pyrobustfs 0.1.1

A robust feature selection library using ensemble mRMR and optional refinement.

pyrobustfs

A robust feature selection library for Python, leveraging ensemble Minimum Redundancy Maximum Relevance (mRMR) with an optional refinement step. Designed for seamless integration into scikit-learn pipelines.

Features

• Ensemble mRMR: Improves robustness and stability of feature selection by running mRMR on bootstrapped/subsampled data and aggregating results.
• Scikit-learn Compatibility: Implements BaseEstimator and TransformerMixin for easy integration into scikit-learn pipelines, GridSearchCV, and RandomizedSearchCV.
• Flexible Refinement: Allows for an optional second stage of model-specific feature selection using any scikit-learn compatible estimator (e.g., RFE, SelectFromModel).
• Classification and Regression Support: Handles both classification (using mutual information for classification) and regression (using mutual information for regression) tasks.

Installation

Currently, pyrobustfs is not yet available on PyPI. You can install it directly from the source code:

1. Clone the repository:

   git clone https://github.com/yourusername/pyrobustfs.git
   cd pyrobustfs
   

2. Install in editable mode (for development) or standard mode:

   # For development (changes to code are immediately reflected)
   pip install -e .
   
   # For standard installation
   # pip install .
   

Usage

Basic Feature Selection

import pandas as pd
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from pyrobustfs.selectors import RobustMRMRSelector

# Generate synthetic classification data
X, y = make_classification(n_samples=1000, n_features=50, n_informative=10, n_redundant=5, random_state=42)
X = pd.DataFrame(X, columns=[f'feature_{i}' for i in range(X.shape[1])])

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)

# Initialize and fit the selector
# Select 5 features using 10 ensemble runs for a classification task
selector = RobustMRMRSelector(n_features_to_select=5, n_ensembles=10, classification=True, random_state=42)
selector.fit(X_train, y_train)

# Get the names of the selected features
selected_features = selector.get_feature_names_out()
print(f"Selected features: {selected_features}")

# Transform the data to keep only the selected features
X_train_selected = selector.transform(X_train)
X_test_selected = selector.transform(X_test)

print(f"Original X_train shape: {X_train.shape}")
print(f"Transformed X_train shape: {X_train_selected.shape}")

Using with a Refiner Estimator

You can provide an optional refiner_estimator for a second stage of feature selection. This is useful for model-specific refinement.

import pandas as pd
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.feature_selection import RFE # Recursive Feature Elimination
from pyrobustfs.selectors import RobustMRMRSelector

# Generate synthetic classification data
X, y = make_classification(n_samples=1000, n_features=50, n_informative=10, n_redundant=5, random_state=42)
X = pd.DataFrame(X, columns=[f'feature_{i}' for i in range(X.shape[1])])

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)

# Define a refiner estimator (e.g., RFE with Logistic Regression)
# The refiner will operate on the features pre-selected by the ensemble mRMR.
refiner = RFE(estimator=LogisticRegression(solver='liblinear', random_state=42), n_features_to_select=3)

# Initialize RobustMRMRSelector with the refiner
selector_with_refiner = RobustMRMRSelector(
    n_features_to_select=5, # This is the target for ensemble mRMR, refiner might override
    n_ensembles=10,
    refiner_estimator=refiner,
    classification=True,
    random_state=42
)

selector_with_refiner.fit(X_train, y_train)
selected_features_refiner = selector_with_refiner.get_feature_names_out()
print(f"Selected features (with refiner): {selected_features_refiner}")
print(f"Number of features selected by refiner: {len(selected_features_refiner)}")

# Transform data
X_train_refined = selector_with_refiner.transform(X_train)
print(f"Transformed X_train shape (with refiner): {X_train_refined.shape}")

Integrating into a Scikit-learn Pipeline

RobustMRMRSelector can be seamlessly integrated into a scikit-learn Pipeline.

import pandas as pd
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import Pipeline
from pyrobustfs.selectors import RobustMRMRSelector

# Generate synthetic data
X, y = make_classification(n_samples=1000, n_features=50, n_informative=10, n_redundant=5, random_state=42)
X = pd.DataFrame(X, columns=[f'feature_{i}' for i in range(X.shape[1])])

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)

# Create a pipeline with feature selection and a classifier
pipeline = Pipeline([
    ('feature_selection', RobustMRMRSelector(n_features_to_select=5, n_ensembles=10, classification=True, random_state=42)),
    ('classifier', LogisticRegression(solver='liblinear', random_state=42))
])

# Fit the pipeline
pipeline.fit(X_train, y_train)

# Evaluate the pipeline
accuracy = pipeline.score(X_test, y_test)
print(f"Pipeline accuracy: {accuracy:.4f}")

# Access selected features from the pipeline step
selected_features_pipeline = pipeline.named_steps['feature_selection'].get_feature_names_out()
print(f"Selected features from pipeline: {selected_features_pipeline}")

Development

To contribute or run tests, clone the repository and install in editable mode:

git clone https://github.com/yourusername/pyrobustfs.git
cd pyrobustfs
pip install -e .
pip install pytest
pytest

License

This project is licensed under the MIT License - see the LICENSE file for details.