Cool package for robust AI
Project description
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 theTrainer#train_model
function to keep reference to theWriter
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:
- put your source files into
src/<PACKAGE NAME>/
dir and create files:src/<PACKAGE NAME>/__init__.py
,setup.py
,pyproject.toml
,README.md
andLICENSE
(also optionally an emptytest
dir) - install required packages:
python3 -m pip install --upgrade twine build
- build the package, i.e. run from root folder:
python3 -m build
- 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.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for robbytorch-0.3.2-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 0cf0d625d2da52f1b65dd7b538c87b34ddab19f0fd7cd6e7a170e12c2d3104b8 |
|
MD5 | b947e9f8a0947815b975d59ca2871670 |
|
BLAKE2b-256 | 2d3694d7287309172fcd577c0636de1767a3eefa5f7184c8478dcb2b471a98df |