dataclr is a Python library for feature selection, enabling data scientists and ML engineers to identify optimal features from tabular datasets. By combining filter and wrapper methods, it achieves state-of-the-art results, enhancing model performance and simplifying feature engineering.

• Comprehensive Methods:

  • Filter Methods: Statistical and data-driven approaches like ANOVA, MutualInformation, and VarianceThreshold.

    Method	Regression	Classification
    ANOVA	Yes	Yes
    Chi2	No	Yes
    CumulativeDistributionFunction	Yes	Yes
    CohensD	No	Yes
    CramersV	No	Yes
    DistanceCorrelation	Yes	Yes
    Entropy	Yes	Yes
    KendallCorrelation	Yes	Yes
    Kurtosis	Yes	Yes
    LinearCorrelation	Yes	Yes
    MaximalInformationCoefficient	Yes	Yes
    MeanAbsoluteDeviation	Yes	Yes
    mRMR	Yes	Yes
    MutualInformation	Yes	Yes
    Skewness	Yes	Yes
    SpearmanCorrelation	Yes	Yes
    VarianceThreshold	Yes	Yes
    VarianceInflationFactor	Yes	Yes
    ZScore	Yes	Yes

  • Wrapper Methods: Model-based iterative methods like BorutaMethod, ShapMethod, and OptunaMethod.

    Method	Regression	Classification
    BorutaMethod	Yes	Yes
    HyperoptMethod	Yes	Yes
    OptunaMethod	Yes	Yes
    ShapMethod	Yes	Yes
    Recursive Feature Elimination	Yes	Yes
    Recursive Feature Addition	Yes	Yes

• Flexible and Scalable:

  • Supports both regression and classification tasks.
  • Handles high-dimensional datasets efficiently.

• Interpretable Results:

  • Provides ranked feature lists with detailed importance scores.
  • Shows used methods along with their parameters.

• Seamless Integration:

  • Works with popular Python libraries like pandas and scikit-learn.

Install dataclr using pip:

pip install dataclr

Prepare your dataset as pandas DataFrames or Series and preprocess it (e.g., encode categorical features and normalize numerical values):

import pandas as pd
from sklearn.preprocessing import StandardScaler

# Example dataset
X = pd.DataFrame({...})  # Replace with your feature matrix
y = pd.Series([...])     # Replace with your target variable

# Preprocessing
X_encoded = pd.get_dummies(X)  # Encode categorical features
scaler = StandardScaler()
X_normalized = pd.DataFrame(scaler.fit_transform(X_encoded), columns=X_encoded.columns)

The FeatureSelector is a high-level API that combines multiple methods to select the best feature subsets:

from sklearn.ensemble import RandomForestClassifier
from dataclr.feature_selection import FeatureSelector

# Define a scikit-learn model
my_model = RandomForestClassifier(n_estimators=100, random_state=42)

# Initialize the FeatureSelector
selector = FeatureSelector(
    model=my_model,
    metric="accuracy",
    X_train=X_train,
    X_test=X_test,
    y_train=y_train,
    y_test=y_test,
)

# Perform feature selection
selected_features = selector.select_features(n_results=5)
print(selected_features)

For granular control, you can use individual feature selection methods:

from sklearn.linear_model import LogisticRegression
from dataclr.methods import MutualInformation

# Define a scikit-learn model
my_model = LogisticRegression(solver="liblinear", max_iter=1000)

# Initialize a method
method = MutualInformation(model=my_model, metric="accuracy")

# Fit and transform
results = method.fit_transform(X_train, X_test, y_train, y_test)
print(results)

As our algorithm produces multiple results, we selected benchmark results that balance feature count with performance, while being capable of achieving the best performance if needed.

Explore the full documentation for detailed usage instructions, API references, and examples.