zxul767 / fastai-annotated Goto Github PK

In research notebook for chapter 4, an ensemble of binary classifiers (for the MNIST classification problem) is created which achieves ~90% accuracy on the validation set.

However, the strategy used to train the individual classifiers uses a general random sampling, which makes it perfectly legitimate that some of the samples will contain more classes of a certain digit.

Would the validation accuracy rise if we sampled the corresponding proportion from each digit class instead?

This is a common practice in fastai notebooks and while it is probably harmless in the context of the course, I think it's a bad practice in general due to the potential for silent shadowing of function and variable names.

We should rewrite the notebooks to use the more commonly used practice of using a very short but common prefix. pandas uses pd, numpy uses np, seaborn uses sns, etc.

Some simple issues I found after a quick review:

• Typos
• Some paragraphs can be streamlined
• Some images are too large
• Some images seem unnecessary (e.g., the huge one on human inspection)

Some of the answers in the last section could use some correction in style and grammar.