moose-fs 0.1.0

MOOSE-FS: Multi-Objective Optimized Ensemble Feature Selection

MOOSE-FS

Documentation

Overview

MOOSE-FS is a feature selection library that leverages an ensemble-based approach to optimize both predictive performance and stability. By combining multiple feature selection methods, merging strategies, and evaluation metrics, it provides a highly flexible and tunable pipeline for both classification and regression tasks. The package automates feature selection across multiple iterations and uses Pareto optimization to identify the best feature subsets.

Users can define their feature selection process by:

• Selecting feature selection methods from predefined options or implementing custom ones.
• Choosing merging strategies to aggregate feature rankings.
• Specifying performance metrics to evaluate selected features.
• Configuring the number of features to select and the number of repetitions.
• Working with either classification or regression problems.

The library allows defining feature selectors, merging strategies, and metrics either as class instances or as string identifiers, which act as placeholders for built-in methods. The framework is modular and can be easily extended by adding new selection algorithms or merging strategies.

---

Requirements

• Python 3.9 or higher
• Dependencies: Automatically installed from pyproject.toml.

---

Installation

From Source

To install the package from source, run:

pip install git+https://github.com/CI4CB-lab/moosefs.git

Alternatively, clone the repository and install locally:

git clone https://github.com/CI4CB-lab/moosefs.git
cd moosefs
pip install .

---

Using the Library

1. Feature Selection Pipeline

The core of MOOSE-FS is the FeatureSelectionPipeline, which provides a fully configurable workflow for feature selection. Users can specify:

• Feature selection methods
• Merging strategy
• Evaluation metrics
• Task type (classification or regression)
• Number of features to select
• Number of repetitions

Example Usage

# `data` can be a single DataFrame (last column = target)
# or you can pass `X` and `y` separately.
# Assume `data` is a pandas DataFrame whose last column "label" holds the targets.
from moosefs import FeatureSelectionPipeline

fs_methods = ["f_statistic_selector", "random_forest_selector", "svm_selector"]
merging_strategy = "union_of_intersections_merger"

pipeline = FeatureSelectionPipeline(
    X=data.drop(columns=["label"]),
    y=data["label"],
    fs_methods=fs_methods,
    merging_strategy=merging_strategy,
    num_repeats=5,
    task="classification",
    num_features_to_select=10,
)
results = pipeline.run()

This will run feature selection, merge results using the chosen strategy, and return the best-selected features.

2. Extensibility

MOOSE-FS is designed to be easily extended. Users can implement custom:

• Feature selection methods: Define a new feature selector class and integrate it into the pipeline.
• Merging strategies: Implement a custom strategy to aggregate selected features.
• Metrics: Add new evaluation metrics tailored to specific tasks.

New methods can be used directly in the pipeline by passing the class or a corresponding identifier.

---

Using the CLI

Once installed, the pipeline can also be run from the command line using:

efs-pipeline

This command executes scripts/main.py using parameters from scripts/config.yaml. Users can specify a different config file:

efs-pipeline path/to/your_config.yaml

Example config.yaml

experiment:
  name: "example_experiment"
  results_path: "results/"
  data_path: "data/input_data.csv"

preprocessing:
  normalize: true
  handle_missing: true

pipeline:
  fs_methods: ["f_statistic_selector", "random_forest_selector"]
  merging_strategy: "union_of_intersections_merger"
  num_repeats: 5
  task: "classification"
  num_features_to_select: 10

Results

The results are saved in a structured directory under results/example_experiment/, including:

• A text file summarizing the pipeline run.
• A CSV file containing the final results.

---

Code Structure

• core/: Core modules for data processing, metrics, and stability computation.
• feature_selection_pipeline.py: Defines the main feature selection workflow.
• feature_selectors/: Implements feature selection methods (e.g., F-statistic, mutual information, RandomForest, SVM).
• merging_strategies/: Implements merging strategies such as Borda count and union of intersections.

---

Contributing

Contributions are welcome! If you have ideas for improving MOOSE-FS, feel free to open an issue or submit a pull request.

Development (uv)

This project uses uv for local environments and dependency management. The library builds via the existing PEP 517 backend (hatchling); uv only manages the environment, installs, and command execution.

• Install/select Python 3.9+ and ensure uv is installed.
• Create a local virtual environment in .venv:

uv venv --python 3.9

• Install dev dependencies (editable):

uv pip install -e ".[dev]"

• Install pre-commit hooks:

uv run pre-commit install

• Run formatting and linting:

uv run ruff format .
uv run ruff check --fix .

• Run tests:

uv run pytest -q

---

License

This project is licensed under the MIT License.