Model Drift and Retraining (13.3)

Some thoughts on the future of reinforcement learning.
Code for This Video:
github.com/jef...
~~~~~~~~~~~~~~~ COURSE MATERIAL ~~~~~~~~~~~~~~~
📖 Textbook - Coming soon
😸🐙 GitHub - github.com/jef...
▶️ Play List - • 2024 PyTorch Version A...
🏫 WUSTL Course Site - sites.wustl.ed...
~~~~~~~~~~~~~~~ CONNECT ~~~~~~~~~~~~~~~
🖥️ Website: www.heatonrese...
🐦 Twitter - / jeffheaton
😸🐙 GitHub - github.com/jef...
📸 Instagram - / jeffheatondotcom
🦾 Discord: / discord
▶️ Subscribe: www.youtube.co...
~~~~~~~~~~~~~~ SUPPORT ME 🙏~~~~~~~~~~~~~~
🅿 Patreon - / jeffheaton
🙏 Other Ways to Support (some free) - www.heatonrese...
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#reinforcementlearning #gymnasium #PyTorch #DeepLearning #PyTorchTutorial #MachineLearning

[@warssup]

Hi Jeff interesting topic. I actually discussing the topic, that many acadamic (but also industrial) dataset are never tested for data drifts (especially if they are tabular). Basically one simply assumes there is no data drift and a common full random k-forld cross validation with test dataset is executed and the results are seen as stable. However, they are not. My approach for the dataset of my specific application was, that I compressed the dataset with a PCA (but any other dimensional reduction method would work as well) to plot it in a 2D scatter plot, but as color index I display class and data point index. By this, anyone could see that there is a clear correlation between time and the clustered derived by the PCA and this delivered a perfect explanations for the experimental results, that also have shown amongst others that reported reality in academic paper is often not the true reality that one would face in the real world deploying a model.

[@Beauty.and.FashionPhotographer]

Off topic question,...please don't laugh,..... Mac M2,M3 Silicon and running windows on it, solely for the purpose to run and do gen ai or even LLMs(?) so much faster ,.....(via eGPU docks with thunderbolt4 to the macs).... there is almost no info anywhere to be found,....would you ever consider dedicating a video to this topic? ..... There are so many in the creative advertising world, who by default all only work on macs,... and who want to run gen ai faster with such external graphic cards, be it Nvidias RTX 3090 or 4090,.... soooo many who are "willing" to run windows just for the much faster speed of rendering gen ai (auto1111 comfyui or LLms),..... but there is nothing really out there to help get this going. Yet gen ai "real life application" can only be found in advertising, where money can be made with gen ai is being used in ads and eventually in tv commercials. And still almost no one out there showing how to do this, which hardware, what cables, what drivers, which macs and so on..... it would be totally amazing if a top Expert like You, could show , how it gets done