True</strong>))<br>    out = torch.cat(out, dim=-1)<br><br>    <strong>if</strong> (dim != -1) <strong>or</strong> (dim != x.ndimension() - 1):<br>        out = out.transpose(-1, dim)<br>    <strong>return</strong> out<br><br><br><strong>def</strong> reverse_logcumsumexp(x, dim):<br>    <strong>return</strong> torch.flip(logcumsumexp(torch.flip(x, dims=(dim, )), dim), dims=(dim, ))<br><br><strong>def</strong> smart_perm(x, permutation):<br>    <strong>assert</strong> x.size() == permutation.size()<br>    <strong>if</strong> x.ndimension() == 1:<br>        ret = x[permutation]<br>    <strong>elif</strong> x.ndimension() == 2:<br>        d1, d2 = x.size()<br>        ret = x[<br>            torch.arange(d1).unsqueeze(1).repeat((1, d2)).flatten(),<br>            permutation.flatten()<br>        ].view(d1, d2)<br>    <strong>elif</strong> x.ndimension() == 3:<br>        d1, d2, d3 = x.size()<br>        ret = x[<br>            torch.arange(d1).unsqueeze(1).repeat((1, d2 * d3)).flatten(),<br>            torch.arange(d2).unsqueeze(1).repeat((1, d3)).flatten().unsqueeze(0).repeat((1, d1)).flatten(),<br>            permutation.flatten()<br>        ].view(d1, d2, d3)<br>    <strong>else</strong>:<br>        <strong>ValueError</strong>(&quot;Only 3 dimensions maximum&quot;)<br>    <strong>return</strong> ret</pre><pre><strong>class</strong> <strong>PlackettLuce</strong>(Distribution):<br>    <em>&quot;&quot;&quot;</em><br><em>        Plackett-Luce distribution</em><br><em>    &quot;&quot;&quot;</em><br>    arg_constraints = {&quot;logits&quot;: constraints.real}<br><br>    <strong>def</strong> __init__(self, logits):<br>        <em># last dimension is for scores of plackett luce</em><br>        super(PlackettLuce, self).__init__()<br>        self.logits = logits<br>        self.size = self.logits.size()<br><br>    <strong>def</strong> sample(self, sample_shape=torch.Size()):<br>        <em># sample permutations using Gumbel-max trick to avoid cycles</em><br>        <strong>with</strong> torch.no_grad():<br>            logits = self.logits.unsqueeze(0).expand(sample_shape, *self.size)<br>            u = torch.distributions.utils.clamp_probs(torch.rand_like(logits))<br>            z = self.logits - torch.log(-torch.log(u))<br>            samples = torch.sort(z, descending=<strong>True</strong>, dim=-1)[1]<br>        <strong>return</strong> samples<br><br>    <strong>def</strong> log_prob(self, samples):<br>        <em># samples shape is: num_samples x self.size</em><br>        <em># samples is permutations not permutation matrices</em><br>        <strong>if</strong> samples.ndimension() == self.logits.ndimension():  <em># then we already expanded logits</em><br>            logits = smart_perm(self.logits, samples)<br>        <strong>elif</strong> samples.ndimension() &gt; self.logits.ndimension():  <em># then we need to expand it here</em><br>            logits = self.logits.unsqueeze(0).expand(*samples.size())<br>            logits = smart_perm(logits, samples)<br>        <strong>else</strong>:<br>            <strong>raise</strong> <strong>ValueError</strong>(&quot;Something wrong with dimensions&quot;)<br>        logp = (logits - reverse_logcumsumexp(logits, dim=-1)).sum(-1)<br>        <strong>return</strong> logp</pre><p>Ok, so let’s do a little simulation. It’s very simple. Imagine that we have 5 race car drivers. We are going to generate some “worth” vector using an Exponential distribution. We see that Driver 3 has by the far the highest rating and Driver 0 the lowest. Given those parameters we then generate random samples of “races” using the PL distribution. We do this 10 times.</p><pre>n_drivers = 5<br>torch.manual_seed(39)<br>theta = dist.Exponential(0.1).sample((n_drivers,))<br>ranks = list(zip(range(n_drivers),theta.numpy()))<br>[print(f&#39;driver <strong>{</strong>i<strong>}</strong>: <strong>{</strong>j<strong>:</strong>0.3<strong>}</strong>&#39;) <strong>for</strong> i, j <strong>in</strong> sorted(ranks,key=<strong>lambda</strong> x: -x[1])]<br>pl = PlackettLuce(theta)<br>data = pl.sample(10)<br>print(data.numpy())</pre><pre># output<br>driver 3: 31.7<br>driver 4: 23.0<br>driver 2: 2.02<br>driver 1: 1.02<br>driver 0: 0.294<br>[[3 4 1 2 0]<br> [3 4 1 2 0]<br> [3 4 2 0 1]<br> [3 4 2 1 0]<br> [3 4 2 1 0]<br> [3 4 1 0 2]<br> [3 4 2 1 0]<br> [3 4 1 2 0]<br> [3 4 2 1 0]<br> [3 4 0 2 1]]</pre><p>You can see that drivers 3 and 4 come in 1 and 2 in all 10 of our simulated races.</p><p>Now let’s do the same simulation but increase the Exponential rate parameter such that the drivers are more “even”.</p><pre>n_drivers = 5<br>torch.manual_seed(39)<br>theta = dist.Exponential(1.0).sample((n_drivers,))<br>ranks = list(zip(range(n_drivers),theta.numpy()))<br>[print(f&#39;driver <strong>{</strong>i<strong>}</strong>: <strong>{</strong>j<strong>:</strong>0.3<strong>}</strong>&#39;) <strong>for</strong> i, j <strong>in</strong> sorted(ranks,key=<strong>lambda</strong> x: -x[1])]<br>pl = PlackettLuce(theta)<br>data = pl.sample(10)<br>print(data.numpy())</pre><pre>#output<br>driver 3: 3.17<br>driver 4: 2.3<br>driver 2: 0.202<br>driver 1: 0.102<br>driver 0: 0.0294<br>[[1 4 3 2 0]<br> [3 1 4 2 0]<br> [3 4 2 0 1]<br> [4 3 2 0 1]<br> [4 3 2 1 0]<br> [3 4 0 1 2]<br> [3 4 2 0 1]<br> [3 4 1 2 0]<br> [3 4 2 0 1]<br> [4 0 3 2 1]]</pre><p>You can see that now driver 4 wins some races and even driver 1 gets to stand at the podium. Sweet! Hopefully this is starting to make sense to you. We’re basically ready to go. We just need to read in some data, create the model, and do some inference. Let’s do it!</p><p>I’ve collected data from 17 publicly available mock drafts. Not all drafts mock both rounds though. That is ok, since there is still useful information in even one round. I’ll explain how I deal with that shortly, but let’s read in the data first.</p><pre>df = pd.read_csv(&#39;data/mock-drafts.csv&#39;)<br>df.head()</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*23cdc02hcDab1SUvHWsxxg.png" /><figcaption>Image by Author</figcaption></figure><pre>cols = df.columns[1:]<br>players_set = set()<br><strong>for</strong> col <strong>in</strong> cols:<br>    <strong>for</strong> player <strong>in</strong> df[col].values:<br>        players_set.add(player)<br>players = {player:idx <strong>for</strong> idx,player <strong>in</strong> enumerate(players_set) <strong>if</strong> type(player) <strong>is</strong> str}<br>player_lut = {idx:player <strong>for</strong> player,idx <strong>in</strong> players.items() <strong>if</strong> type(player) <strong>is</strong> str}<br>print(player_lut)</pre><pre># output<br>{1: &#39;Kira Lewis&#39;, 2: &#39;Udoka Azubuike&#39;, 3: &#39;Tyrell Terry&#39;, 4: &#39;Saben Lee&#39;, 5: &#39;Aleksej Pokusevski&#39;, 6: &#39;Cassius Winston&#39;, 7: &#39;Austin Wiley&#39;, 8: &#39;Devon Dotson&#39;, 9: &#39;Naji Marshall&#39;, 10: &#39;Nick Richards&#39;, 11: &#39;Yam Madar&#39;, 12: &#39;Zeke Nnaji&#39;, 13: &#39;Kaleb Wesson&#39;, 14: &#39;Mason Jones&#39;, 15: &#39;Patrick Williams&#39;, 16: &#39;James Wiseman&#39;, 17: &#39;Tyrese Haliburton&#39;, 18: &#39;Theo Maledon&#39;, 19: &#39;LaMelo Ball&#39;, 20: &#39;Tyrique Jones&#39;, 21: &#39;Ty-Shon Alexander&#39;, 22: &#39;Vernon Carey&#39;, 23: &#39;Reggie Perry&#39;, 24: &#39;Xavier Tillman&#39;, 25: &#39;Filip Petrusev&#39;, 26: &#39;Leandro Bolmaro&#39;, 27: &#39;Josh Green&#39;, 28: &#39;Abdoulaye Ndoye&#39;, 29: &#39;Killian Hayes&#39;, 30: &#39;Tyler Bey&#39;, 31: &#39;Isaiah Stewart&#39;, 32: &#39;Lamine Diane&#39;, 33: &#39;Desmond Bane&#39;, 34: &#39;Lamar Stevens&#39;, 35: &#39;Aaron Nesmith&#39;, 36: &#39;Tyrese Maxey&#39;, 37: &#39;Anthony Edwards&#39;, 38: &#39;Yoeli Childs&#39;, 39: &#39;Omer Yurtseven&#39;, 40: &#39;Borisa Simanic&#39;, 41: &#39;Justinian Jessup&#39;, 42: &#39;Malachi Flynn&#39;, 43: &#39;Jahmius Ramsey&#39;, 44: &#39;Obi Toppin&#39;, 45: &#39;Markus Howard&#39;, 46: &#39;Tre Jones&#39;, 47: &#39;Robert Woodard&#39;, 48: &#39;Nate Hinton&#39;, 49: &#39;Isaac Okoro&#39;, 50: &#39;Cole Anthony&#39;, 51: &#39;Jaden McDaniels&#39;, 52: &#39;Jordan Nwora&#39;, 53: &#39;Deni Avdija&#39;, 54: &#39;Jalen Harris&#39;, 55: &#39;Rokas Jokubaitis&#39;, 56: &#39;Paul Reed&#39;, 57: &#39;Jay Scrubb&#39;, 58: &#39;Ashton Hagans&#39;, 59: &#39;Jalen Smith&#39;, 60: &#39;Kristian Doolittle&#39;, 61: &#39;Henri Drell&#39;, 62: &#39;CJ Elleby&#39;, 63: &#39;Isaiah Joe&#39;, 64: &#39;Paul Eboua&#39;, 65: &#39;Killian Tillie&#39;, 66: &#39;Kenyon Martin&#39;, 67: &#39;RJ Hampton&#39;, 68: &#39;Immanuel Quickley&#39;, 69: &#39;Sam Merrill&#39;, 70: &#39;Onyeka Okongwu&#39;, 71: &#39;Grant Riller&#39;, 72: &#39;Payton Pritchard&#39;, 73: &#39;Skylar Mays&#39;, 74: &#39;Karim Mane&#39;, 75: &#39;Giorgos Kalaitzakis&#39;, 76: &#39;Mamadi Diakite&#39;, 77: &#39;Nico Mannion&#39;, 78: &#39;Saddiq Bey&#39;, 79: &#39;Marko Simonovic&#39;, 80: &#39;Paul Ebouta&#39;, 81: &#39;Precious Achiuwa&#39;, 82: &#39;Cassius Stanley&#39;, 83: &#39;Daniel Oturu&#39;, 84: &#39;Elijah Hughes&#39;, 85: &#39;Josh Hall&#39;, 86: &#39;Devin Vassell&#39;}</pre><p>As you can see there are 86 players total in these mock drafts. Now we’re going to put the mocks in a data format that Pyro can understand, a Torch tensor containing 17 rows. Each row will have all 86 players in some permutation. Now I can explain how I “pad” each mock such that whether a mock had 30 or 60 players, each will be padded to contain all 86. I assume something very simple: if a mock doesn’t have a player then every other player can just be randomly shuffled to fill out the full mock to position 86. This is about the simplest way I could do it. A more advanced way would be to start with a randomization like this, but then subsequently refine the “counterfactual” padding with a random order that is based on the inferred ratings from the model. One could keep iterating like this until some equilibrium is reached (basically until there is no more improvement).</p><pre><strong>import</strong> <strong>random</strong><br>n_players = len(player_lut)<br>list_data = []<br><strong>for</strong> col <strong>in</strong> cols:<br>    mock = df[col]<br>    ranks = []<br>    players_cp = players.copy()<br>    <strong>for</strong> i <strong>in</strong> range(1,n_players+1):<br>        <strong>try</strong>:<br>            player = mock[i-1]<br>            players_cp.pop(player)<br>            ranks.append(players[player])<br>        <strong>except</strong> <strong>KeyError</strong>: <br>            random_player = random.choice(list(players_cp.keys()))<br>            <strong>if</strong> players_cp.pop(random_player):<br>                ranks.append(players[random_player])<br>    list_data.append(ranks)<br>data = torch.tensor(list_data)-1<br>print(data)</pre><pre># output<br>tensor([[18, 15, 36,  ..., 44, 73,  6],<br>        [36, 52, 15,  ...,  8, 53, 78],<br>        [18, 16, 36,  ..., 75, 61, 63],<br>        ...,<br>        [18, 15, 36,  ..., 31,  3,  6],<br>        [36, 18, 69,  ..., 72, 51, 63],<br>        [36, 15, 18,  ..., 37, 44, 19]])</pre><p>Let’s look at one of these rows:</p><pre>[player_lut[i+1] <strong>for</strong> i <strong>in</strong> data.numpy()[0]][0:10]</pre><pre># output<br>[&#39;LaMelo Ball&#39;,<br> &#39;James Wiseman&#39;,<br> &#39;Anthony Edwards&#39;,<br> &#39;Deni Avdija&#39;,<br> &#39;Obi Toppin&#39;,<br> &#39;Tyrese Haliburton&#39;,<br> &#39;Isaac Okoro&#39;,<br> &#39;Devin Vassell&#39;,<br> &#39;Onyeka Okongwu&#39;,<br> &#39;Killian Hayes&#39;]</pre><p>Ok, now the exciting part. Let’s define the model:</p><pre>n_players = 86<br><br><strong>def</strong> model(data):<br>    sigma = pyro.sample(&quot;sigma&quot;, dist.Uniform(0,1))<br>    theta = pyro.sample(&quot;theta&quot;, dist.Cauchy(torch.zeros(n_players,),sigma*torch.ones(n_players,)))<br>    <strong>with</strong> pyro.plate(&quot;data&quot;, data.shape[-2]):<br>        pyro.sample(&quot;obs&quot;, PlackettLuce(theta), obs=data)</pre><p>That’s all it is really. I’m using a Cauchy distribution because if you think about the distribution of NBA talent it seems “fat-tailed”. In other words, we want our model to be able to handle the case where there is a “superstar” or consensus top pick(s). If we make our distribution too flat, it would be more like a horse race where any horse has a shot at finishing first (more or less). Now we do the inference step, setting up a guide and optimizer according to the Pyro parlance for doing stochastic variational inference (in the future I plan to investigate the more accurate MCMC sampling approach, but this should fine for now).</p><pre>guide = AutoDiagonalNormal(model)<br>adam = pyro.optim.Adam({&quot;lr&quot;: 0.03})<br>svi = SVI(model, guide, adam, loss=Trace_ELBO())<br>pyro.clear_param_store()<br><strong>for</strong> j <strong>in</strong> range(5000):<br>    <em># calculate the loss and take a gradient step</em><br>    loss = svi.step(data)<br>    <strong>if</strong> j % 100 == 0:<br>        print(&quot;[iteration <strong>%04d</strong>] loss: <strong>%.4f</strong>&quot; % (j + 1, loss / data.shape[-2]))</pre><p>Here are the results of our Meta Mock!</p><pre>1. Anthony Edwards (12.2)<br>2. LaMelo Ball (12.0)<br>3. Deni Avdija (10.3)<br>4. James Wiseman (10.3)<br>5. Obi Toppin (9.9)<br>6. Onyeka Okongwu (8.0)<br>7. Killian Hayes (7.9)<br>8. Tyrese Haliburton (7.6)<br>9. Isaac Okoro (7.6)<br>10. Devin Vassell (6.7)<br>11. Patrick Williams (6.0)<br>12. Aaron Nesmith (5.2)<br>13. Saddiq Bey (4.7)<br>14. Kira Lewis (4.6)<br>15. Precious Achiuwa (4.5)<br>16. RJ Hampton (4.1)<br>17. Jalen Smith (4.0)<br>18. Cole Anthony (3.9)<br>19. Tyrese Maxey (3.7)<br>20. Josh Green (3.0)<br>21. Aleksej Pokusevski (2.6)<br>22. Leandro Bolmaro (2.4)<br>23. Isaiah Stewart (2.3)<br>24. Jaden McDaniels (2.2)<br>25. Theo Maledon (2.0)<br>26. Jahmius Ramsey (1.7)<br>27. Tyrell Terry (1.6)<br>28. Desmond Bane (1.3)<br>29. Nico Mannion (1.2)<br>30. Zeke Nnaji (1.1)<br>31. Malachi Flynn (1.0)<br>32. Tre Jones (0.9)<br>33. Devon Dotson (0.8)<br>34. Paul Reed (0.5)<br>35. Vernon Carey (0.5)<br>36. Isaiah Joe (0.5)<br>37. Elijah Hughes (0.5)<br>38. Cassius Stanley (0.5)<br>39. Daniel Oturu (0.4)<br>40. Killian Tillie (0.4)<br>41. Cassius Winston (0.4)<br>42. Grant Riller (0.3)<br>43. Udoka Azubuike (0.3)<br>44. Tyler Bey (0.3)<br>45. Reggie Perry (0.2)<br>46. Xavier Tillman (0.2)<br>47. Payton Pritchard (0.1)<br>48. Jay Scrubb (0.1)<br>49. Abdoulaye Ndoye (0.1)<br>50. Immanuel Quickley (-0.0)<br>51. Ty-Shon Alexander (-0.0)<br>52. Ashton Hagans (-0.0)<br>53. Jordan Nwora (-0.1)<br>54. Robert Woodard (-0.1)<br>55. Nick Richards (-0.1)<br>56. Yam Madar (-0.1)<br>57. Skylar Mays (-0.2)<br>58. Nate Hinton (-0.2)<br>59. Paul Ebouta (-0.2)<br>60. Filip Petrusev (-0.3)<br>61. Justinian Jessup (-0.3)<br>62. Lamar Stevens (-0.3)<br>63. Marko Simonovic (-0.3)<br>64. Sam Merrill (-0.3)<br>65. Josh Hall (-0.4)<br>66. Karim Mane (-0.4)<br>67. Kaleb Wesson (-0.4)<br>68. Kristian Doolittle (-0.5)<br>69. Tyrique Jones (-0.5)<br>70. Borisa Simanic (-0.5)<br>71. Markus Howard (-0.5)<br>72. Kenyon Martin (-0.5)<br>73. CJ Elleby (-0.6)<br>74. Rokas Jokubaitis (-0.6)<br>75. Lamine Diane (-0.6)<br>76. Yoeli Childs (-0.6)<br>77. Jalen Harris (-0.6)<br>78. Mason Jones (-0.7)<br>79. Paul Eboua (-0.7)<br>80. Saben Lee (-0.7)<br>81. Mamadi Diakite (-0.7)<br>82. Omer Yurtseven (-0.8)<br>83. Henri Drell (-0.8)<br>84. Austin Wiley (-0.8)<br>85. Naji Marshall (-0.8)<br>86. Giorgos Kalaitzakis (-0.9)</pre><p>Ok, not bad, but that’s not the coolest part. The coolest part is now we have probability distributions for each player so we can go and simulate thousands of mock drafts. Let’s do it! Basically we’re just going to sample the posterior distribution (from the Pyro guide function) and generate one sample mock draft order. We then do that 10,000(!) times and place them all in a Pandas dataframe.</p><pre>rank_hist = []<br><strong>for</strong> m <strong>in</strong> range(10_000):<br>    theta = guide.forward()[&#39;theta&#39;]<br>    pl = PlackettLuce(theta)<br>    sample = pl.sample(1).numpy()[0]<br>    <strong>for</strong> i, j <strong>in</strong> enumerate(sample.squeeze()):<br>        <strong>if</strong> i&lt;=59:<br>            rank_hist.append((i+1,player_lut[j+1]))<br>        <strong>else</strong>:<br>            rank_hist.append((61,player_lut[j+1]))<br>rank_df = pd.DataFrame.from_records(rank_hist,columns=[&#39;rank&#39;,&#39;player&#39;])</pre><p>So now we have distributions (histograms) of simulated draft position for each player. For example, let’s look at where Anthony Edwards is predicted:</p><pre>sns.set_style(&quot;ticks&quot;, {&quot;xtick.minor.size&quot;: 1, &quot;ytick.major.size&quot;: 8})<br>players = [&#39;Anthony Edwards&#39;]<br>sns.displot(data=rank_df[rank_df.player.isin(players)],x=&#39;rank&#39;,kind=&#39;hist&#39;,hue=&#39;player&#39;)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/481/1*sSIKBB__3sLRnJ8fKopzMg.png" /><figcaption>Image by Author</figcaption></figure><p>An easier way to understand this distribution is to look at the cumulative probability. Remember when I said this model can be used in a similar way to survival analysis? This is what I’m talking about:</p><pre>sns.set_style(&quot;ticks&quot;, {&quot;xtick.minor.size&quot;: 1, &quot;ytick.major.size&quot;: 8})<br>players = [&#39;Anthony Edwards&#39;]<br>sns.displot(data=rank_df[rank_df.player.isin(players)],x=&#39;rank&#39;,kind=&#39;ecdf&#39;,hue=&#39;player&#39;)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/481/1*ec8-QCk0impUOoJrFJqkkQ.png" /><figcaption>Image by Author</figcaption></figure><p>From this we see Edwards has around a 45% chance of being taken first and around 80% to go top 2 and so on. Pretty neat! Now let’s look at someone with a wider range of predictions, Saddiq Bey:</p><pre>sns.set_style(&quot;ticks&quot;, {&quot;xtick.minor.size&quot;: 1, &quot;ytick.major.size&quot;: 8})<br>players = [&#39;Saddiq Bey&#39;]<br>sns.displot(data=rank_df[rank_df.player.isin(players)],x=&#39;rank&#39;,kind=&#39;ecdf&#39;,hue=&#39;player&#39;)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/450/1*8zE6Y_CtiRYdsN8ID0VKbQ.png" /><figcaption>Image by Author</figcaption></figure><p>Ooh isn’t that cool? A very wide range indeed. Let’s take a look at a more fringey player, Nick Richards:</p><pre>sns.set_style(&quot;ticks&quot;, {&quot;xtick.minor.size&quot;: 1, &quot;ytick.major.size&quot;: 8})<br>players = [&#39;Nick Richards&#39;]<br>sns.displot(data=rank_df[rank_df.player.isin(players)],x=&#39;rank&#39;,kind=&#39;ecdf&#39;,hue=&#39;player&#39;)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/462/1*F2XzCNs2IFyc1Y8h4vnblA.png" /><figcaption>Image by Author</figcaption></figure><p>Ok, so I should explain that in the simulation if a player is not drafted in the top 60, I just put him at 61, thus essentially making him an undrafted free agent (UDFA). These plots get kind of messy but before I’m done, here’s one that has the top 10 guys in one chart.</p><pre>sns.set_style(&quot;ticks&quot;, {&quot;xtick.minor.size&quot;: 1, &quot;ytick.major.size&quot;: 8})<br>players = [&#39;Anthony Edwards&#39;, &#39;LaMelo Ball&#39;, &#39;Deni Avdija&#39;, &#39;James Wiseman&#39;, &#39;Obi Toppin&#39;, &#39;Onyeka Okongwu&#39;,<br>          &#39;Killian Hayes&#39;, &#39;Tyrese Haliburton&#39;, &#39;Isaac Okoro&#39;, &#39;Devin Vassell&#39;]<br>sns.displot(data=rank_df[rank_df.player.isin(players)],x=&#39;rank&#39;,kind=&#39;ecdf&#39;,hue=&#39;player&#39;)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/481/1*KMHsZ0tkgDXoeCN64jiJZQ.png" /><figcaption>Image by Author</figcaption></figure><p>Well, that’s all I got for now. Be on the lookout (via my Twitter) for more things related to this as we get closer to the Draft (Nov. 18). If you have any suggestions (or corrections) let me know! Here’s a link to the github repo for this project:</p><h3></h3><p>Read for some Friday Night Stats? Check out: &quot;Bayesian Inference of Mock NBA Draft Order&quot;. https://t.co/GXQ1gZa8td</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f4fa4a26be17" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science/bayesian-inference-of-mock-nba-draft-order-f4fa4a26be17">Bayesian Inference of Mock NBA Draft Order</a> was originally published in <a href="https://medium.com/data-science">TDS Archive</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Simple Neural Network for In-Game Win Probability Modeling of NCAA Basketball Games]]></title>
            <link>https://medium.com/analytics-vidhya/a-simple-neural-network-for-in-game-win-probability-modeling-of-ncaa-basketball-games-58ab4e3ca0f9?source=rss-949f03393eeb------2</link>
            <guid isPermaLink="false">https://medium.com/p/58ab4e3ca0f9</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[sports-statistics]]></category>
            <category><![CDATA[basketball]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[neural-networks]]></category>
            <dc:creator><![CDATA[Evan Zamir]]></dc:creator>
            <pubDate>Fri, 03 Jan 2020 23:15:33 GMT</pubDate>
            <atom:updated>2020-01-07T12:53:22.808Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7zKjE0bIN7pzZaU_y5558Q.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@franki?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Franki Chamaki</a> on <a href="https://unsplash.com/s/photos/artificial-intelligence?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>Win probability (WP) is the probability of winning (duh) — given a certain <em>context </em>(or <em>prior knowledge</em> in the parlance of statistics). That context is everything. For example, bettors typically have some idea about the win probabilities for an upcoming game based on their prior knowledge of the strengths and weaknesses of the two teams in general and the context of the specific game at hand. Is it a home game? Are the players coming off a day of rest or a bye week? Is a star player injured?</p><p>In-game win probability — as the name implies — additionally models the context of moment-by-moment factors <em>during</em> the course of a game. This can be useful not only for sports betting but also for fans who want to be more engaged with their teams. Although WP can be modeled for any sport, in this post, I’m going to focus on the sport of basketball, and specifically, men’s Division I college basketball for which I have an appropriate database of plays that I started collecting this season. Here is a sample of what the dataset looks like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1018/1*z_IJNWHQBaE0RtXQdgXBnA.png" /><figcaption>Dataset for win probability modeling</figcaption></figure><p>Each row here is a moment in time during an NCAA game where an event happened that changed the score or time on the clock — field goal attempt, rebound, turnover, etc. The relevant features for the model described here will be the <strong>lead</strong> (home score - visitor score), <strong>net rating</strong> which is just the difference in <a href="https://kenpom.com/">kenpom</a> ratings between the two teams (home - visitor), and <strong>time</strong> remaining in the game in seconds. The target variable — the one we are trying to predict with our model — is, of course, the outcome <strong>win</strong>, whether the home team won (1) or lost (0) the game in regulation. Note that for the current model I have just chosen to look at games that did not end in a tie — I will deal with ties…<em>eventually</em>. This was just a choice I made, for now, to simplify and get things working <em>fast</em>. Another simplification was to leave out the context of possession — which team has the ball, which is obviously critical to know very late in the game as time winds down.</p><p>Prior models for win probability in basketball (and other sports as far as I can tell) have mostly relied on some form of local logistic regression (<a href="https://lukebenz.com/post/ncaahoopr_win_prob/">here</a> and <a href="https://www.inpredictable.com/2013/06/nba-in-game-win-probability.html">here</a>). I actually tried doing pure logistic regression, and sure enough, that just doesn’t work well at all. When I read that people have previously gone the local logistic regression route, part of me thought, well, there’s no fun in doing exactly what has been done before! As I started to think about it, I wondered if a neural network could figure out the same kind of locality and non-linearities essentially that are needed to model win probability. As it turns out, I think the answer to that question is at least to some extent a clear <em>yes</em>.</p><p>To help guide me on this journey, I relied heavily on one of the figures from <a href="https://www.inpredictable.com/2013/06/nba-in-game-win-probability.html">Mike Beuoy’s blog post</a> (please read it and follow him @<a href="https://twitter.com/inpredict">inpredict</a> on Twitter) which shows the win probability (for NBA games) for a team ahead by 3 points that is either a 5-point favorite or underdog as a function of time during the game. The solid lines are the model, and the dots represent bucketed win probabilities for the actual game outcomes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m1FRrKIb2BGz9aplRdNPIA.png" /><figcaption>Screenshot from Mike Beuoy’s WP post</figcaption></figure><p>My goal was basically to build a model the could replicate something like this. Before doing that, however, let’s see how similar the NCAA data are.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/752/1*nyilPJa1_15I5LY1YYyukw.png" /><figcaption>NCAA win probabilities for teams with a lead between 2–4 points and net ratings around 5 (blue) or -5 (orange). Time buckets here are 60 seconds.</figcaption></figure><p>You can see from the figure — if you squint hard enough — that the raw win probabilities are actually not too different from Mike’s data. One thing to note is that time is going “backwards” on my plots — 0 is the buzzer at the end of game. Ok, so now on to my model.</p><p>I should mention that I tried a few different things, including a sort of “exponential transformation” of time which looked like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*tsFI1AUXd0Rh0Wn8481Wgg.png" /></figure><p>As it turned out, I just needed to follow the mantra of deep learning which is to let the neural network learn the features it needs. So I abandoned the time transformation and just made my neural network one layer deeper and that seemed to do the trick.</p><p>In TensorFlow 2.0 the model — which is simple AF really is just this:</p><pre>model = tf.keras.models.Sequential([<br>    tf.keras.layers.Dense(12, activation=&#39;relu&#39;, input_shape=[3,]),<br>    tf.keras.layers.Dense(12, activation=&#39;relu&#39;),<br>    tf.keras.layers.Dense(1, activation=&#39;sigmoid&#39;)<br>])<br>model.compile(optimizer=tf.keras.optimizers.RMSprop(1e-3), loss=&#39;binary_crossentropy&#39;, metrics=[&#39;accuracy&#39;, &#39;AUC&#39;])</pre><pre>model.summary()</pre><pre>model.fit(df[[&#39;lead&#39;,&#39;net&#39;,&#39;time&#39;]].to_numpy(),<br>          df[&#39;win&#39;].to_numpy(),<br>          verbose=1,<br>          shuffle=True,<br>          epochs=50,<br>          batch_size=512)</pre><p>That’s it, just an input layer, 2 hidden layers with 12 neurons each, and an output layer with sigmoid activation. And it basically works as you can see here for team with a 3-point lead that is either a 5 point favorite or dog in terms of net rating:</p><pre>times = list(range(0,2400,30))<br>winp_favorite = [model.predict_proba(x=[[3,5,time]])[0][0] for time in times]<br>winp_underdog = [model.predict_proba(x=[[3,-5,time]])[0][0] for time in times]</pre><pre>plt.plot(times, winp_favorite, &#39;b-&#39;)<br>plt.plot(times, winp_underdog, &#39;r-&#39;)<br>plt.title(&#39;NCAA Win Probability&#39;)<br>plt.xlabel(&#39;Time elapsed (s)&#39;)<br>plt.savefig(&#39;ncaa_win_prob.png&#39;)</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/432/1*pEnAXpUZRtQHLXZxn0trkA.png" /><figcaption>The model basically works!</figcaption></figure><p>By the way if you just have 1 hidden layer you can get wonky-looking curves like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/432/1*d2PctUFg_uAmBI9euKzJtg.png" /><figcaption>Wonky-looking model with only 1 measly hidden layer</figcaption></figure><p>Finally, to show you how the features work individually here are win probabilities for a team with a +5 net rating and 30 seconds left on the clock with a lead between 1–5 points:</p><pre>model.predict_proba(x=[[1, 5, 30], [2, 5, 30], [3, 5, 30], [4, 5, 30], [5, 5, 30]])</pre><pre>&gt;<br>array([[0.7975986 ],<br>       [0.8999447 ],<br>       [0.95134   ],<br>       [0.978909  ],<br>       [0.99045944]], dtype=float32)</pre><p>As expected win probability jumps way up with just a few extra points as it becomes very difficult for the team that his behind to gain multiple possessions in so short a time. Finally, we can see that the win probability of a team with a 4-point lead with 30 seconds to go does not really depend much at all on the net rating (as <a href="https://lukebenz.com/post/ncaahoopr_win_prob/">Luke Benz showed</a>):</p><pre># varying net rating from -5 to +5<br>model.predict_proba(x=[[4, -5, 30], [4, 0, 30], [4, 5, 30]])<br>&gt;<br>array([[0.94858927],<br>       [0.97111243],<br>       [0.978909  ]], dtype=float32)</pre><p>I’m excited that this works at all, although I guess it’s not too surprising. But I wanted to share it with the hopes that other people might go down this route as well and add more features, improved architectures, and such. As I said earlier a next obvious step would be to add possession as an additional feature. Have fun with it!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=58ab4e3ca0f9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/analytics-vidhya/a-simple-neural-network-for-in-game-win-probability-modeling-of-ncaa-basketball-games-58ab4e3ca0f9">A Simple Neural Network for In-Game Win Probability Modeling of NCAA Basketball Games</a> was originally published in <a href="https://medium.com/analytics-vidhya">Analytics Vidhya</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Stochastic Gradient Descent for Coin Flips]]></title>
            <link>https://thecity2.medium.com/stochastic-gradient-descent-for-coin-flips-8e6475fd70cc?source=rss-949f03393eeb------2</link>
            <guid isPermaLink="false">https://medium.com/p/8e6475fd70cc</guid>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Evan Zamir]]></dc:creator>
            <pubDate>Thu, 12 Dec 2019 05:11:53 GMT</pubDate>
            <atom:updated>2019-12-12T05:16:41.249Z</atom:updated>
            <content:encoded><![CDATA[<p>Here I’m going to show you in explicit detail how to perform stochastic gradient descent (SGD) for a Bernoulli simulation (eg coin flips). This is arguably the simplest non-trivial problem which fully demonstrates SGD. The likelihood function for a single Bernoulli trial is defined as follows:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/400/1*7yUuDg9Wq04jlB1hGkFDIw.png" /><figcaption>The likelihood function for a single Bernoulli trial</figcaption></figure><p>The first thing we’re going to do is simulate a coin-flipping experiment. Let’s do it in Python. Here’s the code to generate 1000 random coin flips:</p><pre>import numpy as np<br>import matplotib.pyplot as plt</pre><pre>theta = 0.5 # a &quot;fair&quot; coin<br>num_flips = 1000 #simulate 1000 coin flips</pre><pre>#Python list comprehension<br>flips = [np.random.random() &lt; theta for flip in range(1000)]<br>print(f&#39;Total number of heads = {sum(flips)}&#39;)<br>&gt; Total number of heads = 499</pre><p>Unsurprisingly, even though the coin is perfectly fair, it comes up heads (very) slightly less than 50% of the time simply due to chance. We assume our coin flips are independent and identically distributed (i.i.d.), and derive the total log-likelihood:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/880/1*eKayV0CeUbPBpodlwJmBCw.png" /><figcaption>Derivation for the total log-likelihood for a series of (independent) coin flips</figcaption></figure><p>In the above equations, H and T, respectively, are the total number of heads and tails in the experiment. The final step in formulating the SGD update is to calculate the derivative of the likelihood with respect to the theta parameter:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/686/1*6Jwo3XdERLutPwaZfDIYeg.png" /></figure><p>In the above equation, we have divided by the total number of coin flips so that we can work with average values that are independent of the mini-batch size in our optimization step. The SGD update is now simply:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/1*GAAqVpV8zSg1OnYRWbY3Mw.png" /></figure><p>Here, alpha 𝛼 is the learning parameter. With all that setup here is the Python code to implement the SGD step:</p><pre>dtheta = lambda theta, h, t: (1/(h+t))*((h/t) - (t)/(1-theta))<br>theta = 0.4<br>alpha = 0.001<br>num_epochs = 500<br>batch_size = 100<br>thetas = []<br>thetas.append(theta_opt)</pre><pre>for epoch in range(num_epochs):<br>    rflips = np.random.permutation(flips)<br>    num_batches = int(np.floor(len(flips)/batch_size))<br>    for i in range(num_batches):<br>        batch = rflips[i*batch_size:(i+1)*batch_size-1]<br>        theta -= alpha*(1-epoch)/num_epochs * dtheta(theta,   sum(batch), batch_size-sum(batch))<br>        thetas.append(theta)<br>plt.plot(thetas)<br>print(f&#39;Final theta = {np.average(thetas[-batch_size*10:])}&#39;)<br>&gt; Final theta = 0.6089699285788563</pre><p>Note that I added the term (1-epoch)/num_epochs which slows down the learning rate and seems to help with convergence. I suggest playing around with all the other hyperparameters as well, num_epochs, batch_size, num_epochs. At any rate (no pun), here’s what the evolution of 𝜃 looks like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/378/1*EDnAuEvf8l6jUMnvw3IoAg.png" /><figcaption>Evolution of theta to convergence.</figcaption></figure><p>Here’s another convergence plot when starting at a theta=0.9:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/378/1*ZS7OVZbfZm3E7ZYMGkhbKA.png" /><figcaption>Starting at theta=0.9</figcaption></figure><p>That’s all for now. I hope this simple example of SGD helped you in your own learning process. Keep in mind this problem can be easily solved by deriving the maximum likelihood estimate. Simply set the derivative of the likelihood function to zero and solve for 𝜃. You should get H/T!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8e6475fd70cc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[2019 NBA Draft Board “Others”]]></title>
            <link>https://thecity2.medium.com/2019-nba-draft-board-others-a948133b2e8a?source=rss-949f03393eeb------2</link>
            <guid isPermaLink="false">https://medium.com/p/a948133b2e8a</guid>
            <category><![CDATA[nba-draft]]></category>
            <category><![CDATA[nba]]></category>
            <category><![CDATA[basketball]]></category>
            <category><![CDATA[sports]]></category>
            <dc:creator><![CDATA[Evan Zamir]]></dc:creator>
            <pubDate>Thu, 20 Jun 2019 19:05:52 GMT</pubDate>
            <atom:updated>2019-06-20T19:05:52.862Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*u1gWBT2q6J17l_1IrvzmAg.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@tjdragotta?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">TJ Dragotta</a> on <a href="https://unsplash.com/search/photos/basketball?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><ul><li><a href="https://medium.com/@thecity2/2019-nba-draft-board-tiers-1-3-bc12261629d8">Tiers 1–3 here</a></li><li><a href="https://medium.com/@thecity2/2019-nba-draft-board-tiers-4-5-64651a684d3">Tiers 4–5 here</a></li></ul><p>This last of 3 posts on the 2019 Draft is just going to be a list (with some brief thoughts sprinkled in) of players I like that I would target in the second round or as UDFA’s. <strong>Full disclosure</strong>: I have decided to leave out all the Europeans not named Goga because I simply didn’t have enough time this year to watch actual game footage — and I don’t want to make decisions simply based on highlights. If you’re wondering, I actually did watch Goga in some full game footage, so I felt like I had a good enough read on him. If a guy is on this list, I have watched at least a half of basketball. Consider the rankings here nominal, because essentially this is one big tier of guys. They are all “sleepers”.</p><p><strong>31 Daniel Gafford — Arkansas — 20y8mo</strong></p><p>I thought Gafford made a mistake coming back for his sophomore year, but I’m not sure in the end it changed his draft stock that much. He’s an athletic rim runner, and I don’t think he’s much more than that.</p><p><strong>32 Jalen McDaniels — San Diego St. — 21y4mo</strong></p><p>Another guy I was kind of hoping would declare last year, he didn’t really “explode” this year, and probably lost most of whatever intrigue he had going for him. Still, he’s got length, some creation skills, and a good motor.</p><p><strong>33 Keldon Johnson — Kentucky — 19y8mo</strong></p><p>I’m not really into Keldon at all, but there’s some chance he grows into a rotation player who can shoot and defend well enough to get a second contract.</p><p><strong>34 Dylan Windler — Belmont — 22y8mo</strong></p><p>May go way higher than warranted based on a good tournament showing. He’s a good shooter, but I’m not sure he’s strong enough to be a 4 or quick enough to be a 3. Also doesn’t seem like much of a playmaker. And he’s pretty old. But he is a good shooter, which always has some value.</p><p><strong>35 Alen Smailagic — Santa Cruz Warriors — 18y10mo</strong></p><p>Really young and talented defender the Warriors are <a href="https://www.forbes.com/sites/jefffeld/2018/12/27/alen-smailagic-is-the-warriors-draft-target-they-dont-want-you-to-know-about-yet/">trying to hide in the GL</a>. Really intriguing if he can develop a consistent outside shot.</p><p><strong>36 Ty Jerome — Virginia — 21y11mo</strong></p><p>Very good college basketball player. I have a lot of concerns about how he translates to the NBA simply because he’s so damn unathletic.</p><p><strong>37 Dedric Lawson — Kansas — 21y8mo</strong></p><p>You think Ty Jerome is unathletic? Dedric Lawson says hold my beer. Yet he is a great basketball player who does a lot of things well. Personally, I hope he makes it in the league.</p><p><strong>38 Isaiah Roby — Nebraska — 21y4mo</strong></p><p>Kind of a versatile 4 or small ball 5 (eventually). Huge hands and a 7&#39;1&quot; wingspan combined with some playmaking skills. If he can shoot it, he’s a player to watch.</p><p><strong>39 Eric Paschall — Villanova — 22y7mo</strong></p><p>He’s old and did not look as good this year not being surrounded by as much talent, but Paschall is a solid big who knows how to play, might defend a bit, and shoot a bit. Could have some value if he is not overdrafted.</p><p><strong>40 Carsen Edwards — Purdue — 21y3mo</strong></p><p>A natural scorer who is super strong for his size and has a decent wingspan. He’s a combo guard but there’s no doubt he might bring some value off the bench. Again, not a guy I’d overdraft.</p><p><strong>41 Sagaba Konate — West Virginia — 22y3mo</strong></p><p>Another guy should have come out last year. Injuries derailed him, but he’s one of the rare guys who is a potential 3&amp;B(lock) big. Not a lot of 3pt attempts, but great free throw shooting suggests potential there.</p><p><strong>42 Tremont Waters — LSU — 21y5mo</strong></p><p>One of my favorite players in the draft. He’s just a couple inches too small and maybe not quite as good a shooter as he needs to be to run a team, but I have a feeling he will stick around as a backup point guard for a while. He’s really fun to watch though.</p><p><strong>43 Justin Robinson — Virginia Tech — 21y8mo</strong></p><h3>I&#39;m not Patrick Beverley 🏀 on Twitter</h3><p>Just putting this out there in case he ever hits with an NBA team. I like Justin Robinson from VT. Wouldn&#39;t draft him but he could get a shot in SL as a UDFA. Plays like this pretty special for a guy his size. Competitive little guy. https://t.co/bYZ15744tZ</p><p>My favorite “find” of the year. He was the engine that made VaTech run. Jumped off the screen every time I watched him with cross-court passes and effort plays. I love him. He’s one of the top UDFA targets for me.</p><p><strong>44 Justin Wright-Foreman — Hofstra — 21y7mo</strong></p><p>A question I pose to you is why draft Carsen Edwards when you might be able to get JWF as a UDFA?</p><p><strong>45 Dean Wade — Kansas St — 22y7mo</strong></p><p>Dude can shoot. Good UDFA target.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a948133b2e8a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[2019 NBA Draft Board Tiers 4–5]]></title>
            <link>https://thecity2.medium.com/2019-nba-draft-board-tiers-4-5-64651a684d3?source=rss-949f03393eeb------2</link>
            <guid isPermaLink="false">https://medium.com/p/64651a684d3</guid>
            <category><![CDATA[basketball]]></category>
            <category><![CDATA[nba-draft]]></category>
            <category><![CDATA[sports]]></category>
            <category><![CDATA[nba]]></category>
            <dc:creator><![CDATA[Evan Zamir]]></dc:creator>
            <pubDate>Thu, 20 Jun 2019 01:28:33 GMT</pubDate>
            <atom:updated>2019-06-20T01:28:33.937Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JwFE8zRrKFi5oojgx4XYpQ.jpeg" /><figcaption>Photo by <a href="https://www.pexels.com/@rawpixel?utm_content=attributionCopyText&amp;utm_medium=referral&amp;utm_source=pexels">rawpixel.com </a>from <a href="https://www.pexels.com/photo/grayscale-photography-of-person-holding-basketball-1451362/?utm_content=attributionCopyText&amp;utm_medium=referral&amp;utm_source=pexels">Pexels</a></figcaption></figure><p>(<a href="https://medium.com/@thecity2/2019-nba-draft-board-tiers-1-3-bc12261629d8">Draft Board Tiers 1–3 here</a>)</p><p>This is not a mock draft. It’s my big board — a ranking of draft-eligible players roughly in the order I estimate their talent level/usefulness to NBA teams. As a point of emphasis, tiers are more important than rankings within tiers. The tier labels correspond primarily to perceived upside value — as opposed to a player’s floor. Of course, some players will invariably play above or below the tiers as I’ve constructed them. Hopefully, I am within +/- 1 tier when it’s all said and done! I wish I had time to write up more than these little blurbs, but hopefully, my thoughts will be somewhat useful to you — and perhaps, different in some little way from other draft profiles you’ve read. At the end of each prospect blurb I give subjective rankings in the following 10 offensive and defensive categories that I find useful to think about (scale is 1–5 for each):</p><h3>Offense</h3><ul><li>Spacer — ability to space the floor (“gravity”) with shooting off the dribble or on the move.</li><li>Facilitator — ability to run an offense or keep an offense flowing (ball stopping dribble pounders will have low scores here).</li><li>Creator — ability to create shots off the dribble, coming off screens, rolling to the basket, cutting, etc.</li><li>Finisher — ability to finish shots, regardless of shot location (but taking into account relative efficiencies of 3pt vs 2pt, layups vs mid-range, etc).</li><li>Rebounder — should be self-explanatory</li></ul><h3>Defense</h3><ul><li>Stopper — point of attack defense, penetration stopper.</li><li>Protector — rim protector + ability to hold position in paint and keep opposing big men from establishing deep position.</li><li>Exploder — (“play exploder”) ability to deflect passes, create turnovers off-ball, generally wreak havoc on opposing offenses</li><li>Extender — (“play extender”) ability to switch, hard closeouts to the 3pt line, team defense</li><li>Rebounder — self-explanatory</li></ul><h3>Tier 4 — Starter/High-end Rotation</h3><p><strong>12 P.J. Washington — Kentucky — 20y9mo</strong></p><p>PJ is the prototypical “solid” prospect. I think he tops out as a good — but not great — starter for a team, based on his combination of strength, length, and well, solid skillset for a 4 or small ball 5. David West and Paul Millsap come to mind as optimistic comps — think 95%-ile outcomes.</p><ul><li>Spacer — 3 Facilitator — 3 Creator — 3 Finisher — 3 Rebounder — 4</li><li>Stopper — 2 Protector — 3 Exploder — 3 Extender —3 Rebounder — 4</li></ul><p><strong>13 Coby White — North Carolina — 19y4mo</strong></p><p>I had Coby quite a bit higher than this for most of the season, but I chickened out. I really like his shooting and creation ability, but at the end of the day, I don’t think I’m convinced that he can be a lead guard on a good team or a starting wing with his limitations on size-wise on defense. I can’t quite get there, totally get it if you can. I will not be mad if he proves me wrong. My gut tells me he tops out as a useful bench scorer.</p><ul><li>Spacer — 5 Facilitator — 4 Creator — 5 Finisher—4 Rebounder — 2</li><li>Stopper — 2 Protector — 1 Exploder — 3 Extender — 2 Rebounder — 2</li></ul><p><strong>14 Brandon Clarke — Gonzaga — 22y8mo</strong></p><p>I like the idea of Brandon Clarke as a crazy pogostick finisher and weakside rim protecting nightmare for opponents — heck, he might even shoot it eventually — but his severe lack of length — I mean even if he were a wing he wouldn’t be considered long — and near eligibility for AARP just don’t let me push him up much higher than here. He deserves to be good, if that makes any sense, I just don’t fully buy into it.</p><ul><li>Spacer — 2 Facilitator — 3 Creator — 2 Finisher — 5 Rebounder — 4</li><li>Stopper — 4 Protector — 4 Exploder — 4 Extender — 4 Rebounder — 4</li></ul><p><strong>15 Bol Bol — Oregon — 19y7mo</strong></p><p>You might not have known that Darius Garland is the son of former Warrior Winston Garland, but you almost surely know that Bol Bol also plays basketball. While strength is one of Bol’s weaknesses, he does have legitimate strengths that might offset his, er, weakness. (Note to self, improve sentence construction.) Seriously, though, somewhat like his father Manute who was known for shooting 3’s way before it was fashionable for centers to do so (enabled by the great Don Nelson), Bol the latter possesses a surprisingly high level of basketball skill for a player size. Although his season was shortened by injury (another major concern, but not one I’ll worry about here), he shot 63% TS on 33% USG to go along with a 12.4% BLK rate, which gives some idea how unusual a prospect he is. He can handle the ball and create his own offense better than most wings half his size (not sure that’s even exaggerating). Not the best passer, but not a complete black hole probably. Hopefully. Worth the risk around here.</p><ul><li>Spacer — 5 Facilitator — 2 Creator — 4 Finisher — 5 Rebounder — 5</li><li>Stopper — 3 Protector — 5 Exploder — 3 Extender — 4 Rebounder — 5</li></ul><p><strong>16 Cam Reddish — Duke — 19y9mo</strong></p><p>I’m not a huge believer in Cam’s upside, but I could certainly see him starting or coming off the bench as a 3&amp;D 4 and helping a rotation eventually. He just doesn’t seem to have much dynamism to his game and I think he’s way overhyped for what he is. He “looks” like a star, whatever that means. Seriously, what do people mean? I get a lot of Harrison Barnes vibes with Cam which scares me. I like his high steal rate, though.</p><ul><li>Spacer — 4 Facilitator — 3 Creator — 2 Finisher — 3 Rebounder — 3</li><li>Stopper — 4 Protector—3 Exploder — 4 Extender — 4 Rebounder — 3</li></ul><p><strong>17 De’Andre Hunter — Virginia — 21y6mo</strong></p><p>Honestly, take whatever I said about Cam above and apply that to Hunter with a bit less animosity and take out the compliment about steals. Cam is ever so slightly ahead simply by virtue of being younger and, perhaps, being a bit better facilitator. The thing I’ll say for Hunter that I appreciate is he seems to know himself as a player and looks unflappable on the court.</p><ul><li>Spacer — 4 Facilitator — 2 Creator — 2 Finisher — 3 Rebounder — 3</li><li>Stopper — 4 Protector — 2 Exploder — 2 Extender — 4 Rebounder — 3</li></ul><p><strong>18 Talen Horton-Tucker — Iowa St. — 18y6mo</strong></p><p>One of the few “fat upside” guys in this Draft. THT has a lot of intrigue going for him — a combination of youth, length for the position and high skill floor (ceiling?). This is a high variance prospect. If he gets in bonkers shape and hits on shooting, he might be really good. If not, he could flop. I could easily talk myself into putting him significantly higher, because I do like the bet on ceiling, but man his shot seems so bricky to me. I’ll keep him here.</p><ul><li>Spacer — 3 Facilitator — 4 Creator — 3 Finisher — 3 Rebounder — 3</li><li>Stopper — 4 Protector — 3 Exploder — 4 Extender — 4 Rebounder — 3</li></ul><p><strong>19 Matisse Thybulle — Washington — 22y3mo</strong></p><p>Welcome to “Crazy Stats” prospect. I really thought Thybulle would test the waters in the 2018 Draft, but he came back for another year for whatever reason. He proceeded to put up even more crazy steals and blocks numbers for a wing. Combination of athleticism, length, and insane anticipation instincts and plays in a system that enabled him to gamble a ton. There’s a lot to be skeptical about with him, but…what if his defense actually does translate and he shoots a bit (his 78% career free throw shooting is encouraging)? That’s a really valuable player obviously. Also, I think he’s an underrated playmaker.</p><ul><li>Spacer — 3 Facilitator — 3 Creator — 2 Finisher — 4 Rebounder — 3</li><li>Stopper — 4 Protector — 3 Exploder — 5 Extender — 5 Rebounder — 3</li></ul><h3>Tier 5 (Rotation Players)</h3><p>Warning: At this point, my takes will get terser (and probably meaner).</p><p><strong>20 Romeo Langford — Indiana — 19y7mo</strong></p><p>I like his size, and not doubting his ability to bucket-get, but he’s only going to go as far as his shot and effort on d take him. Hopefully better than Rodney Stuckey.</p><ul><li>Spacer — 2 Facilitator — 3 Creator — 5 Finisher — 5 Rebounder — 2</li><li>Stopper — 2 Protector — 2 Exploder — 3 Extender — 3 Rebounder — 2</li></ul><p><strong>21 Kevin Porter Jr — USC — 19y1mo</strong></p><p>I like his size, and not doubting his ability to bucket-get, but he’s only going to go as far as his shot and effort on d take him. Hopefully better than Nick Young.</p><ul><li>Spacer — 4 Facilitator — 2 Creator — 5 Finisher — 5 Rebounder —2</li><li>Stopper — 3 Protector — 2 Exploder — 3 Extender — 3 Rebounder — 2</li></ul><p><strong>22 Nickeil Alexander-Walker — Virginia Tech — 20y9mo</strong></p><p>Better shooting, but worse overall basketball player of the basketball playing Alexander-Gilgeous-Alexander-Walker-Alexander cousins. Do like his length and 3pt shooting potential. Whereas Shai is a wingy-guard, Nickeil seems more like a guardy-wing. Didn’t look as good to me when Justin Robinson was out, which is not surprising given his increased responsibilities.</p><ul><li>Spacer — 4 Facilitator — 3Creator — 3 Finisher — 3 Rebounder — 2</li><li>Stopper — 3 Protector — 2 Exploder — 4 Extender — 4 Rebounder —2</li></ul><p><strong>23 Tyler Herro — Kentucky — 19y4mo</strong></p><p>We don’t need another Her — sorry, I had to slap myself just now. There’s a good chance Herro will be an elite shooter, and if he turns out to be “next JJ Redick” I’ll look pretty dumb having him this low. The thing is I don’t really see that much other than the shooting (and to be fair, his 3PT% was surprisingly pedestrian given his virtually flawless mechanics). Also, his short arms just seem so limiting on defense for being essentially a small wing. So yeah, call me skeptical. But I totally do get why you might have him higher.</p><ul><li>Spacer — 5 Facilitator — 2 Creator — 3 Finisher — 4 Rebounder — 1</li><li>Stopper — 2 Protector — 1 Exploder — 2 Extender — 2 Rebounder — 1</li></ul><p><strong>24 Iggy Brazdeikis — Michigan — 20y5mo</strong></p><p>Really good size, strength, and length for a wing to the point where he might be considered a big wing or maybe a small ball 4 eventually. He kind of reminds me of Tom Tolbert physically — do you ever get tired of my former Warriors references in these profiles? Defense and playmaking will be concerns, but shooting and motor will keep him in the league somehow.</p><ul><li>Spacer — 4 Facilitator — 2 Creator — 3 Finisher — 3 Rebounder — 3</li><li>Stopper — 2 Protector — 2 Exploder — 2 Extender — 3 Rebounder — 3</li></ul><p><strong>25 Cameron Johnson — North Carolina — 23y3mo</strong></p><p>Injury concerns and age aside, Cameron Johnson is probably one of the top 3–4 shooters in this class in a vacuum (which is how I’m told they do testing for that). He also has a knack for timely playmaking and “winning”. Some concerns on defense given lack of athleticism, but he does have good length and he’s smart which could help him get by on that end.</p><ul><li>Spacer — 5 Facilitator — 3 Creator — 2 Finisher — 4 Rebounder — 3</li><li>Stopper —3 Protector — 2 Exploder — 2 Extender — 3 Rebounder — 3</li></ul><p><strong>26 Bruno Fernando — Maryland — 20y10mo</strong></p><p>NBA-ready body, efficient scorer and finisher, decent passer for his position, and probably versatile enough to play in different offensive systems. Doesn’t really have an outside shot now, but he’s a great free throw shooter, and it wouldn’t surprise me at all if he’s shooting 3’s eventually, which would be huge for his game because he can put the ball on the deck a bit. He’s got a lot going for him, but for some reason, it feels like the whole has been somewhat less than the sum of his parts. Supposedly has a great work ethic, but he doesn’t always seem as tough as he looks on the court. Could definitely see him overachieving his draft position if everything clicks.</p><ul><li>Spacer — 3 Facilitator — 3 Creator — 3 Finisher — 5 Rebounder — 5</li><li>Stopper —2 Protector — 5 Exploder — 3 Extender — 3 Rebounder —5</li></ul><p><strong>27 Mfiondu Kabengele — Florida St — 21y10mo</strong></p><h3>I&#39;m not Patrick Beverley 🏀 on Twitter</h3><p>Kabengele really might end up a first round pick at this rate. Maybe like this year&#39;s Omari (in terms of tourney boost) if FSU goes a bit further.</p><p>Well, it happened. I was in on Kab early — even starting <a href="https://forums.realgm.com/boards/viewtopic.php?f=3&amp;t=1790437">the Kab thread</a> on RealGM back in January (a fun read btw)— but I think the hype has gotten a bit out of a control. Kab is like “One job; Did it” personified. He’s a big man bucket getter and can do it in a bunch of ways, and he may even end up being a very good shooter. It’s fun to watch Leonard Hamilton inexplicably bring Kab off the bench and unleash him as a bench scorer — and actually, that may end up being his best role in the NBA. He’s not as athletic or mobile as people are telling you. He’s kind of like Thomas Bryant to me, a bit worse on offense probably, but better on defense. Oh, and he’s Dikembe’s nephew. I am contractually obligated to mention that every time I talk about him.</p><p>Oh yeah, sorry. I forgot to mention he doesn’t pass. He had 11 assists. This season. Remember: One job. Did it. He will block some shots, so let’s call it Two jobs.</p><ul><li>Spacer — 3 Facilitator — 1 Creator — 4 Finisher — 5 Rebounder — 5</li><li>Stopper — 2 Protector — 5 Exploder — 3 Extender — 3 Rebounder — 5</li></ul><p><strong>28 Rui Hachimura — Gonzaga — 21y4mo</strong></p><p><a href="https://twitter.com/thecity2/status/1059863662477758464">https://twitter.com/thecity2/status/1059863662477758464</a></p><p>I think Rui is destined to be a bench scorer/energy big. I will say for the record, I think there’s a decent chance he becomes a 3pt threat. We’re probably looking at a Jabari Parker more or less — to be clear, not a compliment at this point — just don’t pay him like one and you’re fine. Lack of defense and playmaking will keep him from becoming a real star (could be a fake one if he scores enough).</p><ul><li>Spacer — 3 Facilitator — 2 Creator — 3 Finisher — 4 Rebounder — 3</li><li>Stopper — 2 Protector — 2 Exploder — 2 Extender — 3 Rebounder — 3</li></ul><p><strong>29 Nassir Little — North Carolina — 19y4mo</strong></p><p>Ok, folks, I guess we’ve come to the “Don’t overthink it” portion of my board. I thought I was going to be a Nas fan before the season started, but like most everyone else, I was disappointed. Good body and motor, but not much else to hang your hat on. But maybe that’s enough at this point in the Draft. He better bring it on d every night.</p><p>Also 77% free throw shooter…maybe there’s a 3pt shooter in there eventually.</p><ul><li>Spacer — 2 Facilitator — 2 Creator — 2 Finisher — 4 Rebounder — 4</li><li>Stopper — 4 Protector — 2 Exploder — 3 Extender — 4 Rebounder — 4</li></ul><p><strong>30 Shamorie Ponds — St. Johns — 20y11mo</strong></p><p>I’ve gone back and forth with Ponds a lot. Is he good enough to be a starting point guard on an average-ish, let alone a playoff team? Most of the time I’m thinking not, but he definitely makes high level plays passing and scoring-wise, and he has the requisite “very high steal rate” to be just intriguing enough that I’m going to take him here. On the other side of 30 in this draft you’re not missing much if he fails — i.e. is just another backup point/combo guard. There’s a small chance he’s more than that, and I guess it’s worth the gamble.</p><ul><li>Spacer — 4 Facilitator — 5 Creator — 5 Finisher — 4 Rebounder — 1</li><li>Stopper — 1 Protector — 1 Exploder — 3 Extender — 2 Rebounder — 1</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=64651a684d3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[2019 NBA Draft Board Tiers 1–3]]></title>
            <link>https://thecity2.medium.com/2019-nba-draft-board-tiers-1-3-bc12261629d8?source=rss-949f03393eeb------2</link>
            <guid isPermaLink="false">https://medium.com/p/bc12261629d8</guid>
            <category><![CDATA[zion-williamson]]></category>
            <category><![CDATA[nba]]></category>
            <category><![CDATA[ja-morant]]></category>
            <category><![CDATA[nba-draft]]></category>
            <category><![CDATA[basketball]]></category>
            <dc:creator><![CDATA[Evan Zamir]]></dc:creator>
            <pubDate>Wed, 19 Jun 2019 06:12:38 GMT</pubDate>
            <atom:updated>2019-06-19T06:12:38.037Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YMNhJE3_oEia6lpQak4Mww.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@diante21?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">DiAnte Squire</a> on <a href="https://unsplash.com/search/photos/basketball?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>This is not a mock draft. It’s my big board — a ranking of draft-eligible players roughly in the order I estimate their talent level/usefulness to NBA teams. As a point of emphasis, tiers are more important than rankings within tiers. The tier labels correspond primarily to perceived upside value — as opposed to a player’s floor. Of course, some players will invariably play above or below the tiers as I’ve constructed them. Hopefully, I am within +/- 1 tier when it’s all said and done! I wish I had time to write up more than these little blurbs, but hopefully, my thoughts will be somewhat useful to you — and perhaps, different in some little way from other draft profiles you’ve read. At the end of each prospect blurb I give subjective rankings in the following 10 offensive and defensive categories that I find useful to think about (scale is 1–5 for each):</p><h3>Offense</h3><ul><li>Spacer — ability to space the floor (“gravity”) with shooting off the dribble or on the move.</li><li>Facilitator — ability to run an offense or keep an offense flowing (ball stopping dribble pounders will have low scores here).</li><li>Creator — ability to create shots off the dribble, coming off screens, rolling to the basket, cutting, etc.</li><li>Finisher — ability to finish shots, regardless of shot location (but taking into account relative efficiencies of 3pt vs 2pt, layups vs mid-range, etc).</li><li>Rebounder — should be self-explanatory</li></ul><h3>Defense</h3><ul><li>Stopper — point of attack defense, penetration stopper.</li><li>Protector — rim protector + ability to hold position in paint and keep opposing big men from establishing deep position.</li><li>Exploder — (“play exploder”) ability to deflect passes, create turnovers off-ball, generally wreak havoc on opposing offenses</li><li>Extender — (“play extender”) ability to switch, hard closeouts to the 3pt line, team defense</li><li>Rebounder — self-explanatory</li></ul><h3>Tier 1 — All-NBA/MVP</h3><p>1 <strong>Zion Williamson — Duke — 18y11mo</strong></p><h3>I&#39;m not Patrick Beverley 🏀 on Twitter</h3><p>This is nitpicking Zion, but my only real concern with him (let&#39;s say other than shooting, duh) is that I think his hands are not that big. He compensates obviously with being able to jump 2 feet over the rim, but that&#39;s why he&#39;s always dunking with his hand on top of the ball.</p><p>As evidenced by the above tweet I had to struggle to come up with <em>anything</em> negative to say about Zion this season. He’s simply an incredible prospect. The crazy thing is that I actually think he might be a better <em>basketball player</em> than people give him credit for — as opposed to seeing him purely as a uniquely gifted athletic specimen, which he absolutely is, of course. I think he will go as far as his basketball IQ and conditioning/health takes him because I’m pretty sure about everything else. This is as easy a number 1 pick as it gets.</p><ul><li>Spacer — 3 Facilitator — 4 Creator — 5 Finisher — 5 Rebounder — 4</li><li>Stopper — 5 Protector — 4 Exploder — 5 Extender — 4 Rebounder — 4</li></ul><h3>Tier 2 — All-Stars</h3><p><strong>2 R.J. Barrett — Duke — 19y0mo</strong></p><h3>I&#39;m not Patrick Beverley 🏀 on Twitter</h3><p>I&#39;m going to go out on a limb and say Barrett is going to be better than Andrew Wiggins.</p><p>I think Barrett ends up quite a bit better than Wiggins — even if he disappoints me. Barrett is a bit like Marvin Bagley last year — not physically, please hear me out — in the sense that even though both are player archetypes I’m typically not enamored with, there is no denying the productivity — at least, in terms of bucket-getting. I think Barrett showed more passing ability than I expected as the season progressed to the point where I think that will eventually be a considerable strength of his. I suspect his first few seasons in the league will be about him proving he’s “the man”, but my hope is he continues to develop the rest of his game — because I think lurking underneath the bravado and killer instinct is a pretty talented and smart player.</p><ul><li>Spacer —3 Facilitator — 3 Creator — 4 Finisher — 5 Rebounder — 3</li><li>Stopper — 3 Protector — 2 Exploder — 3 Extender — 4 Rebounder — 3</li></ul><p><strong>3 Jarrett Culver — Texas Tech — 20y3mo</strong></p><h3>I&#39;m not Patrick Beverley 🏀 on Twitter</h3><p>I really hope Culver is better than Evan Turner. A lot of that will depend on his shooting. Most likely a better defender too.</p><p>Statistically, and perhaps, in some other ways, Culver is eerily similar to ET — but that doesn’t doom him necessarily. Culver is a smooth, if not an explosive athlete who has an uncanny ability to slither through and carve up defenses — especially adept at using picks for his size. In some ways, he’s a bit like a bigger more wingy version of Shai Gilgeous-Alexander. I’m not exactly sure how it will play out for Culver at the next level, but after a few months of falling out of love with him, I think I’m back to seeing All-Star upside if he hits on his shooting. Regardless, he should provide legit value on both ends of the court during the regular season and Playoffs.</p><ul><li>Spacer — 3 Facilitator — 4 Creator — 3 Finisher — 4 Rebounder — 3</li><li>Stopper — 4 Protector — 3 Exploder — 4 Extender — 5 Rebounder — 3</li></ul><p>4 <strong>Darius Garland — Vanderbilt — 19y4mo</strong></p><p>Let me get this out of the way…I have a Warriors bias going on here for two reasons: 1) Garland’s dad Winston played for the Warriors when I was a kid. 2) I can’t help but think about Steph when I watch Darius on and off the court. Darius is a super skilled and super-cute babyface assassin with a killer personality just like Steph! <em>Ok, ok. I’ll stop.</em> But seriously, I get all the reasons people have not to be high on DG. He could definitely fail to live up to being a top 5 pick. But in this draft, which feels really weak to me, I think you have to place some high variance bets, and Garland just feels like one of the better ones to make with his general skill level. If I’m a real GM, maybe I don’t make this pick, but I’m a fake GM, so here we are.</p><ul><li>Spacer — 5 Facilitator — 4 Creator — 5 Finisher — 4 Rebounder — 1</li><li>Stopper — 1 Protector — 1 Exploder — 2 Extender — 2 Rebounder — 1</li></ul><p>5 <strong>Ja Morant — Murray St — 19y10mo</strong></p><p>Given that Ja is likely to be the #2 pick, I think I don’t need to explain much more. I guess I’m not quite as high on him as the consensus, because I’m not sure his shooting ever reaches truly elite levels, but he can be a star regardless given his vision, playmaking ability, handles and general athletic goodness — the latter may be a bit overrated but still seems quite considerable to my eyes. Hopefully, he gains a bit more strength.</p><ul><li>Spacer — 4 Facilitator — 5 Creator — 5 Finisher — 5 Rebounder — 1</li><li>Stopper — 1 Protector — 1 Exploder — 2 Extender — 2 Rebounder — 1</li></ul><h3>Tier 3 — High-End Starters</h3><p><strong>6 Jontay Porter—Missouri — 19y7mo</strong></p><p>If you follow me at all, you know I absolutely hate Jontay Porter. What a dick.</p><p>Ok, so here’s the thing. Jontay’s a mess right now. His injuries have injuries. But I absolutely love Jontay’s basketball talent — especially his ability to make plays and handle the ball for his size. I decided that I’m just going to rank him as high as I reasonably could even despite all his myriad injury issues. I would have ranked Jontay 5 or 6 in last year’s stacked draft, so in the end, it’s pretty easy for me to put him here. There’s no chance a sane GM would take him this high, but this slot just reflects where his talent level is for me. You can discount him as far down as you want for injury reasons, but if he ever does get fully healthy and fit — great news to hear he measured at 8.5% BF at the Combine — he’ll be a winning player for a very lucky team. He does so many things right.</p><ul><li>Spacer — 5 Facilitator — 5 Creator — 3 Finisher — 3 Rebounder — 3</li><li>Stopper — 2 Protector — 4 Exploder — 3 Extender — 3 Rebounder — 3</li></ul><p><strong>7 Grant Williams — Tennessee — 20y6mo</strong></p><p>Grant is like a shorter chiseled football player version of Jontay that isn’t quite as talented in any one area. But that still makes him good enough to be up this high. He’s a winning player and bonus he’s a fairly young Junior — for example, he’s a couple of months younger than P.J. Washington. Right now Grant is not much of a 3pt shooter but given his FT% and touch I think that will come in time.</p><ul><li>Spacer — 3 Facilitator — 4 Creator — 4 Finisher — 3 Rebounder — 4</li><li>Stopper — 3 Protector — 4 Exploder — 3 Extender — 4 Rebounder — 4</li></ul><p><strong>8 Goga Bitadze — Buducnost — 19y10mo</strong></p><p>I started doing research on Goga last year when I thought he was going to be in the 2018 Draft. At this point he just feels like one of those no-brainer Euro bigs that will probably be drafted like 10–15 slots below where he should be. When he fully matures he’s probably going to anchor a defense and stretch the floor a bit. I don’t know if he’s a Gasol or a Jokic or even a Nurkic, but he’s probably something — and have I told you this is a weak draft?</p><ul><li>Spacer — 4 Facilitator — 3 Creator — 3 Finisher — 4 Rebounder — 5</li><li>Stopper — 3 Protector — 5 Exploder — 3 Extender — 4 Rebounder — 5</li></ul><p><strong>9 Chuma Okeke — Auburn — 20y10mo</strong></p><p>Admittedly, Chuma did not get on my radar until maybe a month or two before the Tourney this year, but he quickly rose in my personal rankings once he did. Not sure if he’s a big wing or a small big, but he checks a lot of boxes — shooting, passing, some ball handling ability, team defense. It’s unfortunate that he had a major injury a few months before the Draft and may miss his entire rookie season, but like Jontay, I’m really just ranking him based on talent. You can apply your own injury discounting from here.</p><ul><li>Spacer — 4 Facilitator — 4 Creator — 3 Finisher — 4 Rebounder — 3</li><li>Stopper — 3 Protector — 3 Exploder — 3 Extender — 4 Rebounder — 3</li></ul><p><strong>10 Jaxson Hayes — Texas — 19y0mo</strong></p><p>Jaxson Hayes was kind of the hipster #DraftTwitter pick early in the NCAA season this year, but for legit reasons. He has a great athletic profile, requisite size and huge football-catching able hands. I’m not sure about the rest of his game — I’m openly skeptical about his passing ability — but the good stuff is good enough to squelch those doubts and take him here.</p><ul><li>Spacer — 2 Facilitator — 1 Creator — 2 Finisher — 5 Rebounder — 5</li><li>Stopper—3 Protector — 5 Exploder — 2 Extender — 4 Rebounder — 5</li></ul><p><strong>11 Nicolas Claxton — Georgia — 20y2mo</strong></p><p><a href="https://twitter.com/thecity2/status/1094273013850071041">https://twitter.com/thecity2/status/1094273013850071041</a></p><p>If Jaxson Hayes was the hipster pick early in the season, Claxton might be that guy now. As the above tweet shows, I was in on Claxton before he started gaining much Draft momentum. Although similarly sized to Hayes they are very different players and prospects. Claxton is kind of a weirdo big who was asked to basically run the offense at Georgia (weird!). Of course, he won’t be responsible for running an offense in the NBA, but the fact that he can do a lot of different things is super intriguing to me. On defense, he kind of reminds me of a Kevon Looney — who by the way, was a considerably high ranked prospect at UCLA before serious injury (and asthma) questions arose. I’m glad Claxton ended up declaring, now he has to prove me right!</p><ul><li>Spacer — 3 Facilitator — 3 Creator — 3 Finisher — 4 Rebounder — 4</li><li>Stopper — 4 Protector — 4 Exploder — 3 Extender — 4 Rebounder — 4</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bc12261629d8" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>