🚀 ez-automl-lite

A lightweight, serverless-optimized AutoML library for Python. Build, evaluate, and export high-performance machine learning models with just 3 lines of code.

✨ Features

• 3-Line API: Designed for simplicity and speed across 5 different ML tasks.
• Serverless-First: Optimized for AWS Lambda/Azure Functions and low-memory environments.
• Premium Reports: Professional HTML/CSS reports for all tasks (No external JS or Internet required).
• Comprehensive Analytics: Supports Regression, Classification, Clustering, Anomaly Detection, and Time Series Forecasting.
• ONNX Export: One-click export for cross-platform deployment.

📦 Installation

# Full installation (recommended)
pip install "ez-automl-lite[all]"

# Or install with specific optional dependencies:
pip install "ez-automl-lite[onnx]"       # ONNX export support
pip install "ez-automl-lite[reports]"    # Enhanced EDA reports
pip install "ez-automl-lite[cluster]"    # DBSCAN automatic eps selection
pip install "ez-automl-lite[timeseries]" # Time series forecasting (ARIMA/Prophet)

🚀 The 5 Core Modules

1. Regression

Automated training with residual analysis and error diagnostics.

from ez_automl_lite import AutoML
aml = AutoML(target="target").fit(df)
aml.report("regression_report.html")

2. Classification

Visual Confusion Matrices and detailed class-wise performance metrics.

from ez_automl_lite import AutoML
aml = AutoML(target="label").fit(df)
aml.report("classification_report.html")

3. Clustering (Unsupervised)

Automated optimal K-search using Silhouette and Calinski-Harabasz scores. NEW: Automatic DBSCAN eps parameter selection via k-distance elbow detection.

from ez_automl_lite import AutoCluster
ac = AutoCluster(max_clusters=8).fit(df)
ac.report("clustering_report.html")

4. Anomaly Detection

Profile-based detection using Isolation Forest with detailed sample analysis.

from ez_automl_lite import AutoAnomaly
aa = AutoAnomaly(contamination=0.05).fit(df)
aa.report("anomaly_report.html")

5. Time Series Forecasting

Automated forecasting with ARIMA/SARIMA and Prophet (1.2.1+), including decomposition, stationarity analysis, and 95% confidence intervals.

from ez_automl_lite import AutoTimeSeries
ats = AutoTimeSeries(
    time_column="date",
    target_column="sales",
    forecast_horizon=30,
    scaling="absmax"  # Optional: 'absmax' or 'minmax'
).fit(df)
forecast = ats.predict(30)
ats.report("timeseries_report.html")

📂 Examples & Scripts

Check the examples/ directory for full implementation scripts:

• examples/regression_example.py
• examples/classification_example.py
• examples/clustering_example.py
• examples/anomaly_example.py
• examples/timeseries_example.py
• examples/timeseries_advanced_example.py (demonstrates scaling and holidays_mode options)

🛠️ Performance & Export

• ONNX Export: Cross-platform models in one line: aml.export_onnx("model.onnx").
• EDA: Generate pre-training analysis: aml.eda(df, "eda.html").
• UUIDs: Every training session generates a unique ID for easy tracking.

🗺️ Roadmap

• Core Package Refactor
• Premium CSS-only Reports
• AutoCluster & AutoAnomaly implementation
• Cross-platform ONNX support
• PyPI Automated Release Workflow

🤝 Contributing & License

Created by Cristopher Coronado. Distributed under the MIT License.