Photon — ML Framework

Open-source machine learning framework supporting the full ML lifecycle, from data preparation through training, monitoring, evaluation, deployment, and serving.

Photon extends TensorFlow, Keras, and PyTorch for deep learning on financial time series with a custom object-oriented API: built-in subclassing of Keras/TensorFlow APIs plus custom models, layers, optimizers, and loss functions, each behind a highly customizable interface for extending them to specific algorithms and networks. The framework is model/algorithm agnostic and container-native.

Key capabilities:

• Native TensorFlow distributed-strategy support with shared inputs/outputs across networks for deep ensemble learning and dynamic learning-rate scheduling
• Real-time preprocessing for dataset splitting, normalization, scaling, aggregation, and time-series resampling, with custom batching, padding, and masking
• Simple interface for saving, serializing, and loading entire networks, including learned and hyperparameters
• Detailed parameter logging throughout the pipeline for interpretability and reproducibility