gschramm
commented
3 years ago Thanks for all the feedback! I included all your comments and also added the automatic data download in the 00_*.ipynb notebook since the data is now available on Zenodo.
Could you run a final test of all notebooks including the data download?
AnderBiguri
Hi!
Just to be clear, the notebooks are awesome, definitely really good, so this is more of a nitpick than a serious criticism, so feel free to ignore it.
01_tf_data.ipynbThe reason of these data indices is not clear in the code/test, may be confusing?x[3*i + sim,...,0] = data[0][40:-40,30:-30,40:-40]The exercise says "Modify
train_data" but there is no such function/variable astrain_data.02_tf_models.ipynbYou will inevitably get the question "why MSE, why Adam, why 1e-3". One option is to write something in the notebook (albeit I know the answer is probably "we tried it and it works"), the other is just to be ready for the question :).The section that does
# generate artificial training and validation datamay use some code-comments, but admittedly, with a quick look I can infer what everything is doing. Maybe its good as is, so people scratch their head a bit.That's all! Let me repeat that this notebooks are awesome. I can only be described as someone with a mild interest on NN that only programmed a fully contented perception network 10 years ago and I could follow this tutorial (with some google searches) quite well.