Lentil - Latent Skill Embedding

A package for training, evaluation, and visualization of the Latent Skill Embedding model. Read more about the model at http://siddharth.io/lentil.

Usage

You can install the package's dependencies with

pip install -r requirements.txt

You can install the package in your environment with

python setup.py install

If you wish to run the tests, make sure you have tox installed and then run

tox

Once installed in your environment, command-line interfaces for training and evaluation are available through lse_train and lse_eval. The appropriate format for input interaction log data is given in the documentation for lentil.datatools.InteractionHistory. IPython notebooks used to conduct experiments are available in the nb directory, and provide example invocations of most functions and classes. It is recommended that you read the notebooks in the following order: toy_examples, synthetic_experiments, data_explorations, model_explorations, evaluations, sensitivity_analyses, and bubble_experiments.

Documentation

Build the documentation with

tox -e docs

Once run, open doc/_build/html/index.html for Sphinx documentation on modules in the package.

Questions and comments

Please contact the author at sgr45 [at] cornell [dot] edu if you have questions or find bugs.