coreset 0.0.1

A flexible framework for experimenting with and evaluating different sample selection strategies

Exemplar Sample Selection Framework

A flexible framework for experimenting with and evaluating different sample selection strategies. This framework allows you to:

• Compare different selection strategies
• Evaluate using multiple metrics
• Work with various datasets
• Extend with custom strategies and metrics

Installation

1. Clone the repository:

git clone https://github.com/yourusername/exemplar-sample-selection.git
cd exemplar-sample-selection

2. Install dependencies using Poetry:

# Install Poetry if you haven't already:
# curl -sSL https://install.python-poetry.org | python3 -

# Install dependencies and create virtual environment
poetry install

3. Activate the virtual environment:

poetry shell

Quick Start

Run the example experiment:

python examples/run_experiment.py

This will:

1. Load the IMDB dataset
2. Extract features using a sentence transformer
3. Run random selection (baseline strategy)
4. Evaluate using coverage metrics
5. Save results to outputs/imdb_random/

Framework Structure

exemplar-sample-selection/
├── src/
│   ├── data/              # Dataset handling
│   ├── selection/         # Selection strategies
│   ├── metrics/           # Evaluation metrics
│   ├── experiments/       # Experiment management
│   └── utils/             # Utilities
├── tests/                 # Unit tests
├── configs/               # Experiment configs
├── examples/              # Example scripts
└── docs/                  # Documentation

Core Components

1. Dataset Management

• Standardized dataset interface
• Built-in support for text datasets
• Feature extraction and caching
• Easy extension to other data types

2. Selection Strategies

• Base strategy interface
• Random selection baseline
• Support for both supervised and unsupervised selection
• Easy addition of new strategies

3. Evaluation Metrics

• Coverage metrics
• Distribution matching
• Performance metrics
• Extensible metric system

4. Experiment Management

• Configuration-based setup
• Automated logging
• Result tracking
• Reproducible experiments

Adding New Components

Adding a New Selection Strategy

1. Create a new file in src/selection/:

from .base import SelectionStrategy

class MyStrategy(SelectionStrategy):
    def select(self, features, labels=None, n_samples=100):
        # Implement your selection logic here
        return selected_indices

2. Register in src/selection/__init__.py

Adding a New Metric

1. Create a new file in src/metrics/:

from .base import Metric

class MyMetric(Metric):
    def compute(self, selected_features, full_features, 
                selected_labels=None, full_labels=None):
        # Implement your metric computation here
        return {'my_metric': value}

2. Register in src/metrics/__init__.py

Running Experiments

1. Create Configuration

from src.experiments import ExperimentConfig
from src.experiments.config import DatasetConfig, SelectionConfig, MetricConfig

config = ExperimentConfig(
    name="My Experiment",
    dataset=DatasetConfig(
        name="dataset_name",
        split="train"
    ),
    selection=SelectionConfig(
        name="strategy_name",
        params={"param1": value1},
        n_samples=1000
    ),
    metrics=[
        MetricConfig(
            name="metric_name",
            params={"param1": value1}
        )
    ]
)

2. Run Experiment

from src.experiments import ExperimentRunner

runner = ExperimentRunner(config)
results = runner.run()

3. Examine Results

Results are saved in the output directory:

• config.json: Experiment configuration
• results.json: Detailed results
• summary.txt: Human-readable summary
• experiment.log: Execution log

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

License

MIT License - see LICENSE file for details