<?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 Riley Feltner on Medium]]></title>
        <description><![CDATA[Stories by Riley Feltner on Medium]]></description>
        <link>https://medium.com/@rileyfeltner?source=rss-9e64efae5337------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*CZG0LQsabIv_tc61uYI5SA.jpeg</url>
            <title>Stories by Riley Feltner on Medium</title>
            <link>https://medium.com/@rileyfeltner?source=rss-9e64efae5337------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 02:02:26 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@rileyfeltner/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[ISA 630 Project Part I: Feed Forward Neural Network]]></title>
            <link>https://medium.com/@rileyfeltner/isa-630-project-part-i-feed-forward-neural-network-ee4eeace74dc?source=rss-9e64efae5337------2</link>
            <guid isPermaLink="false">https://medium.com/p/ee4eeace74dc</guid>
            <dc:creator><![CDATA[Riley Feltner]]></dc:creator>
            <pubDate>Sat, 13 Apr 2024 03:03:14 GMT</pubDate>
            <atom:updated>2024-04-13T03:15:36.028Z</atom:updated>
            <content:encoded><![CDATA[<h4>RJ Menter, Steven Wintringham, Riley Feltner</h4><h4>Introduction</h4><p>Image recognition has been a hot topic in machine learning, especially now with computer vision making strides towards understanding visual data on a level that rivals human perception. A simple yet useful first step in this space is utilizing feedforward neural networks as the architecture is fairly straightforward for processing input data and learning representations. A feedforward neural network is one that has a uni-directional flow meaning that information only flows forwards. This can pave the way for more complex neural networks to build upon.</p><p>Neural networks have three distinct types of layers: input layers, hidden layers, and output layers. We do not change the input layers and output layers. The hidden layers are hyperparameters which do need to be specified.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*uaPRSTGg3OYsR46_.png" /><figcaption>Blue: Input Layer, Black: Hidden Layers, Green: Output Layer. Image: <a href="https://victorzhou.com/series/neural-networks-from-scratch/">https://victorzhou.com/series/neural-networks-from-scratch/</a></figcaption></figure><p>Neural networks are a machine learning algorithm inspired by biological neurons. The function of each neuron is to increase the signal, decrease the signal, or stop the signal. The input layer of the neurons are just individual variables. In the following neurons, the hidden layer neurons, each one has a weight and a bias which makes a change to those inputs. The Σf below represents the summation of the weights derived from the model’s activation function. The activation functions most typically used are sigmoid, tanh, and relu. The type of problem will determine which activation is optimal as each activation function has its strengths in certain situations. A bias is also tacked on to reduce or amplify the signal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/634/1*QyYtxmsxdpi96VeQ2J6O_Q.png" /><figcaption>Neuron with weighted inputs and bias, applying summation and activation function for output.</figcaption></figure><h4>Implementation</h4><p>We created a feedforward neural network using the fashion-MNIST dataset. The shapes of the train and test sets are (60000, 28, 28) and (10000, 28, 28) respectively. So, there are 60,000 images in the train dataset each with 28x28 pixels and 10,000 images in the testing dataset each with 28x28 pixels also. Below is a visualization of the first observation of our training set.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/667/1*eQOLcSFHyijKvMZnKSqWhQ.png" /><figcaption>1st Sample of Training Set Re-shaped to a 28x28 Image</figcaption></figure><p>Now for the fun part. Pixel intensities usually go from 0 to 255, so it is generally accepted to scale images by dividing by 255 which is what we did. This scaled our training and testing sets to range from 0 to 1. Next, we flattened the image to get a single vector as our input column. Below is the code for our neural network.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lDy3uaSerBz_7t2ys1YmQQ.png" /></figure><p>The shape of the input layer is representative of the available pixels: 28x28. Relu activation was used because it promotes sparsity in the network due to it return zero for negative values. A 10% dropout was added after each hidden layer to reduce overfitting. Finally, we used softmax activation in the output layer as it best represents the probabilities in a multiclass classification model.</p><p>We complied the network using an adam optimizer with a small learning rate of 0.0005 and a sparse categorical cross entropy loss function. We chose an adam optimizer as it is an adaptive gradient algorithm meaning that it maintains a per-parameter learning rate that improves performance on sparse gradients. When predicting using the unseen data (the test set) the model presented the following results.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1002/1*e0mzH77HpMq2oRxuOyEvsg.png" /></figure><p>While very impressive, what we found to be most impressive was the increased clarity of the following visualization using the predictions from our model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/819/1*1xDRJeZiD5hfHbUkaT0Ncg.png" /></figure><p>The network detected anomolies in the data which allowed us to “de-noise” it. Feedforward neural networks are extremely powerful in their ability to learn complex patterns and relationships which allows for effective classification and prediction. Feedforward neural networks are useful due to their capacity to approximate complex functions, learn from raw data without feature engineering, and adapt to various tasks through customizable architectures. Their flexibility, scalability, and ability to extract hierarchical features make them a true weapon in image recognition.</p><p>Our code can be found here: <a href="https://github.com/rileyfeltner/ISA630-Feedforward-Neural-Net">https://github.com/rileyfeltner/ISA630-Feedforward-Neural-Net</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ee4eeace74dc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Mechanisms to Decrease Variability in Amateur Pitchers]]></title>
            <link>https://medium.com/@rileyfeltner/mechanisms-to-decrease-variability-in-amateur-pitchers-1bc852f2206a?source=rss-9e64efae5337------2</link>
            <guid isPermaLink="false">https://medium.com/p/1bc852f2206a</guid>
            <dc:creator><![CDATA[Riley Feltner]]></dc:creator>
            <pubDate>Sun, 11 Feb 2024 04:09:44 GMT</pubDate>
            <atom:updated>2024-02-11T04:26:37.558Z</atom:updated>
            <content:encoded><![CDATA[<p>It should come as no surprise that amateur pitchers present less consistency than professional pitchers. This lack of consistency is often attributed to a pitcher being young or lacking experience, which can be true. However, there are ways in which coaches can aid players in becoming more consistent versions of themselves. The problem is very easy to identify. For example, player X needs to throw more strikes, land the breaking ball more, or improve fastball command. A solution based approach to helping a player gain consistency is more difficult than simply identifying the issue, but it is not overly challenging. And in many cases, it is certainly worth the effort instead of letting the player mature or figure things out on their own. The areas I have found to be successful in helping pitchers gain consistency fall under three categories: identify an ideal movement profile for their pitches, develop an understanding of how to create that movement profile, and building a more robust thrower.</p><h4><strong>The Variance is Big… Very Big</strong></h4><p>To examine the difference in professionals versus amateur, I took a quick look at Major League Baseball’s 75th percentile — 25 percentile compared to Division I Baseball’s 75th percentile — 25 percentile of the following metrics: velocity, horizontal break, vertical break, extension, vertical release, and horizontal release. Only pitchers who threw 50 or more or a particular pitch type were included. I then took the median values of all of the characteristics of all pitchers to produce the following results.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0rDektqCdT6chPhmcIV-uQ.png" /></figure><p>Red values show a high discrepancy between the 75th percentile and 25th percentile while green shows a low discrepancy. This should do a good job of depicting how much more consistency is present in professionals compared to amateurs. They repeat their velocity, pitch shapes, movement profiles, and release characteristics more consistently than college pitchers.</p><h4>Identifying and Replicating Ideal Movement Profiles</h4><p>Many amateur pitcher’s pitch shapes vary greatly from outing to outing. This could be a byproduct of amateur pitchers tinkering with pitches more than professionals. However, it can sometimes be attributed to amateur pitchers not just lacking the knowledge of how to create their intend movement profile, but lacking knowledge of what the intended movement profile should be. The visualization below is what sparked my idea on this topic in the summer of 2023. This pitcher made no intentional changes on his splitter throughout the spring season. There was no change in the wrist orientation, grip, or mindset with the pitch.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cqeh2p4LzFFscZMzZJ6ArQ.png" /><figcaption>2023 Spring Splitters</figcaption></figure><p>This pitcher’s splitter ranged from -3.7 IVB to 18.5 IVB. So, how can a pitcher possibly consistently execute pitches or have sustained success when their pitch shapes vary so much without intentional adjustments? The simple answer is it is a whole lot more difficult with such great variance.</p><p>Using high speed video paired with Trackman data to create a new pitch from scratch is very attractive to many. What is less talked about is utilizing these resources to identify the characteristics of good versus bad pitches. This will allow us to identify the qualities that make the good pitches good and the bad pitches bad. Back to the above example of the splitter, the movement profile we tried to target were the ones clustered below 10 IVB. After one bullpen session, we walked away with a solution. The splitters that had carry to them where being lead with the palm whereas the splitters with depth were not being lead with the palm.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*utbZnYkouXy4_M55LwTqtg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/698/1*4z4QDqrOqYmmdF-QqR3JBQ.png" /><figcaption>Left: Bad (78.8 mph, 12.4 IVB, 4.0 HB) | Right: Good (80.0 mph, 3.8 IVB, 11.9 HB)</figcaption></figure><p>The high speed video and still shots were presented to the player. The cues given were: keep the wrist stiff and get the finger tips to the catcher. The below results are pretty impressive! A much more consistent and ideal movement profile; hopefully leading to a more consistent and successful pitch for the 2024 season.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_GymJSTNeE1gCQT0.gif" /><figcaption>2023 Spring vs 2023 Fall and 2024 Winter Splitters</figcaption></figure><h4><strong>Identifying and Replicating Ideal Movement Profiles: Putting Together the Puzzle Pieces</strong></h4><p>Oftentimes, it is not as straightforward as the above example. There can be gaps in sessions where high speed video was obtained or inability to get high speed video (i.e. in-game pitches). The pitcher below showed the ability to generate a good amount of sweep on his slider at the same velocity as ones that were of a shorter movement profile. However, none of these more sweepy sliders were thrown in his early bullpens. And again, the movement profile had shown to be inconsistent in 2023 and in early fall 2023 bullpens. There was no need for a grip change as we knew he could generate sweep out of the grip he had been throwing. So, we just changed the wrist orientation of the pitch from being more behind the ball to being more “around” the ball. The cue given to the player was to try to line up the thumb with the pointer and middle finger horizontally at ball release.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kOL1uDUcS6kzlw0lBrOlNQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Fp2oHx9zDaSjiNd0oL_muQ.png" /><figcaption>Old vs New Slider Wrist Orientation</figcaption></figure><p>The below results are another impressive change. This time obtaining a more optimal movement profile and one that is more consistent.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*fhprNt55UO2q9p3Q.gif" /><figcaption>2023 Spring Slider vs 2023 Fall and 2024 Winter Slider</figcaption></figure><p>Helping players gain an understanding of how their pitches should be moving and exactly how to create that movement will likely result in a pitch that has a more consistent movement profile. A more consistent profile will likely allow for better command and feel for the pitch.</p><h4><strong>Train for Athleticism, Reap the Consistency Benefits</strong></h4><p>I have recently taken the approach of athlete first, pitcher second in training amateurs especially. It may seem counter intuitive, but I believe that adding in variability to a pitcher’s throwing routine can help them to increase their consistency on the mound. We refer to these as “athletic throwing drills”. We will often use these with pitchers who present one of more of the following: limited strike throwing, very tall / larger frames, inconsistent release characteristics on the same pitch (<a href="https://medium.com/@rileyfeltner/adjustability-repeatability-271cb17dd45f">NOT on different pitches </a>).</p><p>The old-school thought process for these pitchers that I have described would be lot of dry reps and structured catch play to “repeat their delivery”. My approach is that adding drills that require high levels of athleticism and have a proprioception component will form a more robust thrower. Someone who will become more comfortable with their movement patterns, thus becoming more consistent. Below are some of my favorite athletic throwing drills that I will give to pitchers who present one or more of the three components listed above.</p><p><strong>Athletic Throwing Drills</strong>: <a href="https://youtu.be/vX2xEXrwzOs">https://youtu.be/vX2xEXrwzOs</a></p><p>The first three drills (walking forward throws, walking backward throws, and hot feet clap to throw) accomplish a multitude of things. The walking forward throws and walking backward throws have a proprioception component and a “self-organization” component. The hot feet clap to throw is more focused on letting the athlete self organize after an abrupt external cue (the clap). This will force them to organize and deliver the throw in a randomized fashion. It’s similar to a quarterback having to deliver the ball after only a split-second to think. The final drill is a double shuffle to throw. This is simply getting the pitcher’s feet moving and letting them feel their bodies create the throw.</p><h4><strong>Limitations and Conclusion</strong></h4><p>The tracking technologies and tagging method for pitch types are not consistent between the MLB and Division I baseball. Therefore, the 2023 75th Percentile — 25th Percentile Pitch Characteristics Division 1 vs MLB visualization should be taken with a grain of salt, especially the Horz Break and Vert Break values. Player names were not included in this article as I did not want to give away any information on our pitchers.</p><p>There are certainly benefits to letting an athlete find their own way and develop consistency through trial and error. However, there are also benefits in helping an athlete to expedite the process through: identifying an ideal movement profile for their pitches, developing an understanding of how to create that movement profile, and building a more robust thrower. Creating new pitches and developing elite velocity are flashy topics that get a great deal of attention as they should. There has not been many new approaches to aiding pitchers in their consistency. Hopefully this quick blog can provide some helpful insight on mechanisms to decrease a pitcher’s variability, thus increasing consistency and success.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1bc852f2206a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My Stuff + Model: The Process, Results, and Reproducible Code]]></title>
            <link>https://medium.com/@rileyfeltner/my-stuff-model-the-process-results-and-reproducible-code-f77ef8456657?source=rss-9e64efae5337------2</link>
            <guid isPermaLink="false">https://medium.com/p/f77ef8456657</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[r]]></category>
            <category><![CDATA[baseball]]></category>
            <dc:creator><![CDATA[Riley Feltner]]></dc:creator>
            <pubDate>Fri, 16 Jun 2023 19:15:54 GMT</pubDate>
            <atom:updated>2023-06-16T19:34:23.422Z</atom:updated>
            <content:encoded><![CDATA[<p>I’d like to start by giving some credit where it is due. There were several individuals who let me bounce questions off of them and offered me some advice along the way: <a href="https://twitter.com/Johnny_Davis12">Johnny Davis </a>, <a href="https://twitter.com/JoeKoetters1">Joe Koetters</a>, and <a href="https://twitter.com/RobertFrey40">Robert Frey</a>. All of whom post great content on their respective blogs and twitters. I originally used Johnny’s code as a roadmap for my model. For that reason, there are similarities between our code (the calculation of RV/100 and use of differential break in particular).</p><p>I had wanted to make a Stuff + model for a long while now. While not overly challenging, there were some nuances that had stumped me and caused me to lose traction then pick it back up months later. This blog should help to showcase my model and process. More importantly, it will hopefully be useful in providing reproducible code and information to those interested! This should allow anyone to create their own Stuff + model in R.</p><h4><strong>Variable Selection</strong></h4><p>The selection of the dependent variable in pitch quality models carries substantial weight and impact. Initially, I wanted to use some form of Whiff% or CSW%. However, there are some issues with making a Stuff + model around these metrics in my opinion. The main issue being that pitchers who rely on weak contact rather than swing and miss could be undervalued. While whiffs are a big indicator of nasty stuff, I don’t think it is the only indicator. I believe there is a place for statistics like xWhiff or xCSW, but that place is not a Stuff + model in my opinion. So, I decided to use run values for my model. Briefly summarized, run values are either positive or negative values that are be applied to certain events. A positive run value would indicate that the offense is benefiting from an event. A negative run value would indicate that the defense is benefiting from an event.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/723/1*tt1j_alwIXjSg9mPukfN4w.jpeg" /><figcaption>About half of the run values used</figcaption></figure><p>The independent variables I used are fairly typical for stuff models. I used: velocity, vertical break, horizontal break, vertical release, horizontal release, extension, and differential break. I took the absolute value of the horizontal metrics to account for Righty vs Lefty. Differential break was an idea I gathered from Johnny’s model. Many people will use metrics that measure the difference between an off-speed pitches movement and velocity as compared to their main fastball. While likely a good predictor, I did not like this approach for my model. I wanted to look solely at how nasty a pitch was in a vacuum. Differential break (vertical — horizontal break) helped to strengthen my model while only looking at the pitch being analyzed. I decided not to use plate location for this model either as I just wanted it to predict how nasty a pitch was, not execution or command. Once run value is predicted, it is important to account for pitch count. To do some, I did: 100 * sum(xRV)/n. This will scale the model the same way as Baseball Savant scaled their RV/100 metric.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1002/1*3x5AOyb0EhTBpauRnrLFEA.png" /></figure><h4><strong>Model Selection, Information, and Accuracy</strong></h4><p>My model uses 2019–2022 data to train, then makes predictions on the 2023 data. I initially trained on 2019–2022 data and a slice of 2023 data. The results were fantastic, but to me this is a no-no in machine learning if it can be avoided to limit overfitting. Originally, I was using a randomForest model. I soon switched to an xgboost model because of the quicker run time and slightly better performance. I used a simple grid search approach to tune the model. The hyperparameters that were of interest to me were ntree, depth, and alpha. The best values for ntree, depth, and alpha are extracted from the grid search and applied to the model. This approach greatly strengthened my model and saved me a lot of time in tinkering with the features of the model.</p><p>I was fairly pleased with the accuracy of my final model. There was a correlation of 0.33 which resulted in an R² of 0.11. The RMSE is a 3.25 which is a bit high but not too bad. Considering that command, execution, sequencing, and batter were not included, the results seem solid to me.</p><h4><strong>Leaderboards</strong></h4><p>Now for the interesting part! Here are the leaderboards per pitch type:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*YKNVXCDwqm-4TSBxc61fuQ.png" /><figcaption>2023 Fastballs Top 10 Stuff+ (≥ 100 Pitches)… Wow does my model love 4S FBs!</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*932rORdaqaVOUeZX8Ls7LA.png" /><figcaption>2023 Sliders Top 10 Stuff+ (≥ 100 Pitches)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*vkxpxc_rh2H5tGRnw_PGdA.png" /><figcaption>2023 Curveballs Top 10 Stuff+ (≥ 100 Pitches)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*yEDUdO8EUmhCiQWOevQpcg.png" /><figcaption>2023 Changeups Top 10 Stuff+ (≥ 100 Pitches)</figcaption></figure><h4>Final Thoughts and Ideas Moving Forward</h4><p>I think this is a solid start to my initial Stuff + model! This will likely go through more iterations as I learn more advanced machine learning techniques or as new ideas come to mind. One aspect of the model I was still unsure on is how people typically scale their Stuff + models. To me, it makes the most sense to scale 100 as average for all pitches. However, I can see how it is also beneficial to scale 100 as average for individual pitches so pitchers know if that particular offering is above or below average based on the same pitch type. If you have a Stuff + model, let me know what you did! Future iterations could consist of prioritization of tuning for the RMSE instead of correlation and assess model performance between the two methods. I may consider adding plate location in to build a separate (but similar) model to this Stuff + one.</p><p>My hope and main purpose of this blog is that the code is totally reproducible, which I imagine it is. Scraping the data off savant first is necessary and left out as I scraped in separate files. Here is a great <a href="https://www.rdocumentation.org/packages/baseballr/versions/0.4/topics/scrape_statcast_savant">resource</a> on how to scrape the data into R.</p><p>My code for this project can be found here: <a href="https://github.com/rileyfeltner/Final-Stuff-Plus">https://github.com/rileyfeltner/Final-Stuff-Plus</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f77ef8456657" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data-driven In Season Reports Using Gerrit Cole’s April ’22 Data]]></title>
            <link>https://medium.com/@rileyfeltner/data-driven-in-season-reports-using-gerrit-coles-april-22-data-7969f9adbc05?source=rss-9e64efae5337------2</link>
            <guid isPermaLink="false">https://medium.com/p/7969f9adbc05</guid>
            <category><![CDATA[pitching]]></category>
            <category><![CDATA[r]]></category>
            <category><![CDATA[baseball]]></category>
            <category><![CDATA[scouting]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Riley Feltner]]></dc:creator>
            <pubDate>Fri, 07 Apr 2023 20:39:06 GMT</pubDate>
            <atom:updated>2023-04-08T01:27:24.608Z</atom:updated>
            <content:encoded><![CDATA[<p>Code available upon request.</p><p>Dating back to his first start of last season, I have made reports after every outing for my brother, Ryan Feltner. When we were younger, we would mostly just spitball ideas to see what would work for one another. As he’s grown, so has my process in how I present information and ideas to him. For 2023, I will make a report after each start and a report for each month. They are very similar to one another but they do have some important differences. In the post-start reports, I will not include any xBA maps or miss heat maps as their is not a large enough sample size. Below you will find the statistical information that is provided for a monthly report. I also watch video after each outing and touch on anything I see in in the report. This article will just be focused on the various statistics and information that is provided to him. I only handle post-game reports as people with greater coding ability and time handle any sort of pre-game reports. For this blog,<strong> </strong>I will <strong>NOT</strong> be using any of Ryan’s information as I do not want to share any information on him publicly. I will be using Gerrit Cole’s April 2022 data to depict what a monthly report for Ryan will look like!</p><p><strong>Pitch Usage</strong></p><p>Sometimes something simple can make a big difference. It’s important to keep an eye on pitch usage by count type to ensure there are no major trends that a pitcher is falling into. Especially for pitches that are not extreme outliers metrically.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*ySSNBNIcMQu9UuM5M9LFrQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*J5tce19HvmPTWXcTLV2mVA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*bR_U7nzLWs0qpe3IRNlddA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*NWGzCC7Phc1qpw4NISOHbw.png" /><figcaption>Gerrit Cole April 2022</figcaption></figure><p><strong>Pitch Break Chart</strong></p><p>Here is another simple graphic that will show the movement profile by pitch type.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*CyePklE0Ar0KDLaYgXpqXw.png" /></figure><p><strong>Pitch-Type Breakdown Using Various Visualizations and Statistics</strong></p><p>For the sake of brevity, I will just show an analysis of Gerrit Cole’s fastball. In a typical report I would follow the same process for all pitch types. The first piece of information that is provided is a breakdown of the mean values of important metrics. These will likely change or be added to throughout the season, but this is what we will be starting with.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/786/1*zd8HXNnmurGLCpBY3zCRsg.png" /><figcaption>Mean Values of Important Statistics</figcaption></figure><p>Next, I show frequency heat maps separated by batter handedness.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*6Tg-gP2nNJhOdgQOdW_bGQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*UoKD9s401I-cBlzvjck9kQ.png" /></figure><p>Now, I will put in point plots for swing and misses. These are much more applicable for the post-start report vs the monthly report. But, I am keeping them in for now.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*7-iJijH3xPml3-RQT2qssQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*2EjUIQpopKAzae1BWYnbSQ.png" /></figure><p>Now for the swing and miss heat maps!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*gvLvan3BqGWO6zvNEWsL_A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*6M-kUNYNm_AoFGg1j8LVzQ.png" /></figure><p>The last visualization that will be provided is an Expected Batting Average hexagon chart that is heat scaled. My xBA model is a fairly simple one. I used a random forest for my model to predict the probability of a batted ball becoming a hit or not. Then, I added strikeouts back into the dataset. Strikeouts are given a probability of zero to be a hit as they are not balls in play. The numbers inside the hexagon represent the frequency (number of pitches that the hexagon is accounting for). This is an important addition because weighing the sample size is always important in decision making.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*rTWItQTgBjISHcNBQAg6oA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*GSwW_1RpHlduILoCVy5z5g.png" /></figure><p><strong>Final Thoughts</strong></p><p>These reports are likely to be expanded upon throughout the season as I have more free time in the summer months. But, I am happy with it for now! It will help me tremendously in my suggestions to Ryan and I believe it will help him as well. Post-game and monthly reports are a great way to keep tabs on various statistics to target possible areas of improvement. As mentioned above, there is also a portion of video analysis that goes into the reports as well that is not highlighted in this blog. It is kept brief and simplified greatly due to it being in-season. I also include a thoughts and takeaways section to condense any findings and state my opinions in the reports.</p><p>Thanks for reading this blog. Check me out on Twitter for more content!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7969f9adbc05" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Presentation of Propositions to Players]]></title>
            <link>https://medium.com/@rileyfeltner/the-presentation-of-propositions-to-players-694b5e8da198?source=rss-9e64efae5337------2</link>
            <guid isPermaLink="false">https://medium.com/p/694b5e8da198</guid>
            <category><![CDATA[pitching]]></category>
            <category><![CDATA[coaching]]></category>
            <category><![CDATA[baseball]]></category>
            <dc:creator><![CDATA[Riley Feltner]]></dc:creator>
            <pubDate>Mon, 06 Feb 2023 18:13:18 GMT</pubDate>
            <atom:updated>2023-02-06T18:45:26.927Z</atom:updated>
            <content:encoded><![CDATA[<p>The way in which ideas are communicated to players matters as much as the ideas themselves. I will discuss five characteristics that I have believe are essential to most effectively present propositions to players through my coaching and playing experiences.</p><p><strong>Know When to Say Something</strong></p><p>A very valuable trait for a player is to have a strong filter for information that is presented to them. But, not all players have a good filter and take almost any advice that is given by a coach. This means that coaches need to possess the filter and present information in a timely fashion. The vast majority of coaches have very good intentions which leads them to be eager to help when and where they can. However, I’d argue that this over eagerness can be detrimental for many players. Different parts of the year call for different types of coaching styles. Here is a very simple layout of how aggressive coaches should be with propositions dependent upon the different times of the year.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oc4B4NbkP36V2KONt9LtGQ.png" /></figure><p>Immediately after the season is the time to controllably and carefully unload the information you have compiled over the season or at an end of the season analysis. I think it is best to have a general idea of the changes a player needs to make going in to the off-season. The plan begins to become more specific during the early off-season and deployment of the plan takes place during this time and throughout the mid off-season. Tinkering and greater specification of the plan occurs as well. Tinkering and adjusting should start to subside around the late off-season / pre-season and the implementation of the plans should start to take place. This would include higher intensity sessions and live at-bats. The season can be a tough time for coaches who are focused on the development of their players. Obviously you want to help, but making any major overhauls in-season could be catastrophic for the player’s performance. This is an old tweet from Matt Daniels but a good one that I’d like to reference:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CVVpInnX-_0QEGnmL8_ntw.png" /></figure><p>In-season adjustments should ideally be subtle. Something like a mechanical overhaul is not subtle and should almost always be avoided. Things such as game management, arsenal usage, basic pitch design (minor tweaks), minor mechanical cues (apply intent later, more back hip connection, etc) could be worked on in-season.</p><p><strong>Always Be Ready</strong></p><p>Being ready is always preached to players. Coaches should take their own advice. If a player approaches you with a questions, they want an answer. Having advanced analytics and information somewhat readily available will allow coaches to address the player’s question in a timely fashion. Being unprepared could cause a rushed analysis that contains iffy information with subpar suggestions. Any suggestions need to have great confidence as one poor suggestion could lead to major issues for the player.</p><p><strong>Know Your Crowd</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/990/1*G9GP00qK-2qFbKqbmfje2w.png" /></figure><p>It goes without saying that the majority of players I coach (ohio high schoolers) are very different from big leaguers. So, propositions to players should be approached differently dependent upon the playing level. With high schoolers, I have found that it is a lot more trial and error to determine an actionable development plan. There is such a high level of variance in everything that they do. From a pitching perspective, their pitch shapes, velocity, and command have much greater variance than a Major League pitcher. Their movement patterns and pitch shapes are not as concrete as a professional pitcher as they do not have large amounts of innings under their belt. In my experience, working with higher level pitchers means you have to be more calculated in your propositions and plans. The large workload that they likely possess comes with more concrete movement patterns, pitch shapes, and better feel for pitching. Hence, they are oftentimes easier to mess up than they are to help. Working with lower level (younger) pitchers means there can be much more trial and error to help them figure out how they can be most successful.</p><p>Knowing your crowd is also applicable to different types of players at the same level. There are players who can handle large amounts of information well and those who cannot. Grouping players into those two categories can certainly help in how information can be presented to them.</p><p><strong>Create an Open Dialogue</strong></p><p>Before approaching a player and just saying your proposition, ask questions first. This will give better insight as to why they are doing what they are doing and what they are feeling. As much as you think you are teaching your players, they oftentimes teach you as well. There’s the potential that they provide insight or information that does not support your proposition at all or that changes it. Asking questions creates an open dialogue where the player can voice their opinion on a topic. This makes certain that they player has a say in their development. It is their career and they should always be in the driver’s seat.</p><p><strong>Monitor the Progress</strong></p><p>After a proposition is made, the progress needs to be monitored. Even with well thought out and seemingly safe plans, there is always the potential that it goes poorly. Keeping tabs on how the player feels / what they think, video, data, and in game results will be key. This can help to make further tweaks to the plan or to quickly abort the plan if it starts to go south.</p><p><strong>Conclusion</strong></p><p>The role of a modern baseball coach should consist of: coach, data interpreter / filter, and even make shift psychologist. In my opinion, these characteristics are all essential in being able to make effective propositions to players. The timing, readiness, knowledge of your crowd, creation of an open dialogue, and monitoring of progress are all very important.</p><p>As my coaching journey continues I am sure that these are likely to change or be added to. But for now, these are the five characteristics that I try to embody to present propositions most effectively.</p><p>For more baseball content checkout my twitter!</p><p><a href="https://twitter.com/RileyFelt29">https://twitter.com/RileyFelt29</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=694b5e8da198" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Adjustability > Repeatability]]></title>
            <link>https://medium.com/@rileyfeltner/adjustability-repeatability-271cb17dd45f?source=rss-9e64efae5337------2</link>
            <guid isPermaLink="false">https://medium.com/p/271cb17dd45f</guid>
            <category><![CDATA[baseball]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[r]]></category>
            <category><![CDATA[pitching]]></category>
            <dc:creator><![CDATA[Riley Feltner]]></dc:creator>
            <pubDate>Sun, 22 Jan 2023 18:19:31 GMT</pubDate>
            <atom:updated>2023-01-22T18:54:49.689Z</atom:updated>
            <content:encoded><![CDATA[<p>Many pitchers have heard that they need to “repeat their delivery” from pitch to pitch, pitch type to pitch type, inning to inning, game to game. A recent video I watched from Randy Sullivan of the Florida Baseball Armory highlights this topic. Randy mentions that it is impossible to repeat your delivery. However, the highest level pitchers are very good at adjusting. So, our focus should be more geared towards adjustability and not repeatability. In this blog, I will use the vertical release height (release z) and horizontal release height (release x) to highlight the repeatability (or lack there of) of MLB pitchers.</p><p>In my findings, the RHP sweeping slider best depicts adjustability in the pitching delivery. To get elite sweep on a slider, a pitcher often cannot simply adjust their wrist angle, grip, or finger position. They are adjusting their vertical and horizontal release points as well. Let’s take a look at the MLB’s most horizontally breaking sliders and those pitcher’s average differences in vertical and horizontal release between their fastball and slider. This is very likely subconscious and is their bodies way of allowing them to really brush the side of the baseball to get elite glove-side movement. With that being said, mechanical cues after front foot strike can be a very slippery slope. I would be hesitant to suggest a change in the vertical and horizontal release with the majority of pitchers. In the <strong>Conclusion, </strong>I will discuss how this information can be useful for players and coaches.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Dkw2gvuHGIVfrSwLNG07HA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*b8RFIqDXIKxn4XdSGSap1A.png" /></figure><p>On the left is the vertical release difference between the pitcher’s fastball and slider. On the right is the horizontal release difference between the pitcher’s fastball and slider. To create elite sweep on a slider, the release point of the pitch likely will be different from the fastball.</p><p><strong>Hitters Probably Cannot Tell</strong></p><p>Here’s some information to give you a context of how much MLB pitcher’s vertical release height changes between fastballs and sliders.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/702/1*axAfhjue9iJfUw9t_LUJ8Q.png" /></figure><p>I took a look at the absolute value top 10 vertical release height differences in 100+ inning starting pitchers. Here were 3 pitchers who’s vertical release height changed greatly between fastballs and breaking balls. Let’s start by taking the most extreme example of this in José Urquidy (98th percentile vertical release height difference). José has a 9.4&quot; vertical release difference between his fastball and slider and a horizontal release difference at 5.9&quot;</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RKwK8c9j7tvpQCw049NMfw.png" /></figure><p>José’s results- FB: 0.271 ExAVG, 0.533 ExSLG | SL: 0.231 ExAVG, 0.380 ExSLG.</p><p>The fastball results are not great, but not awful either. However, José does not have many outlier fastball qualities besides the above average vertical break.</p><p>Now for a less extreme example. Max Scherzer (82 percentile vertical release height difference) has a 2.8&quot; release height difference and a 1.7&quot; horizontal release difference.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RIWGura_yrFWpPA-nUhpXA.png" /></figure><p>Max’s results- FB: 0.229 ExAVG, 0.425 ExSLG | SL: 0.154 ExAVG, 0.215 ExSLG</p><p>One more example. Here is the largest deviation between fastball and curveball vertical release height amongst starting pitchers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VBByzJki7ErX2_5grZgg9A.png" /></figure><p>Shohei’s results- FB: 0.277 ExAVG, 0.372 ExSLG | CB: 0.180 ExAVG, 0.269 ExSLG.</p><p>José Urquidy, Max Scherzer, and Shohei Ohtani are fairly extreme examples of pitchers who’s release height differs between their fastball and breaking ball. If hitters could pick up on this difference, they would be taking advantage of this. None of the above pitchers are anywhere near unsuccessful and none have terrible results on either pitch type. This tells me that any sort of tipping after foot strike is highly unlikely to be recognized by hitters. Therefore, release point differences should not be a cause of concern unless it is extremely dramatic.</p><p><strong>Low, Median, and High Variance Fastball Horizontal and Vertical Release</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5mAPoKy6ZZQJd6DcHIC2Tw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mEYA1NBf445EnE-g7y_QZA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*39df9dJpJFxsVBZysEbG_A.png" /><figcaption>From Left to Right: Low, Median, High Variance Horizontal Release</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sS9tiJnSc1ul2yV4ZpH-cA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_JZxu84Zfa8UcfNbyiHlVw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S-R2pd-08fzMiOMAHs6CKQ.png" /><figcaption>From Left to Right: Low, Median, High Variance Vertical Release</figcaption></figure><p>I picked pitchers that were in the ≤ 10th percentile, around 50th percentile, and ≥ 90th percentile for their release variance. As we can see, there is a lot of variance in a pitcher’s fastball release point throughout a season. This still holds true when filtering by zone location.</p><p><strong>Conclusion</strong></p><p>I am in agreeance with Randy Sullivan in that we should not preach repeatability, but rather adjustability to pitchers. This could be reflected in drills that allow pitchers to throw from different positions. Drills that preach athleticism and adapting to certain “feels” / movements rather than hyper focusing on repeating them. To create the desired movement on a pitch, changes in grips and wrist positioning are likely not enough. This is obvious when looking at the MLB’s most horizontal breaking sliders. Most of their vertical release heights change rather significantly.</p><p>I do not think that changing the release point on certain pitches should be taught to pitchers. I could see this being a very slippery slope that could have very negative effects as specific mechanical changes after front foot strike can be very hard to implement. If this is an avenue that a coach thinks could be beneficial to a player, it could be presented in a round about fashion. For instance: a coach could tell the player to really get on top of a curveball, or to feel like they are really getting around their slider. In turn, this will probably create some change in their release point. Coaches should also be aware that this is a natural adaptation for some pitchers to create the desired movement on certain pitches. It should not be discouraged if it is already present or begins to occurr. Pitcher’s fastball release points will change throughout a season, outing, and even pitch to pitch. Hence, repeating your delivery is impossible. Adjust don’t repeat.</p><p>Checkout the code for this blog on my GitHub: <a href="https://github.com/rileyfeltner">https://github.com/rileyfeltner</a> . Follow me on Twitter for all baseball content: <a href="https://twitter.com/rileyfelt29">https://twitter.com/rileyfelt29</a> . As my final undergrad semester and first coaching season is heating up, I will not be quite as active on here as I’d like. But, I will post whenever I am able!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=271cb17dd45f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Using Driveline’s Motion Capture Data to Target Important Lead Leg Pitching Attributes]]></title>
            <link>https://medium.com/@rileyfeltner/using-drivelines-motion-capture-data-to-target-important-lead-leg-pitching-attributes-e65d979f94e1?source=rss-9e64efae5337------2</link>
            <guid isPermaLink="false">https://medium.com/p/e65d979f94e1</guid>
            <category><![CDATA[biomechanics]]></category>
            <category><![CDATA[pitching]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[baseball]]></category>
            <category><![CDATA[r]]></category>
            <dc:creator><![CDATA[Riley Feltner]]></dc:creator>
            <pubDate>Wed, 11 Jan 2023 19:49:38 GMT</pubDate>
            <atom:updated>2023-01-12T00:19:53.988Z</atom:updated>
            <content:encoded><![CDATA[<p>The lead leg in the pitching delivery has been a hot topic for the past several years — and rightfully so. However, I have found that there is still some confusion surrounding what a “good” lead side leg looks like. In this analysis, we will be looking at the lead side leg independent of other variables. There’s a great deal that goes into the pitching delivery and achievement of the “optimal” movement patterns. And notably, the “optimal” movement patterns are not consistent from player to player. The reason I will put “good” and “optimal” in quotation marks because “good” and “optimal” are not always the best fit for every player.</p><p><strong>Defining The Variables (source: </strong><a href="https://github.com/drivelineresearch">https://github.com/drivelineresearch</a><strong>)</strong></p><p>“lead_knee_absorption_fp_br”: energy absorption across lead knee between foot plant and ball release (J)</p><p>“lead_hip_absorption_fp_br”: energy absorption across lead hip between foot plant and ball release (J)</p><p>“lead_knee_transfer_fp_br”: energy transfer across lead knee between foot plant and ball release (J)</p><p>“lead_hip_transfer_fp_br”: energy transfer across lead hip between foot plant and ball release (J)</p><p>“lead_knee_extension_angular_velo_max”: peak knee extension angular velocity (deg/s)</p><p>&quot;lead_knee_extension_from_fp_to_br&quot;: difference in lead knee extension angle between foot plant and ball release (deg)</p><p>&quot;lead_grf_z_max&quot;: peak vertical ground reaction force on the lead leg (N)</p><p><strong>Lead Leg Metrics And Peak Vertical Ground Reaction Force</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/970/1*TgFqqGC3dVHSqZIgzrqgOA.png" /><figcaption>R² of lead_grf_z_max and lead leg metrics</figcaption></figure><p>The correlation between lead_knee_transfer_fp_br, lead_hip_absorption_fp_br, and lead_knee_extension_angular_velo_max all seem to have fairly significant correlations to the amount of force vertically applied into the ground.</p><p>The lead knee transfer is shown to have the strongest correlation to peak vertical ground reaction force. A pitcher’s lead leg will strike the ground in a flexed position and as they initiate the throw, the lead leg will begin to stiffen out. The lead leg stiffening is what captures the potential energy built up and fires that flow of energy back up the kinetic chain into kinetic energy.</p><p>The lead hip absorption showed positive correlation as well. In a lot of the pitchers I train, I see a “leaky” front side hip. Meaning it is bowing out towards their glove-side. Let’s compare baseball to skiing so we can get a good understanding of why a “leaky” front hip could be an issue.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/612/1*QJPegVAM4YMJP_voHPv-tA.jpeg" /><figcaption>Credits: Onfokus</figcaption></figure><p>The skier wants to go to the right. So, his hip is leaking over his right foot to go in that direction. In baseball, we are throwing the pitch straight ahead. So, any leaking of the hip is typically not ideal. We ideally want our hip to stay mostly in place between foot strike and ball release as we rotate around it. This is what I would refer to as a lead hip block.</p><p>The lead knee angular velocity makes sense in the context of vertical ground reaction force but does not matter much in terms of what really matters: velocity. I will discuss this further below.</p><p>I bet I know what you were thinking: Why don’t we just see how it correlates to velocity?! Well, defining a “good” lead side leg should not ignore how much force you are putting into the ground. In fact, it has the largest correlation to pitch velocity when examining the lead side leg.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*k1LIkGt5Q7QxfEh66yqoaA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qNtzt2mQtcZAfpudWY3yjA.jpeg" /></figure><p>When looking at the non-linear model, there is much greater effect of increases in lead leg vertical ground reaction force on velocity for pitchers in the 75mph to 85mph range. This tells us that there could be a lot of low hanging fruit in amateur pitchers to improve their lead leg. In throwers that are 85mph +, there usually will not be as much room for improvement.</p><p><strong>Lead Leg Metrics And Velocity + Debunking A False Assumption</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/884/1*6uS7VnqFqjhvyv8SsfBk1Q.png" /><figcaption>R² of velocity and lead leg metrics</figcaption></figure><p>Here we are! We already discussed lead knee transfer and lead hip absorption. Now we will touch on the lead knee extension from foot plant to ball release. I think this is best described by a couple pictures of me when I played to show what is bad, then I will use a couple more pictures to show what is good.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NcSORGFBc6mA5JiINSEtCA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*K6xUMDh6JPeHKQMRmcmb8w.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BXDLMXP5bddCH6zWxxH-Dg.jpeg" /><figcaption>Riley Feltner</figcaption></figure><p>As you can see here, the difference between my lead side knee extension at foot strike compared to ball release and even after ball release did not change much. Now, let’s take a look at someone who does have a good deviation between their foot plant and ball release knee extension.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FGI0R6xy7bY0E3869pK-UQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_VmrMwbDDO1qq3DKl7uWLQ.png" /><figcaption>Spencer Strider</figcaption></figure><p>The difference here should be obvious.</p><p>Oftentimes, I see coaches or athletes who think a lead leg is good because it reaches full extension at some point in the delivery. Without slowing down the video of me, it would appear that I had a stiff lead side leg at ball release, which I did not. Therefore, the angular velocity of the lead knee has the least correlation to velocity. Sure, you can produce more ground reaction force by having a high knee angular velocity. However, the timing of it matters much more. The timing can be represented by the lead knee extension from foot plant to ball release.</p><p><strong>Conclusion And Takeaways</strong></p><p>The lead side leg is a vital piece to the large puzzle of pitching biomechanics. Improvements on the lead side leg for lower-level pitchers appears to reap more velocity benefits. The most important pieces of the lead leg are the lead knee transfer from foot plant to ball release, lead hip absorption from foot plant to ball release, and lead knee extension from foot plant to ball release. Coaches oftentimes look for a quick and easy way to evaluate a pitcher’s mechanics. In the context of the lead leg, it is often evaluated on the visual basis of how much knee extension they get.</p><p>Here are two things coaches can visually look for (ideally at least through slow motion iPhone video) that would somewhat accurately evaluate a lead side leg if they are not able to quantify movements through a motion capture system:</p><p>1.) How does knee flexion change from foot plant to ball release?</p><p>2.) What does the lead hip look like? Is it bowing out to the side as ball release is initiated? Or is it staying in place and the pitcher is rotating around / through the hip?</p><p>My code for this blog can be found on my GitHub: <a href="https://github.com/rileyfeltner">https://github.com/rileyfeltner</a> . The data that I used can be found on Driveline’s GitHub: <a href="https://github.com/drivelineresearch">https://github.com/drivelineresearch</a> . Huge shoutout to Driveline for making the data open-source! I barely scratched the surface of the data on this blog so check it out!</p><p>For all of my baseball content checkout my twitter: <a href="https://twitter.com/rileyfelt29">https://twitter.com/rileyfelt29</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e65d979f94e1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Lou Trivino Analysis and Advanced Report of Significant 2022 OAK & NYY Performance Differences]]></title>
            <link>https://medium.com/@rileyfeltner/lou-trivino-analysis-and-advanced-report-of-significant-2022-oak-nyy-performance-differences-ebba02119c0f?source=rss-9e64efae5337------2</link>
            <guid isPermaLink="false">https://medium.com/p/ebba02119c0f</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[r]]></category>
            <category><![CDATA[baseball]]></category>
            <category><![CDATA[pitching]]></category>
            <dc:creator><![CDATA[Riley Feltner]]></dc:creator>
            <pubDate>Tue, 03 Jan 2023 18:32:20 GMT</pubDate>
            <atom:updated>2023-01-03T18:42:31.077Z</atom:updated>
            <content:encoded><![CDATA[<p>Performance naturally ebs and flows throughout a season to some degree. When taking a look at Lou Trivino’s 2022 campaign, the drastic difference between his performance in Oakland versus that in New York does not appear to be caused by this natural “ebbing and flowing”. The veteran spent the first 4 months (April — July) in Oakland before being traded to the Yankees where he would finish his season (August — October). In 32.0 innings in Oakland, Trivino posted a 6.47 ERA, 4.08 FIP, 0.256 xBA, 0.445 xSLG. In 21.2 innings in New York, Trivino posted a 1.66 ERA, 3.34 FIP, 0.252 xBA, 0.346 xSLG. Trivino should not have given up too many more hits between the two teams indicated by the xBA. However, he was getting hit harder in Oakland than he was in New York indicated by his xSLG values and barrel %: (OAK: 9.4%, NYY: 3.4%).</p><p><strong>2019 And 2021 Arsenal Breakdown</strong></p><p>I took a deep dive into Lou’s 2019 and 2021 arsenal. I excluded the shortened 2020 season due to the small sample size. Some of his pitches are significantly different while others have not changed much.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zBeGxluNb-pLVdMH5hbwPQ.png" /><figcaption>Lou Trivino 2019 and 2021 Pitch Break Charts</figcaption></figure><p>Lou’s fastball is not near the premium velocity that it had in 2019 (97.4 mph avg) as it averaged around 95.6 mph for the previous two seasons. This makes the location of the pitch more noteworthy now once we dive into the 2022 season. The cutter is obviously lacking glove-side horizontal break as we compare 2019 (-0.9&quot;) to 2021 (2.9&quot;). The results suffered, 2019: 0.198 ExAVG and 2021: 0.317 ExAVG. The changeup showed significant improvement from 2019 to 2021 as it gained depth (3.8&quot; to 0.7&quot;) and ExAVG improved (0.204 to 0.166). The sinker also showed improvements from 2019 to 2021 as it gained about an inch of run and half an inch of depth — ExAVG improved (0.362 to 0.265).</p><p><strong>2022 Arsenal Breakdown OAK vs NYY</strong></p><p>As previously mentioned, 2022 was an odd year for Lou Trivino. He struggled greatly in Oakland and thrived in New York.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vND5Ma-f5TN7A4qr_BNklw.png" /><figcaption>Lou Trivino OAK and NYY Pitch Break Charts</figcaption></figure><p>There was one pitch that got significantly worse after he transitioned to the Yankees: the cutter. I will address these below in the section: <strong>What Lou Trivino Could Improve Upon For 2023.</strong></p><p>Bye Bye curveball. Trivino threw his final curveball of the 2022 season on May 21. This is a very impressive and courageous move to make mid-season. There was no dipping the toes in the water and seeing if it worked. Trivino threw his first Major League slider the next outing at -20.1 HB and -3.4 VB. In translation: an amazing sweeper. This was a good move considering the curveball was getting significantly worse results than previous years (2022: 0.578 ExAVG, 2021: 0.093 ExAVG). Completely abandoning a pitch with such good results in 2021 could not have been easy. But, the movement profile was worse. It had 1.1&quot; less of depth on a pitch that already lacked depth. It was also thrown much slower (-1.2 mph from 2021) likely in attempt to regain some of its depth. Due to the large amount of sweep and lack of depth, it is easy to see why he decided to bail on the curveball and switch to a slider in attempt to add carry and a little more sweep. After the trade, the cutter got worse, the fastball had a more defined role, the changeup and sinker stayed consistent, and he already had been throwing the sweeper in Oakland. So what changed? Let’s take a look!</p><p><strong>What Changed From The A’s To The Yankees?</strong></p><p>The movement profile on his pitches didn’t improve. However, the execution and pitch usage did improve significantly.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ypDqta6rlEGqYNPep7Hd8A.png" /></figure><p>Lou Trivino was throwing his worst pitches most often in Oakland. He heavily relied on a sinker and fastball that performed the worst when looking at both their xWOBA and Miss %. His slider was his best pitch and he was not using it nearly enough. To be fair, he had just started throwing this pitch in late May. That is asking a lot of a player to immediately start throwing their newest pitch most often. But, that was what he did after the trade in New York.</p><p>In two strike counts, he cut his fastball usage almost in half, rarely threw his cutter anymore, dropped his change-up usage, and went away with the curveball. This was all in an effort to throw the sweeper more often.</p><p>In non-two strike counts, Lou slightly decreased his fastball, sinker, cutter, and changeup usage while dropping the curveball to get to his sweeper more frequently.</p><p><strong>The Slider</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3WUzZRTZVJB0uITJxXq4YQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AxUYonArI6rdg0e70cMZDA.png" /></figure><p>The heat maps I produced show that his slider was much more on the plate in New York. Let’s check out the difference by two strike and non-two strike counts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VR4v4waj8u_h-orYIfP0hA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5nZrInnb0D0m5BMRixY7RA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iyl9AtV7TSY346sQmw2e1Q.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o_2UL0EnSFcjM8mHUxumhQ.png" /></figure><p>His New York heatmaps resemble what I would refer to as “owning your zones”. In Oakland, Lou’s sweepers were not getting down and glove side as consistently in two strike counts and it was off the plate too often in non-two strike counts. The movement profile was there from the start. But, Lou showed much better command of the pitch once he got to New York. Lou showed the ability in New York to consistently get his slider to the intended location for punch outs and for strikes. The uptick in command and usage was beneficial for Lou. The results were favorable considering the increased amount that the slider was in the zone in New York.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*y3krh6xSMP0fpGzGwMni4A.png" /></figure><p><strong>The Fastball</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*E0nVb_fShg6saX914atMrw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BjPrRUvRaKmRBZTJ5qW2xA.png" /></figure><p>Lou’s fastball had a much more defined roll in his arsenal in New York: get swing and misses. This seems obvious that Lou was trying to get his fastball up more often in New York. Although the xWOBA is higher, the miss% went from 26.3% to a whopping 40.0%. He dropped his fastball usage in both two strike and non-two strike counts once he got to New York. However, in New York he was throwing his fastball more often in two strike counts than he was in non-two strike counts. This tells me that he realizes the elite miss% he obtains with the pitch up in the zone and is trying to leverage it.</p><p><strong>What Lou Trivino Could Improve Upon For 2023</strong></p><p><strong>Sacrifice Velocity for More Movement on the Cutter: </strong>Lou’s cutter has not averaged any glove-side break since 2020 (-1.0&quot;). In both 2021 and 2022, it was backing up to the arm-side at 2.0&quot; and 2.9&quot; respectively. Lou is still throwing his cutter at the same velocity he was when his fastball was averaging 97.4 in 2018 and 2019. I think the pitch would fare much better if he could regain the tilt he had in 2018 (11:15) and sacrifice some velocity to get more on the side of the ball which would create the glove-side break the had in 2018–2020. His 2020 cutter gave us a glimpse of what this could look like: 90.6 mph, 6.1 VB, -1.0 HB, 12:15 Tilt, 0.112 ExAVG, 0.150 ExSLG. I think if Lou could get closer to 11:00 tilt and have it more 90–91 mph, we could see a pitch that is getting much more glove-side movement with not too significant of a velocity decrease.</p><p><strong>Closer Than You Think</strong></p><p>Lou Trivino’s season was very interesting to me because it shows that many high level pitchers are just a few changes away from becoming elite. Lou did not drastically change a whole lot. He noticed his curveball was suffering so he made a change. A fairly significant change, but it obviously was not all that difficult for him as the slider quickly became his best pitch. Lou also adjusted the plan of attack for his fastballs and sliders.</p><p>Lou showed something very admirable in 2022. He was struggling, badly. He took it into his own hands to fix the issue. It appears that Lou is a pitcher who is very open to suggestions and change. I believe we will see a big 2023 season from Lou Trivino. One that is armed with his newly found slider, better pitch usage, and better zone locations.</p><p>Thanks for reading my first blog. For more active (but less in depth) baseball content visit my Twitter: <a href="https://twitter.com/rileyfelt29">https://twitter.com/rileyfelt29</a></p><p>Code for this blog is posted GitHub: <a href="https://github.com/rileyfeltner/Lou-Trivino-Analysis">https://github.com/rileyfeltner/Lou-Trivino-Analysis</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ebba02119c0f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>