robbytorch 0.3.6

Cool package for robust AI

Installation

Robbytorch requires Pytorch, however it's not specified in the dependencies - we recommend installing Pytorch manually via conda and only later installing Robbytorch by pip. Pytorch has to be in version 1.6 or higher.

Use your conda env or create a new one:

conda create --name <ENV NAME> python=3.8 pip
conda activate <ENV NAME>

Install Pytorch. If you have older drivers for GPU you may require older version of CUDA, i.e.:

conda install pytorch torchvision torchaudio cudatoolkit=10.1 -c pytorch -c conda-forge

or even older Pytorch version:

conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.1 -c pytorch

Then run:

pip install robbytorch

Usage

The basics of the Robbytorch library are explained in the ipython/RobbytorchTutorial.ipynb juputer notebook.

TODO

• Streamline logging - currently we need to pass arguments to MLFLowWriter at init which we want to do outside of the Trainer#train_model function to keep reference to the Writer instance. We'd like to invent better flow that would allow us to pass config dict just once;
• Create helper for iterating over rows and cols and printing tensor:

diffs = show - show.roll(shifts=1, dims=0)

cols, rows = 3, 6
for i in range(rows):
    curr = (show[i*cols:(i+1)*cols])
    diff = vis_grad(diffs[i*cols:(i+1)*cols])
    robby.get_image_table(curr, diff, size=(10,10))

robby.widen_outputs()

Packaging

Development mode

Run from this repository's root dir:

python setup.py develop

Now you can use the package as if it was installed by pip.

When you’re done with a given development task, you can remove the project source from a staging area using:

python setup.py develop --uninstall

Creating python package

Use this tutorial. In short:

1. put your source files into src/<PACKAGE NAME>/ dir and create files: src/<PACKAGE NAME>/__init__.py, setup.py, pyproject.toml, README.md and LICENSE (also optionally an empty test dir)
2. install required packages:

python3 -m pip install --upgrade twine build

3. build the package, i.e. run from root folder:

python3 -m build

4. upload the package:

python3 -m twine upload dist/*

In order to perform the last step you need to have a https://pypi.org/ account and create the API token, which could be conveniently placed in $HOME/.pypirc:

[pypi]
  username = __token__
  password = pypi-<API TOKEN>

Warnings:

If you update the package you should increment the version param in setup.py and empty the dist dir.