Attention Diffusion Network

PyTorch implementation of the Attention Diffusion Network from "Structured Time Series Prediction without Structural Prior".

This project is a work in progress.

Requirements

Python 3.6+
Torch ~1.10
einops

Install

From pip:

pip install adn-torch

Or from the source code (in editable / developer mode):

git clone https://github.com/radandreicristian/adn.git
cd adn
pip install -e .[dev]

Usage

The package exposes the ADN model as an API. Example usage:

from adn import ADN

model = ADN(d_features = 1,
            d_hidden=32,
            d_feedforward=256,
            n_heads=4,
            p_dropout=0.3,
            batch_size=64,
            n_blocks=3,
            spatial_seq_len=12,
            temporal_seq_len=315)

The model takes 6 arguments in its forward pass - For both the source and the target sequence, the features, the temporal descriptors and the spatial descriptors. A detailed example is provided in the adn/example.py file.