Skip to content

Latest commit

 

History

History
66 lines (54 loc) · 2.49 KB

CONTRIBUTING.md

File metadata and controls

66 lines (54 loc) · 2.49 KB
Raw

How to contribute

TreeCat is designed to be a fun research platform both for Bayesian algorithms for multivariate multinomial inference, and for practical data analysis. We welcome issues and bug reports on GitHub. Before you submit a pull request, please set up an environment and test as described below.

Setting up a conda env

To install TreeCat for development, clone and install in a fresh conda env using miniconda or Anaconda:

git clone git@github.com:posterior/treecat
cd treecat
conda env create -f environment.3.yml    # Or environment.2.yml for Python 2.7.
source activate treecat3
pip install -e .

Some changes are difficult to get working in both Python 2 and 3. We recommend setting up a pair of conda envs when testing unicode support as in the treecat.format module.

Testing

We use standard tools for testing, linting, and formatting during development. You can run these easily using the Makefile

make lint     # Check for syntax errors using flake8.
make format   # Automatically reformat code using isort and yapf.
make test     # Run tests with py.test.

Please make sure you make format and make test before submitting a pull request.

Coding Conventions

We use the following variable name conventions for often-used variables:

  • N = number of rows in a dataset. This is the n of statisticians.
  • V = number of features = number of vertices in a latent tree. This is the p of statisticians.
  • E = V - 1 = number of edges in a latent tree.
  • K = V * (V - 1) // 2 = number of edges in the complete undirected graph on V vertices.
  • M = config['model_num_clusters'] = number of latent classes.
  • R = number of columns in any ragged array in TreeCat's internal format of multinomial data. This is the sum of the number of categories of all observed features.
  • ragged_index is a V+1-long numpy array of offsets into ragged arrays.
  • config is a global configuration dict, as defined in config.py.
  • probs is any array of competing probabilities, often non-normalized and interpreted as odds.
  • logits is any array of competing log-probabilities, often non-normalized and interpreted as log-odds.
  • data is either an [N, V]-shaped array of data in external dense format or an [N, R]-shaped array of data in internal ragged format. This is "data" in the sense of statisticians.