Open Source — MIT Licensed

Machine Learning for
Time-Series Prediction

DeepAlpha is an open-source Python toolkit for time-series prediction. LightGBM ensemble, SHAP feature selection, and walk-forward validation — all in one pip install.

$ pip install deepalpha-freqai 📋 Copied!

2,500+

Downloads

50+

Features

MIT

Licensed

3.9+

Python

Built for Data Scientists

Every component follows academic best practices. No shortcuts, no lookahead bias, no black boxes.

LightGBM Ensemble

Multi-model consensus system that combines multiple gradient-boosted trees for robust predictions with automatic hyperparameter tuning.

SHAP Feature Selection

Automatic feature pruning using SHAP values. Keeps only the features that actually matter, reducing noise and improving generalization.

Triple Barrier Labeling

Advanced labeling technique that creates well-defined classification targets using upper, lower, and time-expiry boundaries for superior signal quality.

Walk-Forward CV

Expanding-window cross-validation that respects temporal order. No data leakage, no lookahead bias — just honest evaluation.

Meta-Labeling

Secondary model that filters false positives from the primary model, dramatically improving precision on out-of-sample data.

Zero Heavy Dependencies

No GPU required, no TensorFlow, no PyTorch. Runs on any machine with Python and scikit-learn. Fast training on CPU.

Up and Running in 5 Lines

Simple API, powerful internals. Configure your model, fit your data, and get predictions with confidence scores.

$ pip install deepalpha-freqai 📋 Copied!
Read the Docs →
example.py
from deepalpha_freqai import DeepAlphaModel model = DeepAlphaModel(config={ "timeframe": "1h", "features": 50, "shap_selection": True, "walk_forward_folds": 5 }) # Train on your data model.fit(X_train, y_train) # Predict with confidence scores predictions = model.predict(X_test)

Three Steps to ML-Powered Predictions

From installation to inference in minutes, not days.

1

Install

One command, zero configuration headaches. Works on Linux, macOS, and Windows.

pip install deepalpha-freqai
2

Configure

Set your timeframe, feature count, and validation parameters. Sensible defaults included.

config = {"features": 50}
3

Predict

Get ML-powered predictions with confidence scores. Walk-forward validated, out-of-sample tested.

model.predict(X_test)

Validation Metrics on Time-Series Data

All results are walk-forward validated on out-of-sample data. No cherry-picked windows.

60–65%

Model accuracy on out-of-sample test sets using walk-forward validation

50+

Engineered features automatically ranked and pruned via SHAP importance

5-Fold

Walk-forward cross-validation folds to prevent overfitting and data leakage

Walk-forward validation prevents overfitting. Past model performance does not guarantee future results.

Start Building with DeepAlpha

Free, open-source, MIT licensed. Install in seconds.

$ pip install deepalpha-freqai 📋 Copied!

Want the Full Bot? Try It Free.

Get 7 days of the complete DeepAlpha bot with auto-retraining, multi-model ensemble, Telegram alerts, and risk management. No credit card needed.

Get your free license key
7-day full access No credit card All features included

Don't want to run the bot yourself? Copy our AI signals automatically.

Copy Trade on Bybit →