x-trend-architecture

What problem does it solve?

Provides a complete blueprint for implementing the X-Trend architecture, enabling practitioners to build cross-attentive time-series models for trend-following, forecasting, and interpretable trading signals.

Core Features & Use Cases

• Input features: 8-dimensional vectors combining normalized returns and MACD indicators for robust temporal modeling.
• VSN & embeddings: Feature selection networks and asset embeddings to handle multi-asset regimes and few-shot learning.
• Attention & encoder-decoder: Self-attention over context, cross-attention between target and context, plus an encoder-decoder structure for prediction and position forecasting.
• Use Case: Build an end-to-end trading model that predicts positions and forecasts distributions with interpretable attention patterns.

Quick Start

Initialize the X-Trend model and begin training using episodic learning as described in TRAINING.md.