mlt-experiments 0.1.0

Store, manage, and compare machine learning experiments

MLT Experiments Library

A Python library for managing machine learning experiments. Store, retrieve, compare, and analyze your ML experiments effortlessly.

Features

✨ Easy Experiment Tracking

• Save experiment configurations, hyperparameters, models, and results
• Automatic timestamps and metadata management

📊 Experiment Comparison

• Compare multiple experiments side-by-side
• Find best experiments by metric
• Export to CSV for analysis

🏷️ Flexible Organization

• Tag experiments for easy filtering
• List experiments by tag
• Full experiment history

📁 Multiple Data Formats

• JSON export/import
• CSV export for spreadsheets
• Pandas DataFrame integration

Installation

From PyPI (once published)

pip install mlt-experiments

From source

git clone https://github.com/yourusername/mlt-experiments.git
cd mlt-experiments
pip install -e .

Quick Start

from mlt_experiments import ExperimentTracker

# Initialize tracker
tracker = ExperimentTracker(data_dir="my_experiments")

# Save an experiment
tracker.save(
    experiment_name="baseline_v1",
    config={
        "model": "RandomForest",
        "learning_rate": 0.01,
        "batch_size": 32,
        "epochs": 100
    },
    results={
        "accuracy": 0.92,
        "loss": 0.23,
        "precision": 0.89,
        "recall": 0.91
    },
    tags=["baseline", "v1"]
)

# Load an experiment
exp = tracker.load("baseline_v1")
print(exp)

# List all experiments
experiments = tracker.list_all()
print(experiments)

# Compare experiments
comparison = tracker.compare(["baseline_v1", "baseline_v2"], metric="accuracy")
print(comparison)

# Get best experiment by metric
best = tracker.get_best(metric="accuracy", mode="max")
print(f"Best: {best['name']}")

# Export to CSV
tracker.export_to_csv("experiments_summary.csv")

API Reference

ExperimentTracker

__init__(data_dir="experiments_data")

Initialize tracker with storage directory.

save(experiment_name, config, results, model_code="", tags=None)

Save experiment with metadata.

load(experiment_name)

Load specific experiment.

list_all()

Get list of all experiments.

list_by_tag(tag)

Get experiments with specific tag.

compare(experiment_names, metric=None)

Compare multiple experiments.

get_best(metric, mode="max")

Get best experiment by metric.

export_to_csv(output_file)

Export all experiments to CSV.

delete(experiment_name)

Delete an experiment.

Use Cases

• Research: Track different model architectures and hyperparameters
• Production: Compare model versions and performance
• Team Collaboration: Share experiments across team members
• Exam/Assignment: Store all lab work in organized, reproducible format

License

MIT License - see LICENSE file for details

Contributing

Contributions welcome! Please submit pull requests or issues.