The colors used in the network illustration are particularly bright. Consider using a tool like NN-SVG (https://alexlenail.me/NN-SVG/) to render similar network illustrations.
Typo in the Scikit-Learn section of "Neural Networks with Popular Pythong Libraries"; 'nerual' instead of 'neural' in the block of text above the code example.
03_solving_differential_equations_neural_networks
None
04_further_problems
None
Overall comments:
This module is very thorough and well thought-out. Despite having used deep learning professionally for years, I found that there were things I learned from this module. Aside from the notebook-level comments above, my only overall recommendation would be to consider limiting the exercises in the module to JAX and Keras. Scikit's deep learning API is very lean and unlikely to be useful in practice (consisting almost entirely of "multi-layer perceptron" (MLP)-type implementations), whereas Keras/TF is the current standard with JAX being a very interesting up-and-coming library.
Hi @juliebutlerhartley - Great module! A few comments from my end:
Can you please move all of your files gitignore and LICENSE) into a folder and give it a good name (something line "Solving_Differential_Equations_with_NNs")?
Your notebooks are missing the colab link. Could you please add that on the top? Here is an example from Karan.
Please let me know once your changes are complete so I can merge. Thank you!
Julie, great module - here are my comments:
Comments specific to individual components:
01_differential_equations
02_neural_networks
03_solving_differential_equations_neural_networks
04_further_problems
Overall comments:
This module is very thorough and well thought-out. Despite having used deep learning professionally for years, I found that there were things I learned from this module. Aside from the notebook-level comments above, my only overall recommendation would be to consider limiting the exercises in the module to JAX and Keras. Scikit's deep learning API is very lean and unlikely to be useful in practice (consisting almost entirely of "multi-layer perceptron" (MLP)-type implementations), whereas Keras/TF is the current standard with JAX being a very interesting up-and-coming library.