scikit-ptm-fs 0.10

A Python package for integrating scikit-learn feature selection with multi-label problem transformation methods.

Scikit-PTM-FS

Scikit-PTM-FS is a Python library that enables seamless integration of scikit-learn feature selection methods with multi-label classification via problem transformation methods (PTMs). The framework enables transformation-aware feature selection without requiring classification stages, allowing researchers to analyze the behavior of FS methods across various PTMs.

---

📦 Installation

pip install scikit-ptm-fs==0.10

---

🚀 Usage Example

Here is a basic example of how to apply a feature selection method within a problem transformation method using Scikit-PTM-FS:

from scikit_ptm_fs.problem_transformation import LabelPowerset
from sklearn.feature_selection import SelectKBest, f_classif

# Example feature and label matrices (replace with your own)
# X = ...  # shape (n_samples, n_features)
# y = ...  # shape (n_samples, n_labels)

# Step 1: Define a scikit-learn selector
selector = SelectKBest(score_func=f_classif, k=5)

# Step 2: Wrap it using the LabelPowerset PTM
lp_selector = LabelPowerset(selector=selector, require_dense=[False, True])

# Step 3: Apply feature selection
X_selected = lp_selector.fit_transform(X, y)

# Step 4: Get selected feature indices
selected_indices = lp_selector.get_support()
print("Selected features:", selected_indices)

💡 You can use other PTMs like BinaryRelevance, PPT, or PairwiseComparison similarly.

---

🧠 Supported Problem Transformation Methods (PTMs)

• Binary Relevance (BR) — One-vs-All decomposition.
• Label Powerset (LP) — Unique labelset encoding.
• Pairwise Comparison (PW) — One-vs-One label pair modeling.
• Pruned Problem Transformation (PPT) — Labelset pruning before transformation.
• Entropy-based Label Assignment (ELA) — Copy-based transformation for NLP.

---

🔗 Project Links

• 📦 PyPI: https://pypi.org/project/scikit-ptm-fs
• 🛠️ Source Code: https://github.com/Omaimah-AlHosni/scikit-ptm-fs

---

📄 License

This project is licensed under the MIT License.