Unified, centralized and simplified code; added CPU option

[mbuet2ner]

Here are some things I added/ changed:

• dedicated filtervisualizer.py file
• dedicated plot_reconstructions.py file
• easier accessible due to streamlined functions. Easier to use other neural nets, without changing too much code.
• added CPU option
• removed unused imports, codes, parameters
• don't overwrite built in filter function
• pep8 guidelines
• f strings where applicable
• removed imgur code (I am not sure if it is really necessary for the task)

Things I have not changed yet, but will do in the future:

• vgg16 and resnet34 part 2 notebooks

Thus, it should be much easier to change things in the future/ migrate to new fastai version.

[review-notebook-app[bot]]

Check out this pull request on 

You'll be able to see Jupyter notebook diff and discuss changes. Powered by ReviewNB.

[mbuet2ner]

@fg91 sorry for the mess. I know it is hard to review, because it combines quite a lot of changes in one PR. Should I make several pull requests out of it?

[fg91]

Hey, sorry it took me so long to react. First of all, thank you for taking the time to improve this tutorial!

Would you be willing to change the following?

• The experiments folder contains additional notebooks provided by a reader of the blog post, so I would like to keep the FilterVisualizer separated from it. Could you please move the filtervisualizer.py out of the experiments folder? The plot_reconstructions.py can stay in the experiments folder as it is only used for the additional experiments...
• Would you be willing to simplify the Calculate_mean_activation_per_filter_in_specific_layer_given_an_image.ipynb using the most_activated method from FiterVisualizer?

Thanks!

Fabio

[mbuet2ner]

I finally addressed your proposed changes :tada: :

• filtervisualizer.py is now in root. See 12ebf75
• Calculate_mean_activation_per_filter_in_specific_layer_given_an_image.ipynb now uses filtervisualizer.py. See 9b4b794
• train_tfms and val_tfms are now instance attributes. See 84ea7cb
• FilterVisualizer.most_activated() method does not print any mean_act at all (I do not think it is necessary?). Especially no activations.features[0, 5] :grin: . See 9b4b794

[mbuet2ner]

@fg91 If there is anything else you want me to change, just let me know.