cosmicml 0.1.0

From raw data to model insights — for Earth and beyond.

🌌 cosmicml

From raw data to model insights — for Earth and beyond.

A Python toolkit for ML practitioners and space data enthusiasts. cosmicml handles the full pipeline — data loading, preprocessing, model benchmarking, and SHAP explainability — in clean, importable modules.

Built by @Deepali-07 — ML Engineer & Astrophysics enthusiast.

---

✨ Features

Module	What it does
DataLoader	Load CSV, JSON, HDF5, and FITS (astronomy) files
DataSplitter	Stratified train/val/test splitting
DataCleaner	Imputation, outlier clipping, label encoding
SmartScaler	Standard/MinMax/Robust scaling with .revert()
DataBalancer	SMOTE, ADASYN, undersampling, SMOTEENN
ModelBenchmarker	Run N models → ranked comparison table
HyperparamTuner	GridSearch / RandomSearch wrapper
SHAPExplainer	One-line SHAP summary, beeswarm, waterfall
ModelReporter	Auto-generate clean model performance report
timer	Decorator to time any function
TimeIt	Context manager for timing code blocks

---

🚀 Installation

pip install cosmicml

# With all optional extras
pip install cosmicml[all]

# Astronomy FITS support only
pip install cosmicml[astronomy]

---

⚡ Quick Start

from cosmicml import (
    DataCleaner, SmartScaler, DataBalancer,
    ModelBenchmarker, SHAPExplainer, ModelReporter
)
from cosmicml.data.splitter import DataSplitter

# 1. Split
splitter = DataSplitter(test_size=0.2, val_size=0.1, stratify=True)
X_train, X_val, X_test, y_train, y_val, y_test = splitter.split(X, y)

# 2. Clean
cleaner = DataCleaner(strategy="median", outlier_method="iqr")
X_train = cleaner.fit_transform(X_train)
X_test  = cleaner.transform(X_test)

# 3. Scale
scaler = SmartScaler(method="standard")
X_train = scaler.fit_transform(X_train)
X_test  = scaler.transform(X_test)

# 4. Balance
balancer = DataBalancer(strategy="smote")
X_train, y_train = balancer.fit_resample(X_train, y_train)

# 5. Benchmark
bench = ModelBenchmarker(task="classification")
print(bench.run(X_train, y_train, X_test, y_test))

# 6. Explain
explainer = SHAPExplainer(bench.best_model_, X_train)
explainer.summary(X_test)

# 7. Report
reporter = ModelReporter(bench.best_model_, task="classification")
reporter.report(X_test, y_test)

---

🔭 Astronomy / FITS Support

from cosmicml import DataLoader

loader = DataLoader("observations.fits")
df = loader.load()  # Returns a clean pandas DataFrame

---

📁 Project Structure

cosmicml/
├── data/           # DataLoader, DataSplitter
├── preprocess/     # DataCleaner, SmartScaler, DataBalancer
├── training/       # ModelBenchmarker, HyperparamTuner
├── explainability/ # SHAPExplainer, ModelReporter
└── utils/          # logger, timer

---

🤝 Contributing

Pull requests are welcome! Please open an issue first to discuss what you'd like to change.

git clone https://github.com/Deepali-07/cosmicml
cd cosmicml
pip install -e ".[dev]"
pytest tests/

---

📄 License

MIT © Deepali