flake

[pierreablin]

TODO list:

• merge MomentumNetTransform and MomentumNet, needs to only have one remaining class.
• add proper docstrings for MomentumNet with all arguments mentioned and explained, and the same example as in the doc/index.rst
• add a warning when gamma is too small
• change the 'skip_connection' argument to another term
• add more explanation about the parameters in the docstring
• add reference to the paper in the docstring
• replace 'attributes' to "parameters" in MomentumNet docstring
• fix userwarning in exact_rep_pytorch

In the doc:

• remove anaconda mention
• Explain a bit more the class MomentumNet in the index.rst file, and the example (mention that it is 10 layers, etc..)
• Say that you can transform a resnet in two lines of code
• find a way to remove all "methods" in the momentum net class in the doc

Examples:

• No savefig
• Simple classif example with resnet + MLP ? eg on digits dataset
• Example on two nested rings separation

Then, make a new release !

[michaelsdr]

@pierreablin is it ok to merge?

[pierreablin]

Still some work to do !

[michaelsdr]

What remains TODO list:

• add a warning when gamma is too small
• fix userwarning in exact_rep_pytorch In the doc:
• find a way to remove all "methods" in the momentum net class in the doc Examples:
• Simple classif example with resnet + MLP ? eg on digits dataset
• Example on two nested rings separation