shiffman/Neural-Network-p5

Performance - MNIST demo

Opened this issue · 2 comments

Right now it calculates a percentage completed since the sketch first begins. There's probably a better way to think about this.

The traditional way is to query all of testing data at once and calculate the accuracy, and do this every so often, lets say every 100 iterations...

Indeed, one reason why I didn't do this is that I would like the animation to run continuously at a relatively fast framerate. Running through all the data cannot be done reasonably on a per frame basis. However, I could only update the performance rating once it's made a full pass over the test data (and possibly increase to 2, 3, etc. data points per frame.)