Published on

Zeal Machine Learning Docsets

Offline is quicker
Authors

Search Framework APIs online is Tedious

Solution is to download an offline reader : Not only can it be used without an internet connection, but the searching / redrawing is also much quicker.

As root you just need :

dnf install zeal

And then as a user on the command line:

zeal

(or use StartMenu-Develoment-Zeal in the window manager).

TensorFlow 1.4.0rc0 documentation

I found the helpful TensorFlow DocSet via (https://github.com/ppwwyyxx/dash-docset-tensorflow) :

# Download (with redirects enabled)
cd ~/.local/share/Zeal/Zeal/docsets/
curl -L -O https://github.com/ppwwyyxx/dash-docset-tensorflow/releases/download/1.4.0rc0/TensorFlow-1.4.0.tgz
tar -xzf TensorFlow-1.4.0.tgz

This DocSet works as expected, and is a huge help when dabbling with TensorFlow code.

PyTorch documentation

The PyTorch documentation is good, except that the search functionality online is very frustrating, since the individual function search index points to a link that requires the page to redraw several times, causing the display to leap around.

I found a somewhat helpful PyTorch DocSet via (https://github.com/iamaziz/PyTorch-docset) :

# Download (with redirects enabled)
cd ~/.local/share/Zeal/Zeal/docsets/
curl -L -O https://github.com/iamaziz/PyTorch-docset/raw/master/PyTorch.tgz
tar -xzf PyTorch.tgz

One problem is that the DocSet creator doesn't index the individual functions properly. However, if you pull up the main PyTorch page, the original search functions are there within the pane - and render much quicker than using the website online. But still... it's not optimal.

Better PyTorch documentation

Since the original documentation for PyTorch is built via Sphinx, there's a much cleaner way of creating the documentation : use doc2dash within the sphinx build of the PyTorch documentation itself:

git clone https://github.com/pytorch/pytorch.git  # Get the full repo down
cd pytorch/docs/

# Move to the branch of the code that matches your installed version of PyTorch
# this ensures that the docs you get are relevant (PyTorch moves very quickly...)
git checkout v0.2.0

. ~/env3/bin/activate # Or whatever you are using for a VirtualEnv

pip install -r requirements.txt
pip install doc2dash

make docset

cp -r PyTorch.docset ~/.local/share/Zeal/Zeal/docsets/

Once this is confirmed to create decent documents (worked 2nd time for me : once the branch version matched the installed version), you can safely remove the repo.

All done.