Skip to content

Latest commit

 

History

History
70 lines (59 loc) · 4.36 KB

File metadata and controls

70 lines (59 loc) · 4.36 KB

Repository Map

This page gives a quick path through the repository for contributors and power users.

Top-level areas

Path Purpose
README.md GitHub landing page
docs/.vitepress/config.js Navigation and site structure for the /docs/ site
web/ Static landing page assets for the published site root
docs/ VitePress source for the versioned documentation site
examples/ Notebooks and runnable examples
foreblocks/ Main forecasting library
foretools/ Companion tooling

foreblocks/

Path Purpose
foreblocks/__init__.py Top-level public exports
foreblocks/config.py Public configuration dataclasses (ModelConfig, TrainingConfig)
foreblocks/models/ Model-level composition APIs (ForecastingModel, GraphForecastingModel)
foreblocks/layers/ Reusable layer families, including graph convolutions and graph construction
foreblocks/core/ Core forecasting internals and heads
foreblocks/training/ Trainer and training support
foreblocks/evaluation/ Evaluation and metrics
foreblocks/data/ Dataset and dataloader helpers
foreblocks/ts_handler/ Preprocessing and sequence construction
foreblocks/tf/ Transformer stack and advanced attention
foreblocks/darts/ Neural architecture search
foreblocks/mltracker/ Experiment tracking
foreblocks/hybrid_mamba/ Hybrid Mamba SSM blocks (HybridMambaBlock, HybridMamba2Block, SSD)
foreblocks/mamba/ Original Mamba backbone with MoE, positional encoding, and eval tools
foreblocks/kan/ Kolmogorov-Arnold Network backbone

foretools/

Path Purpose
foretools/tsgen/ Synthetic time-series generation
foretools/bohb/ BOHB, TPE configuration, pruning, and optimization plots
foretools/foreminer/ Exploratory analysis and diagnostics
foretools/fengineer/ Feature engineering utilities
foretools/emd_like/ Decomposition tools
foretools/tsaug/ AutoDA-Timeseries: automated data augmentation with adaptive policy

Recommended entry points by task

Task Entry point
Training a baseline model README.md, Getting Started
Understanding architecture composition foreblocks/models/
Working with graph forecasting foreblocks/models/graph_forecasting.py, foreblocks/layers/graph/
Configuring runs foreblocks/config.py
Building dataloaders foreblocks/data/dataset.py
Adding preprocessing logic foreblocks/ts_handler/preprocessing.py
Exploring transformer internals foreblocks/tf/transformer.py
Working on architecture search foreblocks/darts/
Using SSM / Mamba-style blocks foreblocks/hybrid_mamba/layers.py
Generating synthetic data foretools/tsgen/
Running hyperparameter search foretools/bohb/
Augmenting training data adaptively foretools/tsaug/

Related pages