I'm writing a from scratch deep learning set of tools for educational purposes. This will turn into a series of blog posts. The tools may be simple but they are meant to wow you (eventually). Not there yet. The inspiration is @ggerganov's C++ from scratch implementation of OpenAI's Whisper speech model that finally convinced me something very special was going on. I'm a real visionary it appears. I'm hoping to form the *learning* counterpart to Georgi's excellent execution implementation.
Which is the training set accuracy and which is the test set accuracy? (I presume that the "87% correct" is the test set accuracy, but am not sure.)
@robryk voilà: