chinese_restaurant_process

chinese_restaurant_process is a Python package that provides an easy to use interface for simulating the Chinese Restaurant Process (CRP), a popular model in Bayesian nonparametrics.

Installation

---

Installation Methods

You can install Chinese Restaurant Process using pip:

pip install chinese_restaurant_process

You can install the latest version directly from the GitHub repository using pip:

pip install git+https://github.com/jhaberbe/chinese_restaurant_process

Or, if you want to install the package from source, you can use:

git clone https://github.com/jhaberbe/chinese_restaurant_process.git
pip install .

Usage

---

Example Use Cases

To perform the initial inference of classes:

import numpy as np
from crp.process import ChineseRestaurantProcess

# Your data, (n_samples, n_features)
X = np.random.randint(1, 100, size=(1000, 10))

# Run inference on train data.
crp = ChineseRestaurantProcess(X, expected_number_of_classes=1)
crp.run(epochs=1)

After training, you can predict the class of new data points:

# Your data, (n_samples, n_features)
X_new = np.random.randint(1, 100, size=(1000, 10))

# Setting min_membership = 0.01 is recommended usually.
# Since this is random data, we set it to 0
labels = crp.predict(X_new, min_membership=0.0)

Documentation

We have a few tutorials in the notebook/ folder. They go over basic usage, and also try to explain how inference is going to be performed.

Recommended reading order is:

• Usage.ipynb
• What-The-Heck-Is-Collapsed-Gibbs-Sampling.ipynb
• Explaining-Class-Structure.ipynb

Contributing

---

We welcome contributions. If you'd like to contribute, please follow these steps:

1. Fork the repository on GitHub.
2. Create a new branch for your changes.
3. Make the changes to your branch.
4. Commit your changes with a meaningful commit message.
5. Create a pull request against the main branch.

Right now, work is being done to add the following features:

• Improve sampling of inital hyperparameters, right now the hyperparameters are sampled are fixed, and it works well for most cases, but it would be nice to have a more robust sampling method.
• Infinitely Nested Chinese Restaurant Process, as described by Blei et al. (2010). Some work is already done, but it is not yet ready for use (notebook/In Progress/Nested-Chinese-Restaurant-Process.ipynb).
• [1/2] More distributions (Gaussian, Gamma, etc. etc.)
• Plotting utilities to visualize the class structure of the CRP.
• Utilities to extract out which features are most important for each class.
• Testing (I've never done testing before, so this will be a learning experience for me).

License

---

Chinese Restaurant Process is released under the GPL v3 license. See the LICENSE file for more information.