Let's Build Some Cool Sh*t
Our Virtual Monthly Machine Learning Meeting Notes for October 27, 2022
Hey Outliers,
The highlight of the town hall was surrounding individual projects. Currently the community is building through two projects: NBA Stats and Operation Blossom. There was a new member that was working on a Computer Vision project analyzing cancerous tissue in X-ray images. In both projects, we were in the “evaluating the model” phase. We ran Convolutional Neural Networks but now how do interpret these CNNs to understand where the model is failing to learn from the training data. A possible technique is Grad-CAM (Gradient Weighting Class Activation Maps). In summary, Grad-CAM uses gradients in a CNN layer to produce a coarse localization map to highlight regions that contributed to the outcome of the prediction.
In addition, we talked about the trap I fell in when working on NBA project. When starting a project, it is important to know what the end goal is. Why are we starting this project? In a job setting, there might be a business case that needs to be solved such as “Can we forecast sales for the next quarter?” To solve this case, we gather time series data → do some analysis to see if it is ready for modeling → model it → extract some insights → repeat (if necessary). Data is a beautiful concept and a lot of insights can be pulled from data. However if we do not know what we are looking for, then we will be aimlessly looking at the data with no end.
I, unfortunately, fell into this trap. Being a fan of the NBA and having access to a plethora of data of NBA players, I was just plotting visualizations and applying stats to the data to see interesting things. Even though bits of analysis were being made, there was no actual insights because I was going from analyzing players’ points to rebounds to minutes and back to points. I was not sewing a story using the data. After a few days, I got overwhelmed with the data and the analysis with nothing to show for other than graphs and statistics that have no correlation with each other. If I had a goal in mind, I could have made visualizations and analysis towards that goal and have an end-to-end solution. Lesson: Even though it is a personal project, treat it like a business problem. Why are we doing this? What problem are we trying to solve? Once we have this end goal, we can create analysis supporting the goal (Thank you @Tibbee)
Links
Road Map for Operation Blossom
Community Updates
Book Study
Advanced Topics Group
A Huge Shout Out to our Discord member @QwaziRabbit. He is leading the efforts of this group, presenting every day and spearheading the discussions that happen while maintaining a full-time job!
Probabilistic Machine Learning Chapter 7
Check Discord for schedule
Interpretable Machine Learning
Chapter 4 Bayesian Rule Lists
Check Discord for schedule
Interesting Links
https://www.nature.com/articles/s41586-022-05172-4
https://365datascience.com/free-days-2022/
Events
https://datasciencefestival.com/event/dsf-online-2022/ (Virtual)
https://www.techday.tech/techday-expo (New York)
https://www.snowflake.com/build/ (Virtual)
Thank you for being part of this community! Since the holidays are coming up, we will not be having another meeting until the New Year. So Happy Holidays and see you in January!
Feature Photo by Xavi Cabrera on Unsplash