scud 0.0.5

Structured Principal Component Analysis

Package for Statistical Components for Underlying Dimensions.

Installing

pip install scud

Documentation

The documentation can be found here.

Getting Started

import package and load data

from scud import PlnPCA, BinPCA
from scud.data import load_scrna

data_and_labels = load_scrna(return_labels = True)
counts = data_and_labels['endog']
labels = data_and_labels['labels']

Instantiate, fit the model and visualize the results for the Binomial PCA model

binpca = BinPCA(data = counts, latent_dimension = 5)
binpca.fit()
binpca.viz(colors = labels)

Instantiate, fit the model and visualize the results for the Poisson-log Normal PCA model

plnpca = PlnPCA(data = counts, latent_dimension = 5)
plnpca.fit()
plnpca.viz(colors = labels)

Supervised learning using labels

One hot encode the labels

from sklearn.preprocessing import OneHotEncoder
import numpy as np

labels = np.array(labels).reshape(-1, 1)
encoder = OneHotEncoder()
exog = encoder.fit_transform(labels).toarray()
additional_data = {"exog": exog}

Instantiate and fit the PlnPCA model

plnpca = PlnPCA(data = counts, latent_dimension = 5, additional_data = additional_data)
plnpca.fit()
plnpca.viz(colors = labels.reshape(-1))

Instantiate and fit the BinPCA model

binpca = PlnPCA(data = counts, latent_dimension = 5, additional_data = additional_data)
binpca.fit()
binpca.viz(colors = labels.reshape(-1))

CONTRIBUTING

You should run pre-commit install in the repo directory before commiting (if pre-commit is not installed, you can pip install it). This will make sure each python file is well formated and pylint will check the code before any python file is committed. You can check the .pre-commit-config.yaml file for more details on pylint configuration.

🛠 Installation

⚡️ Citations

Please cite our work:

Batardière, Bastien, Joon Kwon, Julien Chiquet, and Julien Stoehr (2024). “Importance sampling based gradient method for dimension reduction in Poisson Log-Normal model.” In: arXiv. url: https://arxiv.org/abs/2410.00476.