search

mli / new-docs

https://beta.mxnet.io/
13 stars 17 forks source link

Updated Autograd Guide #42

Closed thomelane closed 5 years ago

thomelane commented 5 years ago

Added general introductions to why autograd is useful. Added diagrams and video animations.

thomelane commented 5 years ago

@NRauschmayr please review, thanks!

thomelane commented 5 years ago

Made adjustments as per feedback.

astonzhang commented 5 years ago

@thomelane @vishaalkapoor

Sorry for the delayed post-review comments. Overall, this tutorial is meaty and engaging---I like it!

Here are my minor suggestions for improvement. Referring to http://beta.mxnet.io/guide/packages/autograd/autograd.html

  1. (Applicable to other tutorials) Always add a short paragraph at the beginning of a tutorial to provide a high-level overview of the contents. In other words, add such a paragraph before "Why do we need to calculate gradients?" At the end of a tutorial, provide a summary.

  2. "Short Answer"/"Long Answer" should not be a title/subtitle. We can use bold text at the beginning of the paragraph.

  3. (Applicable to other tutorials) Italize latins, such as "e.g." and "i.e." and append a comma to them.

  4. Videos can't be played.

  5. Called the ‘forward pass’ of training. -> This is called ... Called the ‘backward pass’ of training. -> This is called ...

  6. Single quotes should be replaced with double quotes.

  7. Don't expose URL: https://mxnet.incubator.apache.org/api/python/ndarray/ndarray.html?highlight=dropout#mxnet.ndarray.Dropout

  8. I feel that most of "short answers" and "long answers" can be combined to be concise. We may trim short answers into an overview and put it at the beginning of the long answers.

  9. We have many "Advanced". How about putting some "Advanced" in basics, such as "Using Python control flow" and "Switching between training vs inference modes"?