Derek Grifka

A cool dataset was released today on Baseball Savant. It shows the average bat speed, swing length, swing type and player impact for 2024 YTD. Some vocab for the post: - Blasts are pitches that attain at least 80% of the max possible exit velocity based on incoming pitch speed and swing speed that ALSO have a swing speed of 75+ mph. In other words, these are swings with a high speed swing and exit velocity and more propensity to turn into base hits our go out of the park. - Batter run impact is the cumulative run impact of all the at-bats of the player With this data, can we see how "batter run impact" and blasted balls are affected by swing dynamics? Yes! If we first look at just bat speed and swing length, it seems that swings that are too long or too compact don't tend to perform as well as those from 7.25 - 7.5 feet. Similarly, the batters with swings that are too slow (< 67.5 mph) or way too fast (over 77 mph) don't work as well as those running between ~71-75 mph, right near or above the average of 71.8 mph. Meaning the slow swings aren't generating as much power, and the extremely fast ones are almost hacking wildly. Both aren't great. So, hitting in this Goldilocks zone is best, and deviating from it tends to be suboptimal. One case in point, Marcel Ozuna has the highest batter run impact at +13.2 and swings at around 75 mph average with a 7.4 foot swing length. Giancarlo Stanton is at a vastly higher 80 mph bat speed and 8.3 swing length and is only at 1.67 batter run value, vastly lower. But reinforcing the idea that bat speed is important (and how slower swings just don't generate the same impact), you do see how blasts (75+ mph swing speed generating high exit velocity) are correlated with run impact, but there can be too much of a good thing. And with greater bat speed, there is a propensity for more rib/oblique injuries, a fact pointed out in a great article today in the The New York Times. Speaking of which, the NYY's oft-injured Giancarlo Stanton might want to turn his swing down a notch. He's been injured on average 45% of games played back to 2019 and doesn't need to miss more time. Lots more to come from this, I'm sure. For now I am happy to see my Philadelphia Phillies with the best record in baseball at 29-13, and currently in first place over the Barves. And it's always nice to see the Phils pull off a win against the "Stinkin' Mets" as my Dad "fondly" refers to our rivals from Queens. Third baseman Alec Bohm continues his hot season, batting at .329 with a .914 OPS. FWIW, his swing stats averages are 72.2 mph and 7.3 feet swing length...with a 10.5 batter run impact. 4th best in the Majors, and right in that sweet spot. #phillies #philadelphiaphillies #mlb #baseballsavant #sportsanalytics

20

Inspired by the #2024Olympics, I just wrapped up a little project that combines two things I love: volleyball and data science. I put together a Python script called VolleyForm AI. It’s a simple tool that uses AI to help "spot check" my volleyball skills by analyzing my form for passing, setting, attacking, and blocking. 🏐 Here’s what it does: 𝗣𝗮𝘀𝘀𝗶𝗻𝗴: Detects if your arms are too high or if your elbows aren't straight. 𝗦𝗲𝘁𝘁𝗶𝗻𝗴: Checks if your elbows are too close or far apart and if your hands are closed. 𝗛𝗶𝘁𝘁𝗶𝗻𝗴: Evaluates your hitting elbow’s position and whether your guide arm is in the right spot. 𝗕𝗹𝗼𝗰𝗸𝗶𝗻𝗴: Makes sure your hands are open when blocking. 🔗 If you want to try it yourself, the code is up on GitHub: https://lnkd.in/dYaxTSBm Would love to hear what you think or any ideas for improving it! #Python #Volleyball #DataScience #Olympics

313

32 Comments

🚀 New Blog Alert: Understanding regression metrics can be confusing sometimes, but it’s essential if you want to evaluate the performance of your models accurately. In my latest post, I break down the most important metrics in simple terms: 🔸 MAE (Mean Absolute Error) – A straightforward way to understand the average prediction error. 🔸 MSE (Mean Squared Error) – Helps you catch larger errors by squaring the differences. 🔸 RMSE (Root Mean Squared Error) – Gives you error in the same units as your target variable, making it easier to interpret. 🔸 R² (R-Squared) – Want to know how well your model fits the data? This is the metric for you! Whether you’re just getting started with regression models or looking to deepen your knowledge, this guide will help you navigate these metrics with clarity. Check it out and feel free to leave your thoughts! 💬👇 #MachineLearning #DataScience #RegressionMetrics #AI #MAE #RMSE #R²

8

⚙ Enhancing Baseball Data Analysis with R Shiny Widgets 🌟 Want to take your baseball data visualization to the next level? Let’s explore how incorporating widgets into your R Shiny app can create dynamic and interactive dashboards for displaying baseball data. ⚾📊 🌟 What are Shiny Widgets? Widgets in R Shiny are interactive tools that enhance user experience by allowing real-time data manipulation and visualization. From sliders to dropdown menus, widgets provide a user-friendly interface for exploring complex datasets. 🔍 Why Use Widgets? Interactivity: Widgets allow users to interact with data, making it easier to drill down into specific details. Customization: Tailor the user experience with a variety of input controls like sliders, text inputs, and date selectors. Insightful Visualizations: Enhance your data presentations with interactive elements that make insights more accessible and engaging. 🔑 Incorporating Widgets into Your R Shiny App: Here’s a simple example to incorporate widgets into your R Shiny app for displaying baseball data: library(shiny) # Define UI ui <- fluidPage(    titlePanel("Team Pitching Data"),  br(),   sidebarLayout(    sidebarPanel(           # pitcher select     selectInput(      "Pitcher",       label = "Select Pitcher(s):",      choices = levels(as.factor(TrackManCSV$Pitcher)),      multiple = TRUE,       selected = c()),           # date input     dateRangeInput(      "DateRangeInput",      label = "Select Date Range",      start = min(TrackManCSV$Date),       end = max(TrackManCSV$Date)), # Run the application shinyApp(ui = ui, server = server) 🎯 Use Case? Imagine creating an app that allows coaches, scouts, and front office members to explore pitcher performance metrics dynamically. By using sliders to adjust time frames or dropdown menus to select specific players, users can gain tailored insights and make informed decisions. 🤹‍♂️ Mastering Shiny Widgets: Incorporating widgets into your R Shiny app can transform static reports into interactive tools, providing deeper and more actionable insights into baseball performance. ✨ Ready to elevate your baseball data analysis with R Shiny widgets? Start building your interactive dashboards and unlock the full potential of your data! 👉 Have tips on using Shiny widgets? Share your experiences and questions below! #BaseballAnalytics #RShiny #DataVisualization #InteractiveDashboard #SportsData #R #programming #baseballdata #dataanalytics #collegebaseball #MLB #baseballoperations #TrackMan #database

38

Turnover can either derail a project or be just a bump in the road. Losing critical knowledge is expensive—but it doesn’t have to be. With the right documentation practices, high turnover becomes manageable, not disastrous. Here’s my approach to handling turnover as a Data Scientist: 𝗣𝗠𝘀/𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗟𝗲𝗮𝗱𝘀: Capture the big picture—why the project matters, key deadlines, expected outcomes, and the essential stakeholders. Ensure every detail is documented so the project doesn’t stall when someone exits. 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀: Document pipelines, key tables, runtimes, and critical Airflow links. This allows anyone stepping in to pick up exactly where things left off. 𝗗𝗮𝘁𝗮 𝗦𝗰𝗶𝗲𝗻𝘁𝗶𝘀𝘁𝘀/𝗔𝗻𝗮𝗹𝘆𝘀𝘁𝘀: Create a culture of documentation. Ensure data processes, methodologies, and insights are organized and accessible, so no knowledge is lost when team members move on. Being a Data Scientist isn’t just about crunching numbers—it's about being the safety net that keeps knowledge intact when your team changes. Technical skills matter, but adaptability and documentation make you indispensable. 👉 How do you handle turnover in your teams? #DataScience #Turnover #Documentation

2

🚩 We Don’t Gatekeep Here. Free Data Science, Analytics, and Engineering Books 📍Over the years, I’ve built quite a collection of books. Some I bought, some were passed down to me by my mentor, and others were thoughtful gifts. These books have been my go-to references countless times, and now, as part of this amazing community, I want to share some of them with you. 📍But don’t just collect them, make sure you actually read them or refer back to them when you need to. 📍Instead of dropping your email in the comments, please fill out the form, the link is in the comment section and I’ll send a few your way. 🚩One thing to note: I’ve got a mix of Data Science, Analytics, and Engineering books, so I’ll send them as they are, no time to sort them out right now. ♻️ Report to a fellow bookworm #dataliving #favouribude

242

119 Comments

Did a little more work on my sports analytics hobby project modeling #hockey data over the weekend for the start of the NHL playoffs. Below are the top 25 goalies in the regular season as determined by an IRT model with a discrimination parameter. The fitting process involved binning shots into one of ~500 hexagons making up the offensive zone, meaning the "item" in the IRT model is a shot from a specific area of the ice. Interestingly, I had to let the discrimination parameter vary below zero, which is uncommon in typical practice - you wouldn't include a test item that low-ability test-takers do better on than high-ability test-takers, but there are shots that low-ability goalies save better than high-ability goalies. Need to dig into that more... I do think the discrimination parameter overcorrects for bad/good defenses. Blackwood is behind an awful Sharks defense that gives up a lot of high-danger chances, so while he's better than his stats look, I don't think he's top 25. I haven't found a good way to incorporate defense quality in the model, but am open to any suggestions from the Bayesians in my network. Adding data-based priors to the parameters has so far only made the model prohibitively slow to fit. (As a #Canes fan, I'm thrilled to see that the model loves Anderson and Kochetkov's run of play to end the season, so I'm choosing to mostly believe the results.)

7

Learn how fine-tuning Phi-3 Vision on a custom dataset with tracking & evaluating using W&B Weave. From dataset prep to inference logging, see the full process and benefits. Check out the latest article from Brett Young: https://lnkd.in/gxddTTwb

15

PSA for folks working with Posit PBC's #great_tables packages: The package is really, really cool and the examples are very useful. One issue when rendering from e.g. a Quarto document: I don't know if it is a quirk or just user error, but in some contexts I only have to import the main package, but in others I get errors unless I import key submodules The error is along the lines of "Error in `py_call_impl()`: ! NameError: name 'style' is not defined" The fix is to import the submodules, eg "from great_tables import GT, md, html, style, loc" #python #datascience

3

#FlashbackFriday to when I transformed a 4-hour manual task into a 13-minute automated process! 👀 A few years ago, as Team Lead of Data & Development, my team of analysts were stuck in a tedious cycle—downloading Excel files and manually inputting data into a website (using Internet Explorer 😅). It took 4 people a few hours during month-end close, 𝘸𝘩𝘪𝘤𝘩 was 𝘵𝘩𝘦 𝘣𝘶𝘴𝘪𝘦𝘴𝘵 𝘵𝘪𝘮𝘦 𝘧𝘰𝘳 𝘶𝘴! Here’s how I changed that: 1. Created one Jupyter Notebook. 2. Used SQL to extract the data. 3. Implemented Selenium to automate the web navigation and data entry. With these tools, we reduced the process to just 13 minutes—handled by ONE analyst. Learn more about Selenium here: https://buff.ly/2ywa1Ac P.S: I first found Selenium when making bots to scrape Instagram… didn’t realize it had a work function too! 😂 #WebScrape #DataScience #Selenium #Automation

3

Generally agree with the sentiment of this. I assume the unpopular opinion for some is the revenue generation. It’s a North Star metric that only significantly mature data teams likely can realize. But value generation is still a spectrum, with revenue on the far end. If data teams can’t identify / articulate how operational efficiencies, cost avoidance, data literacy improvement, etc. are realized, then they are 1 Jupyter notebook model away from becoming a cost center. Don’t make consumer and product an afterthought

10

2 Comments

Day 37: Correlated Nested Subqueries I’ve been diving into correlated subqueries today, which are subqueries that rely on the outer query. They’re perfect for dynamic filtering based on the outer query’s context. This makes complex data manipulation a lot more intuitive. Example: SELECT player_name, age FROM players p WHERE age > (SELECT AVG(age) FROM players WHERE team_id = p.team_id); Excited to keep progressing 💃 #DataAnalysis #SQL #LearningJourney #90dayschallenge

5

2 Comments

With summer winding down, beach volleyball season is coming to a close. 🏐 As a longtime fan of the AVP Pro Beach Volleyball, I decided to combine my love for the sport with my data science background to create a fun project: 𝗽𝗹𝗮𝘆𝗲𝗿 𝘁𝗿𝗮𝗰𝗸𝗶𝗻𝗴 𝗼𝗻 𝘁𝗵𝗲 𝗰𝗼𝘂𝗿𝘁. This clip is from the 2024 AVP Manhattan Open, featuring Phil Dalhausser, a USA Olympian with 𝟲 𝗚𝗼𝗹𝗱 𝗠𝗲𝗱𝗮𝗹𝘀. The tracking (which is almost spot-on) maps out player movements on both sides of the court. The best part? 𝗧𝗵𝗲 𝗰𝗼𝗱𝗲 𝘄𝗼𝗿𝗸𝘀 𝘀𝗲𝗮𝗺𝗹𝗲𝘀𝘀𝗹𝘆 𝘄𝗶𝘁𝗵 𝗼𝘁𝗵𝗲𝗿 𝗴𝗮𝗺𝗲𝘀 𝘀𝗵𝗼𝘁 𝗳𝗿𝗼𝗺 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲 𝗰𝗼𝘂𝗿𝘁 𝗮𝗻𝗴𝗹𝗲 𝗮𝗻𝗱 𝗱𝗶𝘀𝘁𝗮𝗻𝗰𝗲! 🔗 The code is up on GitHub: https://lnkd.in/dDNbASfD Inspired by AVP Pro Beach Volleyball and Balltime, I’m looking forward to more volleyball data projects ahead. Next up: Ball tracking! If you have any ideas or insights, let’s connect and collaborate! #DataScience #Volleyball #Python #YOLOv8 #ComputerVision

19

This past offseason, I introduced the pitcher evaluation metric xwFIP in a series of three posts on The Paraball Notes. xwFIP is an advanced fielding independent pitching metric that not only takes into account walks, hit batsmen, strikeouts, and homeruns, but also the xwOBA on batted balls. The metric takes into account nearly 100% of plate appearances in its calculation. Moreoever, I have showcased that it has greater predictive power than each of ERA, wOBA, FIP, xFIP, xwOBA, and SIERA in predicting future run prevention performance of a pitcher. Each month this season, I'll post the MLB xwFIP Leaderboard. Who are the best pitchers in 2024 and which pitchers are likely to either improve or regress in the coming months? #BaseballAnalytics #MLB https://lnkd.in/eFX-Z6WG

13

If I had to start from 0 with Tableau here's what I'd do: (how I actually got noticed) → Put the majority of your learning time into your Tableau Public portfolio Here’s why: 1. **Instant Credibility**: Your Tableau Public portfolio acts as a live resume. Employers, clients, and peers can see your work, skills, and creativity in real time. No need to just talk about your expertise—you can **show** it.     2. **Community Recognition**: Tableau Public isn’t just a tool, it’s a community. Having high-quality visualizations can get you noticed, featured, and shared. This exposure can lead to new opportunities and connections that may never have come your way otherwise.     3. **Constant Learning & Improvement**: A public profile keeps you motivated to innovate. Each new visualization is an opportunity to experiment, learn new techniques, and raise the bar. The feedback and inspiration from the community only push you further.     4. **Visibility to Opportunities**: Recruiters, companies, and collaborators use Tableau Public to scout talent. A standout profile could open doors to projects, roles, or collaborations that others simply won’t get access to. The trick is to find a topic from interesting data sources that you enjoy.  You'll stick to it longer. 1. makeovermonday 2. data.world 3. kaggle This is exactly how I have landed interest in my Tableau expertise. And how I've been recruited for multiple positions. Pick a data source, viz it and sell your skills on Tableau Public.

19

5 Comments

The ability to make sense of charts and tables within a complex financial report remains a key challenge which seems to be on track for resolution. this could drastically improve the performance of RAG pipelines in the coming months.

6

“Numbers make people think, emotions make people act.” – I think I first heard this quote on the BiggerPockets Real Estate podcast and I've kept coming back to it throughout my career as a data scientist. While technical skills are crucial, the most valuable skill I’ve learned outside of data science is storytelling. You can have the most precise model and the cleanest data, but if you can’t connect it to people and their real-world challenges, it’s hard to drive meaningful action. The ability to communicate insights in a way that not only informs but inspires action has been a game-changer in my career. Whether it’s convincing leadership to back a project or helping teams understand the impact of data-driven decisions, the emotional connection through storytelling is what makes the numbers truly impactful. What non-technical skill has made the biggest difference in your career?#DataScience #Storytelling #Leadership #Communication

15