FreqAI Architecture and Model Data Lifecycle

Summary

The document explains FreqAI’s main software components and how they support model development. A persistent model object handles data collection, feature engineering, training, and inference; a per-asset data kitchen provides processing tools and metadata; and a persistent data drawer stores predictions and supports saving and reloading. Model classes can override core training, prediction, and data-cleaning functions.

It also describes the disk layout used to preserve configurations, historical predictions, model files, metadata, and optional preprocessing artifacts. Backups of prediction history support recovery after corruption, while saved training data and dates enable post-processing and dissimilarity calculations. The examples illustrate multiple model runs and assets, but provide no trading results or model-quality evidence. The guidance is chiefly about system design and operational resilience; it cautions that the data kitchen depends on the generated file structure, which should not be manually altered.

Key ideas

• FreqAI separates model logic, per-asset processing, and persistent storage into three main objects.
• Model implementations can customize training, prediction, and data-cleaning behavior.
• The generated directory structure stores configurations, predictions, models, metadata, and optional transforms.
• Prediction-history backups and automatic reloads are designed to improve recovery after crashes or file corruption.
• Saved training features and dates support post-processing and dissimilarity analysis.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.