<?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 Fay Wu on Medium]]></title>
        <description><![CDATA[Stories by Fay Wu on Medium]]></description>
        <link>https://medium.com/@mfaywu?source=rss-ec098f820f59------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*MWwT5qYOMng1M7qv.jpg</url>
            <title>Stories by Fay Wu on Medium</title>
            <link>https://medium.com/@mfaywu?source=rss-ec098f820f59------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 07:45:56 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mfaywu/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[Some notes from GitHub Universe 2016]]></title>
            <link>https://medium.com/@mfaywu/some-notes-from-github-universe-2016-5c5f1c3be319?source=rss-ec098f820f59------2</link>
            <guid isPermaLink="false">https://medium.com/p/5c5f1c3be319</guid>
            <category><![CDATA[github]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[remote-working]]></category>
            <category><![CDATA[design]]></category>
            <dc:creator><![CDATA[Fay Wu]]></dc:creator>
            <pubDate>Sun, 18 Sep 2016 03:43:12 GMT</pubDate>
            <atom:updated>2016-09-18T03:43:12.039Z</atom:updated>
            <content:encoded><![CDATA[<p>Last week I was able to go to GitHub Universe thanks to their student scholarships! We use GitHub frequently at school but I didn’t know much about the company. This conference left a big impression that GitHub is dedicated to putting in time and financial resources into increasing diversity and access in the world of software. This year, they threw a concert on the first night with all proceeds benefiting <a href="http://www.blackgirlscode.com/">Black Girls Code</a> which provides opportunities to young women of colour age 7–17. I’d like to share a few of my notes from the conference.</p><h3>Managing a Distributed Team</h3><p>This talk was given by an engineer from Travis CI, which is a testing tool company based in Berlin. I was impressed by their diversity with 52% women on the team and 52% of those in engineering. The engineer talked about the challenges of managing a distributed team where most engineers are remote: with communication, different cultural backgrounds, and establishing a company culture.</p><p><strong>Communication</strong></p><p>Probably the statement that left the biggest impact on me was “If it’s not written down, it doesn’t exist”. With distributed teams, it seems that documentation comes naturally because most, if not all, of the communication is recorded online in message channels like Slack or issues in GitHub. This can actually be really beneficial as opposed to making decisions over phone call or in person because then all the thoughts are written down for anyone else to be able to view and understand — “accidental documentation”. I found it amusing that Travis files issues for almost everything, including book requests, coffee and tea subscriptions, etc. This creates visibility for all decisions. The downside is that the tone can be lost in written communication and certain writing styles can be mistakenly interpreted as passive aggressive or something else.</p><p><strong>Cultural Backgrounds</strong></p><p>The same statement can mean different things to different people with a lot of it depending on their cultural backgrounds. When working with people from different cultural backgrounds (which hopefully we all are because diversity), it’s important to remember that English can be really ambiguous and consider what the meanings of the statements we are sending out are in different cultural contexts.</p><p><strong>Company Culture</strong></p><p>With a distributed team, company culture actually has to be documented and on the internet. An example of this is promoting gender inclusiveness by having a “gender pronoun” box for peoples’ preferred pronouns on their Slack profiles. Since employees are missing face time, culture can get lost as new people join, and it can be confusing what the guidelines are. Having a company offsite can bring everyone together to establish the company culture, then document it to refer back throughout the year.</p><p>I thought it was super interesting that the speaker said that even though a majority of Travis employees are remote, it is still not a remote-first company because the in-house employees get more perks that remote employees don’t, such as team lunches in the office.</p><p>Someday I want to work remotely for a distributed team because you could get to work with the best engineers independent of location that way, but for now I think I need to gain more skills in becoming autonomous and productive with little to no guidance. I think even if we’re not working on remote teams, we can learn from how distributed teams retain tribal knowledge and how to communicate effectively.</p><h3>99% Invisible with Roman Mars</h3><p>We watched a live podcast recording at the end of the second day about unpleasant design. <a href="http://unpleasant.pravi.me/">Unpleasant design</a> is a form of social control through design, used in urban public spaces to create hostile environments. An example is the use of lighting, which is the earliest form of public social control. Pink lighting preys on male teenagers insecurity so that they don’t linger in the public, and blue light in public bathrooms prevent people from seeing their veins so they are unable to take drug injections.</p><p>Another unpleasant design is spikes, which are used on the edges of buildings and on roofs. This prevents people from sleeping on the outside of buildings and also prevents pigeons from gathering in these areas. Spikes can be “a symbol of a heartless approach to homelessness”. But it only causes pigeons to congregate in the only small spaces without spikes!</p><p>Public seating is also a great example of unpleasant design, especially the <a href="https://www.behance.net/gallery/The-Camden-Bench/1498859">Camden bench</a>. This bench won numerous awards because the design makes it anti-theft, anti-sleeping, anti-litter, anti-graffiti, anti-skateboard, and anti-anything-else-but-sitting.</p><p>I thought it was very interesting that this talk was in the conference schedule, but I’m guessing that it was meant to talk about how the way we design things affects how people use the product and that we’re responsible for designing products that encourage inclusiveness and access, not restrictions.</p><p>I really enjoyed my time at GitHub Universe and especially am grateful for some of the engineers I was able to meet — awesome engineers who build SnapCI, Heroku, Electron, and more. Listening to people speak about their projects has inspired me to get started in the world of open source!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5c5f1c3be319" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Interesting Blogs]]></title>
            <link>https://medium.com/@mfaywu/interesting-blogs-f0dc7a68e553?source=rss-ec098f820f59------2</link>
            <guid isPermaLink="false">https://medium.com/p/f0dc7a68e553</guid>
            <category><![CDATA[blog]]></category>
            <dc:creator><![CDATA[Fay Wu]]></dc:creator>
            <pubDate>Wed, 31 Aug 2016 17:18:12 GMT</pubDate>
            <atom:updated>2016-09-17T23:29:13.715Z</atom:updated>
            <content:encoded><![CDATA[<p><a href="https://www.troyhunt.com/?m=1">Troy Hunt</a></p><p><a href="https://zachholman.com/">Zach Holman</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f0dc7a68e553" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Conversation Works]]></title>
            <link>https://medium.com/@mfaywu/how-conversation-works-cdb8d1ad226c?source=rss-ec098f820f59------2</link>
            <guid isPermaLink="false">https://medium.com/p/cdb8d1ad226c</guid>
            <category><![CDATA[books]]></category>
            <category><![CDATA[relationships]]></category>
            <category><![CDATA[conversations]]></category>
            <dc:creator><![CDATA[Fay Wu]]></dc:creator>
            <pubDate>Tue, 26 Jul 2016 15:47:38 GMT</pubDate>
            <atom:updated>2016-07-26T15:47:38.088Z</atom:updated>
            <content:encoded><![CDATA[<p><em>This post is unfinished, work in progress!</em></p><p>These are my notes and thoughts from<a href="https://www.goodreads.com/book/show/18107044-how-conversation-works"> How Conversation Works: 6 Lessons for Better Communication</a> by Professor Anne Curzan. It was recommended to me by someone on Audible and I think it’s worth listening to and sharing.</p><h4><strong>The Conversational Floor</strong></h4><p>On the conversational floor, we can observe many interesting things about how we have conversations by looking at the vocabulary we choose, the noises we make, the timing of interactions, and our physical movements. All these seemingly microscopic things work together to make up a whole conversation. We can measure conversations on the value it provides to us, such as the information it uncovers about the people involved or new knowledge. We can also measure conversations across different factors such as the conversational weight pulled by each person and whether topics are pulled in or die out. In order to become a better conversationalist, we need to become aware of the signals that we give out and the signals that we are getting from our conversation partners. Only then can we learn what to do with these signals and how to use them to understand our partners better and vice versa.</p><h4>The Vocabulary We Choose</h4><p>There are certain words that are called “discourse markers”, which send signals to our partners to give them some idea of what’s coming up next. They also negotiate our position in relation to others. These words include “um”, “well”, “so”, “now”, “you know”, “like”, and “I mean”. The word “um” is incredibly interesting because it can be used in so many different ways. One such way is to indicate an unwillingness or to provide a warning that we’re going to say no. An example would be if someone asked us to do a favour and we replied with “um…”. After this, the asker will certainly be expecting a no. Interestingly, if we drag out our declination long enough, the asker will usually relieve us of having to outright say no!</p><h4>The Noises We Make</h4><h4>The Timing of Interactions</h4><p>Apparently we talk over each other all the time, and this is okay, good even.</p><p>Talk Time</p><h4>Our Physical Movements</h4><h4>Sharing the Conversational Floor</h4><p>There is also the concept of sharing the conversational floor. We want to share the floor so that each person pulls their weight in the conversation and can contribute. When one person hogs the floor by talking endlessly, as I’m sure we’ve all experienced, this makes everyone else uneasy and perhaps annoyed. Generally, we all want to have our voice heard and it helps to even out our talk time. When we are ready to give up the floor to someone else, we can indicate to our partners that we want to transition the conversation to them using verbal and physical cues. We can ask them a question, prompting an answer. Since we usually flicker our eyes between our conversation partner(s) and elsewhere, we could look pointedly at them to indicate we are expecting them to talk.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cdb8d1ad226c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[32-file Commits != Agile]]></title>
            <link>https://medium.com/@mfaywu/32-file-commits-agile-1e2c597b6464?source=rss-ec098f820f59------2</link>
            <guid isPermaLink="false">https://medium.com/p/1e2c597b6464</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[agile]]></category>
            <dc:creator><![CDATA[Fay Wu]]></dc:creator>
            <pubDate>Mon, 25 Jul 2016 16:38:01 GMT</pubDate>
            <atom:updated>2016-07-25T16:57:49.756Z</atom:updated>
            <content:encoded><![CDATA[<p>I want to share some of the things I learned about what it means to be agile and deliver incrementally.</p><p>In software engineering class last quarter, we learned about many different processes that different teams and individuals follow to try to reach the most productivity. One of these processes is to be “agile”, which is based on 12 principles in the <a href="http://www.agilemanifesto.org/">Agile Manifesto</a>. I want to highlight a few of them:</p><ol><li>Customer satisfaction by early and continuous delivery of valuable software</li><li>Working software is delivered frequently (weeks rather than months)</li><li>Close, daily cooperation between business people and developers</li></ol><p>I was definitely <em>not</em> doing these things for my first feature at my summer internship. Three weeks ago, I “finished” writing code to add a small feature to one of our software products which manages experiments. It seemed to me like a small change, just adding a field in the UI and calling an API to create, update, and destroy tags associated with each experiment. Looking through the code base, however, I found that it would require changes to many different files. As you know from the title, it ended up being 32 files for a feature that looks like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/210/1*yzvYUFGSYOS9JrORCa-bcA.png" /></figure><p>It took me almost three weeks to check in the code, which slowed down my development process a significant amount. I want to share some of my learnings on how to be more agile.</p><ol><li><strong>Pull from master at least once every few days.</strong> Not doing this gave me endless headaches. Since other developers are working on the same code base, the master branch will be changing often. If you don’t make sure you have the latest changes, it may be that your code will no longer work right with the current master, <em>which is what really matters.</em> I made the mistake of waiting I was done with everything on my branch and it worked beautifully before pulling the latest changes and found that suddenly, merge conflicts were everywhere and the master branch was not as I remember it (duh, it’s been two weeks).</li><li><strong>Check in completed code at least once a week, even if it’s not the complete feature.</strong> This is really important if you want your code reviewed quickly so you can get tips on what to change. Your fellow busy engineers are more likely to procrastinate and wait to review your code if you send them a code review with more than a few files. I had a friend tell me “Well I wanted to take a look at your code but I saw it was 30 files and said….never mind…I’ll leave that for the others.” More eyes from the developers you admire is better, so make it easier for them. I actually really thought that I was only supposed to check in completed features, but <em>this is not true.</em> At least for my team. You can always do a dark deployment, which is when you comment out the feature so no one can use it, but the code will be there and maybe the database is changed, etc. Just make sure the code you check in has its appropriate tests (which <em>pass, obviously</em>) and the build won’t break.</li><li><strong>Talk to the customers early and often to make sure you’re building what they really want. </strong>I went to meet with one of our team’s customers after I was already working on something I thought would be useful to them. I was met with a lot of “I’m not sure that’s useful for us”, which was discouraging but also made a lot of sense. I’m not the customer. The best way to make sure you’re working on something useful is going to ask them.</li><li><strong>If you have a PM (Program Manager), ask them what they think. </strong>This one’s similar to #3 because your PM is the person who’s been working most closely with the customer and (most likely) knows what they want. You’ll be able to bug your PM more often since they’re hopefully down the hall, whereas you probably don’t want to bug your customer every few days to see if you’re still on track. Your PM could also give useful feedback and connect you with the write people if you have any questions or are blocked.</li></ol><p>My goal in the next few weeks is to work on these four things and “become more agile” :p Hopefully this will speed up my development process and help me get more accustomed to the 2-week sprint cycles of our team.</p><blockquote>Side note: My code was finally checked in after almost 3 weeks with 32 files after 13 iterations and 3 reviewers. Not ideal, lots of room for improvement.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1e2c597b6464" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ACM Farewell Speech]]></title>
            <link>https://medium.com/@mfaywu/acm-farewell-speech-920b457d85f1?source=rss-ec098f820f59------2</link>
            <guid isPermaLink="false">https://medium.com/p/920b457d85f1</guid>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[college]]></category>
            <category><![CDATA[women-in-tech]]></category>
            <dc:creator><![CDATA[Fay Wu]]></dc:creator>
            <pubDate>Mon, 25 Jul 2016 16:00:45 GMT</pubDate>
            <atom:updated>2016-07-25T16:00:45.619Z</atom:updated>
            <content:encoded><![CDATA[<p><em>This is the farewell speech I gave at ACM@UCLA’s formal dinner on May 23. I passed on the presidency for our computer science club to Vic Yeh.</em></p><p>In true ACM fashion, I have prepared this speech less than an hour ago so I will be reading from my notes. It is, however, from my heart. ACM means so much to me.</p><p>I remember my sophomore year, I didn’t really have any friends in computer science. I had an awesome internship experience which convinced me that CS was worth studying no matter how hard it was as long as I could go work at Microsoft when school was over. But I was still having a hard time and often thought about switching to north campus. I remember one night in winter quarter when Calvin and Daniel took me to Fuck it Ship it where I met Akshay and he invited me to join his team of then 2 people to revive ACM. He told me that if I wanted a community, I could create it.</p><p>We’ve gone a long way since then. We’ve had big ideas, some of which came to fruition, and some of which are still in the back of our minds. We brought Local Hack Day, Maria Klawe from Harvey Mudd, engineers from all sorts of companies like the standard Google and Facebook but also smaller ones like Riot and various startups. We took over our own clubhouse, and sent 20 fully funded women to Grace Hopper. Heck, we started UCLA’s first ever women in computing club! We went from a small team of less than 10 to having a bunch of committees and hundreds of members! We have, also, changed organisational structure a few times and titles like PM no longer exist (though perhaps we should have an official retreat planner so that we don’t last minute find out someone died in a lake nearby and have to drive to Vegas).</p><p>I think this is the part of ACM that I love the most though. We are flexible. We have ideas, and we just try to do them, and sometimes it doesn’t work but it’s okay because everyone’s having fun and we’ll do it better next time.</p><p>We are like a small startup, and perhaps this year we grew too big too fast, but after all, that is the number one problem that startups face. I’m not gonna lie, being president was really hard. It was very different from running ACM-W or early outreach or speaker series. I didn’t have my own team, and I wasn’t able to be on the ground running at every single event anymore. I split up overseeing the committees with my VPs and that meant sometimes I didn’t know what was going on with some groups. It is easy to suffer from burnout when attempting to lead and be a student at the same time, as many of you know. There have been many times where I got too tired and forgot why we do this. But then I meet someone who tells me they went to a hack event, or AI workshop, or another event, and they loved it and met people they enjoyed talking to — and I am instantly reminded that that is exactly why we put on events and why we run this club. Because there are people, especially underrepresented minorities, or me 3 years ago, who are curious about cs and love it but for whatever reason have trouble getting support or meeting friends or feeling like they belong.</p><p>I am so grateful to the committee heads who took their positions seriously, knowing that we don’t have defined roles for a reason — so that you can take your passion and do whatever you wish. And that you did. This club could not have been possible without you. Thank you, thank you, thank you — Brooke, Kelly, Carter, Vic, Enyu, Sharon, Li-Wei, Jonathan, James, Summer, Marko, and Hannah. I have learned from each of you and am grateful for having had the chance to work with you all to lead our club together. I was impressed with your ideas, admired y’all for your determination, hard work, and passion, and I know how challenging it is to lead a committee. And to all the committee members, especially those who volunteered a lot, y’all are the bomb dot com. I don’t have time to thank everyone right now by name but you know who you are.</p><p>What I want to leave you with is that you can make a difference in this community, in this school. Have wild ideas, bring Bill Gates or Sheryl Sandberg, hack a 3D printer, try to improve the wifi in Boelter Hall. It’s true that our department makes changes slowly, and we receive lots of unnecessary overhead for simple things like getting keys to the clubhouse. Remember to email lots of times and if they don’t respond, just show up to their office :p</p><p>There are students here, those who are currently in our program and those who have yet to come, who are just like you and me. We just want to build things and have fun. We just want to learn from our classes and make friends and have the best undergraduate experience we can. We can make ACM into whatever kind of community we want, and we can leave a place behind where anyone who is curious about technology or passionate about coding can come and meet us. We can be multi-cultural, diverse, and close the gender gap. We can make a place where alumni can come back and still feel like we have a family. You are all my family, and I am infinitely excited for what is to come.</p><p>Vic — Congratulations, I would like to present you with the server room key and clubhouse key. Disclaimer: I don’t know where the server room key is though.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=920b457d85f1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Making Better Software Choices]]></title>
            <link>https://medium.com/@mfaywu/making-better-software-choices-d622190f5d22?source=rss-ec098f820f59------2</link>
            <guid isPermaLink="false">https://medium.com/p/d622190f5d22</guid>
            <category><![CDATA[technical-debt]]></category>
            <category><![CDATA[choices]]></category>
            <category><![CDATA[software-engineering]]></category>
            <dc:creator><![CDATA[Fay Wu]]></dc:creator>
            <pubDate>Mon, 25 Jul 2016 15:54:06 GMT</pubDate>
            <atom:updated>2016-07-25T15:54:38.487Z</atom:updated>
            <content:encoded><![CDATA[<p>This past Saturday I went to <a href="http://datadayseattle.com/">Data Day Seattle</a>, a company-agnostic conference gathering data scientists, engineers, and anyone else who works with large sets of data in Seattle. I only got to listen to the first talk, but it was entertaining and fun. The speaker was Charity Majors, a startup co-founder and author of Database Reliability Engineering. I admired her spunk and how relatable and engaging she was.</p><p>In the first part of the talk, she expressed a sentiment that I’ve been feeling more and more lately — there are so many choices today when it comes to choosing software platforms, tools, frameworks, etc. Fifteen years ago, programming for the web was mostly using the LAMP stack. The only choices we needed to make were MySQL versus PostgreSQL, and Python versus pHp versus Perl. But now, even just looking at how many data storage choices we have reveals too many to count on my fingers. And there are new choices popping up every so often which aren’t ready for production yet, but since they’re looking for users they persuade you to use it anyway. It’s important to learn how to make good decisions on what software actually serves your technology’s purpose, remembering that technology is only a means to an end, not <em>the end</em>.</p><blockquote>“The best code is no code. The second best code is code someone else wrote. The worst code is everything else.”</blockquote><p>Here are some tips from Charity on how to make better software choices.</p><ol><li><a href="http://blog.codeship.com/growing-tech-stack-say-no/">Optimise globally, not locally.</a> This one confuses me because I’ve been of the opinion so far to use the best technology for each purpose and not to consider too many other factors. But I suppose it makes sense when we consider the technical debt that arises if a team uses too many different technologies that aren’t in the team’s expertise. Perhaps using less suitable tech well is better than using the best tech poorly.</li><li><a href="http://mcfunley.com/choose-boring-technology">Choose boring technology</a> because documentation will likely be better and the chance that someone else has had your failures and solved them is higher. This is why I chose Python Django for my website project last quarter in school — I could always search up my error messages in Stack Overflow and get tips on where to start debugging. When deciding whether or not to use new technology, make sure you understand your appetite for risk and consider the operational impact of switching.</li><li>If you’re going to switch or upgrade, commit to replace, not just add. There should be a gating process for major new components to create friction and make it necessary to justify the cost of switching before dedicating engineers to it. When the team has decided to switch though, it causes more problems if the switching process is dragged out. Two years later you could still have the old tech and new tech mixed together, which can cause a wild dependency tree, technical debt, and various other issues.</li><li>Pay someone else to do it if they can do it better for cheaper. This one’s interesting because I’ve always felt that if we can build it ourselves then we should do it because it’s good practice and learning. But this definitely makes sense in the business context because you’ll want to give yourself the room to do more important things that you can’t really pay someone else to do.</li></ol><blockquote>Celebrate the engineers who remove code, deprecate, and reactor, as much as those who add features.</blockquote><p>The patterns that we celebrate are the ones that will get repeated. I’ve heard a few engineers talk about how unmotivated they are to refactor and improve code quality because that generally doesn’t get as much recognition as coming out with a brand new feature. I can definitely understand why, because new features are shiny and noticeable whereas maintaining code bases don’t change things in a visual and obvious way. But code quality is so important in the long term health of software and decreases future maintenance costs. So why don’t we provide more incentive and positive feedback for engineers to work on code quality?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d622190f5d22" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Bloom Filters]]></title>
            <link>https://medium.com/@mfaywu/bloom-filters-63a0ac3a7e34?source=rss-ec098f820f59------2</link>
            <guid isPermaLink="false">https://medium.com/p/63a0ac3a7e34</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[unit-testing]]></category>
            <dc:creator><![CDATA[Fay Wu]]></dc:creator>
            <pubDate>Fri, 22 Jul 2016 19:20:06 GMT</pubDate>
            <atom:updated>2016-07-22T19:20:06.070Z</atom:updated>
            <content:encoded><![CDATA[<p>This past Friday morning, we hosted an Engineering Excellence session and learned about bloom filters by doing a <a href="http://codekata.com/kata/kata05-bloom-filters/">Code Kata</a>.</p><h3><strong>What is a bloom filter?</strong></h3><p>Bloom filters are probabilistic data structures used to test if items are in a set. Testing for an item will return either True or False, with a false positive rate (items deemed in the set may not actually be, but if it’s deemed missing then it definitely isn’t in the set). Bloom filters are used when space efficiency is needed.</p><h3><strong>Our Session</strong></h3><p>For our code kata, we sat around the conference room table and took turns pair programming with each other to implement the bloom filter. We projected the screen from one laptop so everyone could observe and chat amongst ourselves. This format seems to work well with a group of size 2–6 people where each pair gets 10 minutes to do their thing before we switch. The whole session lasted exactly 90 minutes.</p><p>Our team is attempting to move towards TDD (Test Driven Development), where we write the tests first, clarifying the spec and defining wanted behaviours, and then write code to pass those tests. An issue we faced when applying TDD to solve this kata was the difficulty of writing unit tests for a bloom filter. Since it is impossible to get false negatives when using bloom filters, this seemed an obvious unit test. So was getting a positive given that we know the word exists. However, how do we write unit tests for false positives? It’s impossible to write a conclusive unit test for the “includes” method since returning True does not necessarily mean the word exists. We can run the test repeatedly to determine the probability of getting a false positive and determine if that matches our theoretical calculations, but, as Prashanth brought up, this kind of testing isn’t unit testing because running the same test doesn’t always return True.</p><p>Another issue we discussed was the disconnect between writing code to pass tests and writing good code. Ideally, the tests would have good coverage and make it impossible to pass the tests without implementing the actual solution, but this isn’t always the case. When we write tests that don’t cover all the functionality of the solution, or when we are unable to write unit tests for certain features (see false positive issue above), then it’s quite likely that writing code to pass the tests becomes writing drastically simple, unscalable code. It’s even worse when we hard code things into our code to pass certain tests, though the solution to that would be to generate random items preventing us from knowing exactly what will be tested when we run it. It seems that test-writing is difficult and really brings to light any confusion we might have about the specifications or scenarios in which the code would run.</p><p>An interesting question posed was how to black-box test the bloom filter, i.e., how would we test the defining features of a bloom filter? One idea we came up with is to test the memory allocation for the program. Using a dictionary, each additional word added would require more memory allocated to store the key-value pair. Using a bloom filter, adding each additional word should not use any more memory since the bitmap size stays the same. Of course, if the bloom filter performed other operations as well, the program would still be using more memory.</p><h3><strong>Our Code</strong></h3><p>Here’s the implementation we came up with for our Bloom Filter, on <a href="https://github.com/mfaywu/code-kata/tree/master/BloomFilter">GitHub</a>.</p><p>For bloom filters, there are several variables that are possible to adjust — the size of the bitmap and the number of hash functions used. These factors affect the false positive rate; <a href="http://pages.cs.wisc.edu/~cao/papers/summary-cache/node8.html">check out the math here</a>. We came up with a few ways to generate more indexes per word:</p><p>1. Separate hash functions. We used a C# string method, GetHashCode() to generate an integer. But what other hash functions could we use? We’d have to write our own, maybe using something like SHA but since that generates a string we’d have to take just the integer numbers….seemed too complicated.</p><p>2. Take what our GetHashCode() generates and re-hash it using GetHashCode(). This is what we implemented.</p><p>3. Take what our GetHashCode() generates and use it as a random number generator seed. This guarantees that if we fed in the same GetHashCode() result we’d get the same series of random numbers.</p><p>We weren’t sure if 2 or 3 was better and we didn’t get a chance to discuss what could be pros and cons of each.</p><blockquote>* Side note: Randy recommended<a href="https://www.manning.com/books/the-art-of-unit-testing-second-edition"> The Art of Unit Tests</a></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=63a0ac3a7e34" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[OneAnalyst Conference]]></title>
            <link>https://medium.com/@mfaywu/oneanalyst-conference-16c6a934964f?source=rss-ec098f820f59------2</link>
            <guid isPermaLink="false">https://medium.com/p/16c6a934964f</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[big-data]]></category>
            <dc:creator><![CDATA[Fay Wu]]></dc:creator>
            <pubDate>Fri, 22 Jul 2016 19:18:28 GMT</pubDate>
            <atom:updated>2016-07-25T17:07:44.691Z</atom:updated>
            <content:encoded><![CDATA[<p>Analysis &amp; Experimentation puts on a conference to share learnings from both external and internal groups which use data analytics. The number of people attending these quarterly conferences has increased dramatically over the years, a testament to the growing influence of “big data” on decision making. Here I share some learnings from the most recent conference on July 17, 2016.</p><h4><strong>Beyond Books by Brent Smith</strong></h4><p>The information gathered from each purchase made on Amazon.com is understandably low and difficult to use for future recommendations. For example, if someone buys a stapler, you don’t want to recommend them other staplers because they already own one and are unlikely to buy another (perhaps they’ll even be annoyed!). How does Amazon make recommendations better?</p><p>They use several purchase filtering techniques which filter products to display to shoppers based on their past purchases. Here are some tips from Brent:</p><p>1. Never recommend something the shopper already owns, unless it is a consumable. Funny story — for a long time they wouldn’t show shoppers any toothpaste after a single purchase, even though toothpaste needs to be repurchased every few months.</p><p>2. Consider age progression for consumables. For example, diapers are consumables and parents will need to purchase them again fairly quickly. However, babies grow in size so the next recommendation should be a size bigger.</p><p>3. Items have varying lifespans. This is related to the preceding tips. For example, books have a longer lifespan than seasonal clothes.</p><p>Amazon must closely examine shopper behaviour and what the purchase information can give them to determine how to best advertise and recommend products to their shoppers.</p><blockquote><em>* Side note: Here’s an interesting fact — Amazon started out as a bookstore!</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/540/0*0elLSyuHLlnDNyKa.png" /></figure><h4><strong>Lessons Learned Providing Dashboards, Cubes, and Analyses for SharePoint &amp; OneDrive by Ethan Lexington</strong></h4><p>Ethan’s team delivers insights to the SharePoint and OneDrive team by validating telemetry, aggregating the data sets in COSMOS &amp; DW, and getting Cubes and reports to create responsive analyses, analysis projects, and self-service tools.</p><p>The motivation behind using dashboards and Cubes is to create agreed-upon calculations and reports for business KPIs. Self-service Cubes can actually be more expensive to support than providing service from the team since a lot of users don’t have enough statistics knowledge and healthy skepticism to effectively use the Cubes. There are a lot of assumptions built into the Cubes that require additional help to understand and make meaningful conclusions.</p><p>One of his learnings is that it’s important to correct people if they build things with incorrect logic, but the team needs to let slide things that people build which are logically correct but hideous looking. To help make sure people build the correct things, they’ve created a group to let anyone in the organisation send questions in. That way, anyone can answer the questions and this usually means quick responses.</p><h4><strong>Experimentation at SkyScanner by Colin McFarland</strong></h4><blockquote><a href="http://codevoyagers.com/2016/03/16/design-like-youre-right-test-like-youre-wrong/#more-629">“Design like you’re right, test like you’re wrong.”</a></blockquote><p>Some interesting links:</p><p><a href="http://codevoyagers.com/2015/11/26/common-pitfalls-in-experimentation/">Common Pitfalls in Experimentation (Code Voyagers)</a></p><p><a href="http://www.experimentationhub.com/hypothesis-kit.html">Hypothesis Kit &amp; Power Calculation</a>s</p><blockquote><em>* Side note: The new SkyScanner page looks a lot more like bing.com than the old version. They also have a creepily cool feature which detects if you move your mouse to close the tab after scrolling and they ask why you’re leaving..</em></blockquote><h4><strong>Onboarding to Experimentation by Benjamin Arai</strong></h4><p>Many people think they don’t need to test things they already “know” the result for, but it’s important to run controlled experiments because only users can really tell you whether or not they like a feature and sometimes there are external events that cause some effects that are not related to your deliberate change.</p><p>Here are the “phases” of onboarding:<br>1. <em>Crawl phase</em>: For the initial onboard, the team must define the OEC (Overall Evaluation Criteria) and identify key factors to get the most value from experimentation. They need to identify issues with running experiments since rarely is the data being collected in a complete and clean manner.</p><p>2. <em>Walk phase</em>: Invest early and learn — it’s important to invest engineering time into data quality issues and instrumentation. Around 5–10 experiments in, the team should dive deep into several experiments to develop guard rail metrics and try to really understand how the metrics are moving. A bad practice is to try to experiment on just Preview audience — this customer segment isn’t ultimately who you really care about and it’s not a large enough sample size!</p><p>3. <em>Run phase</em>: This is when feature teams start to scale and experiments are run much more often to help everyone make better decisions!</p><h4><strong>Data Science From A to B by Michael Hochster</strong></h4><p>There isn’t an agreed upon definition of data scientist because there’s a mismatch at companies between what PMs do (business goals) and what data scientists are interested in (machine learning, data, etc). Michael’s definition:</p><blockquote><em>“Data Scientists are people with some mix of coding and statistical skills who work on making data useful in various ways.”</em></blockquote><p>There are three models of data science teams:</p><p>1. Standalone: A separate data science team. Feature teams come to this team to ask for support.</p><p>2. Integrated: Each feature team has their own data scientists.</p><p>3. Partnership (like at Pandora): Have a centralised team but available for everyone.</p><p>Michael believes several things are over-hyped:</p><p>1. Storytelling — Your main job is non-fiction, not to entertain people but to help them make decisions.</p><p>2. “Actionable Insights” — Don’t go digging for actionable insights but have some questions and explore the data, like discovery rather than expecting some conclusion.</p><p>3. Beautiful visualisations — Useless but pretty…but still useless. Your main job is to explain things.</p><p>4. Don’t define yourself in terms of the tools you use because they will be different 5 years from now. Data exploration is still relatively untapped and efficient, large-scale experimentation will have a bigger role in the future. (This is his prediction)</p><h4><strong>A/B Testing in Desktop Apps</strong> <strong>by Viktor Veis</strong></h4><p>It is definitely different to A/B test apps than online features because the user retains a lot of control for preferences in programs, but it’s still possible to do user experience tweaks and performance optimisations. Here’s an example — the smiley face at the top of Visual Studio 2015 which people could click on to give feedback got 122 responses about the smiley itself (that it was creepy, mostly). They decided to run an experiment showing a different, more neutral looking faceless icon. For this icon, they got responses back asking if they could put back the smiley face because they missed it, but ultimately the team decided to keep the new neutral icon because overall, they got less negative feedback from the neutral icon than the smiley face. (Sadly, I liked the smiley face)</p><blockquote>*Side note: One of the downfalls of large companies is that lots of teams can build similar things, somewhat unaware of each other. Someone asked what the novelty was of a data product that our team has made, and we didn’t have a good answer because we were unaware that there was another product like it!</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=16c6a934964f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Hello Seattle (and A&E)]]></title>
            <link>https://medium.com/@mfaywu/hello-seattle-and-a-e-e95a30c8d953?source=rss-ec098f820f59------2</link>
            <guid isPermaLink="false">https://medium.com/p/e95a30c8d953</guid>
            <dc:creator><![CDATA[Fay Wu]]></dc:creator>
            <pubDate>Fri, 22 Jul 2016 06:32:10 GMT</pubDate>
            <atom:updated>2016-07-22T06:32:10.097Z</atom:updated>
            <content:encoded><![CDATA[<p>For my summer internship this year, I’m making some goals for myself to maximise how much I learn and grow. One of these is to write a blog where I will share about the books I’m reading, the project I’m working on, and interesting things about my team or other topics in tech. Though I will optimise for quantity over quality, I believe the quality will come as long as I get myself started. So without further ado, here is some background information my team!</p><p>The Analysis &amp; Experimentation team (Microsoft &gt; ASG &gt; IPG &gt; A&amp;E) currently consists of 77 company employees and 5 interns, 27 of which are data scientists, 7 PMs (Program Managers), and 44 software engineers (1). Our team’s mission statement is “Accelerating software innovation through trustworthy experimentation.” The experimentation method that our team provides is A/B testing which is a scientific method used to prove the causality of a change on the observed impact (2). We have a randomly selected control group and treatment group and we apply a change to the treatment group to watch the effect of the change on certain metrics that we define. Since we use a small sample of users to test on, we can limit the risk and determine if the feature creates desired or undesired effects on user behaviour before rolling out the feature to all users.</p><p>Our customers include Edge, Bing, XBOX, Answers, MSN, Skype, AdCenter, Office and more. Bing is our biggest customer, running thousands of experiments every day. If you visit <a href="http://t.umblr.com/redirect?z=http%3A%2F%2Fbing.com%2F&amp;t=NzJkYzM3MGE5YmMxYjAxNTZmYzVhY2I4YTRjNzI3MDA1MzIxOTE4OSxQbElleXVTMQ%3D%3D">bing.com </a>right now, you will be seeing one of the thousands of combinations of experiments running on the search engine! Our platform works as a “standalone model”(3) which means that we are a shared platform that other teams come and use. There are advantages and disadvantages to a standalone model, as opposed to the partnership or integrated models since the team can serve many products using a single platform and concentrate on creating advancements to improve our one platform, but our team is at least one step removed from any product. As such, our team is funded by our customers who pay for additional engineers and/or analysts to work on their feature requests or guide their experiments. Every sprint, the team looks at the feature requests or bug tickets filed by customers and prioritises which ones to work on. When a new customer wants to come onboard, the team will look at how suitable the customer is for our platform.</p><p>The “ideal” customer is a feature team that works on a product with enough users (at least in the thousands) and has a culture that is ready to integrate experimentation into their decision making. Without enough users of the product, we wouldn’t be able to collect enough data to deem the experiment conclusion statistically significant (4). Interestingly, there are teams which do not believe in experimentation and apparently trust their own judgment over the information gathered from real product users. It seems that the leadership of these feature teams really decides whether or not the team will head towards the direction of experimentation and it clearly helps when a previous member of our data science team joins them (this is what happened with Visual Studio).</p><p>Backtracking a little, here is a bit of history — Ronny Kohavi, our GM (and a distinguished engineer!) started the experimentation platform as a project in 2006 after having moved to Microsoft from Amason where he also led an A/B testing group. Since Bing was the first and heaviest user, the team was moved into the Bing team in 2010. However, since many other products have also started to use experimentation since then, the team official became Analysis &amp; Experimentation. One of our competitors is Optimizely, a mid-size company (~400 employees) which provides enterprise-level A/B testing and personalization for the web and mobile apps (5). Though they are our competitors, we actually encourage people to start off with Optimizely so that they can get started quickly and get a feel for A/B testing before we create a commitment to onboard them onto our system. Optimizely provides a beautiful interface and an easily integrated platform that requires just a few lines of JavaScript but has fewer metrics, decreased performance (pages may flicker), and other downsides. There is no such thing as a free lunch (6)!</p><p>Our team culture is different from some of the other MSFT teams I’ve observed — we have 2-week sprints with a retrospective that happens at each cycle’s end so we can review the good and the bad. Everyone uses user stories (7) to plan features and there’s a push to do TDD (Test-Driven Development) and other agile methods. Our Foray App 6-people team has a daily stand-up scrum at 10:30 to give updates and hash out issues. I like that the team is always active and making new features, not just squashing bugs. The backlog is consistently filled with all these feature requests that Ronny and the other leadership team members go through every few weeks and re-prioritise.</p><p>Overall it’s a pretty cool place to be and I’m excited to learn and build. I think it’s awesome that the team is leading the effort to make experimentation something that all feature teams do across Microsoft. It’s really putting the users first and “listening” to customer feedback in an even better way than simply asking them how they use the products! If any of this sounds interesting to you, check out our website (which has a plethora of cool papers) at <a href="http://t.umblr.com/redirect?z=http%3A%2F%2Fexp-platform.com%2F&amp;t=OWQzNDIxODViOWI5NjYyMmNjMTE3Yjk2OGQyNjFmZjdlZGMyY2JhMCxQbElleXVTMQ%3D%3D">www.exp-platform.com​​</a></p><ol><li>//who, our internal organisation map of people</li><li>Our team’s OneNote notebook</li><li>Michael Hochster’s talk at the One Analyst Conference</li><li>Statistical significance: when the p-value is less than 0.05, meaning the chance of the results occurring randomly rather than as an effect of our changes is less than 5% (or some other value we set, but which is generally 5)</li><li><a href="https://www.optimizely.com/">https://www.optimizely.com/</a></li><li>Says Professor Peter Reiher at UCLA during a Computer Security lecture</li><li><a href="https://www.mountaingoatsoftware.com/agile/user-stories">https://www.mountaingoatsoftware.com/agile/user-stories</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e95a30c8d953" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>