rmst / chi Goto Github PK

Chi provides high-level operations for writing and visualizing experiments

Experiment Scripts can also be defined via decorator. When run from the command line the function arguments are translated into command line parameters.

@chi.experiment
def my_experiment(logdir, a=0.5):
  print(logdir)
  ...

If no logdir is specified it will generate a new one. See examples/experiments.py for the full example

For a more interesting experiment see the Wasserstein GAN example.

Chiboard is a browser based dashboard for managing experiments. Start it with chiboard.

Clicking on an experiment card leads to a detail page about that experiment which automatically spins up and embeds a TensorBoard:

See chi/board for a full overview.

Requires Python 3.6

git clone [email protected]:rmst/chi.git
pip install -e chi

The foundations for this work have been developed during projects at the following institutes.

• Reasoning and Learning Lab (RLLab) at McGill University in Canada
• Montreal Institute for Learning Algorithms (MILA) in Canada
• Intelligent Autonomous Systems Lab (IAS) at TU-Darmstadt in Germany

The structures of this repo and readme were inspired by Keras-RL and Keras respectively.