Editable Graph Temporal Model

Code for editable graph temporal model --- GATRNN, which can jointly learn to infer relational graph and forecast multivariate time series. It is designed to be easily edited from user's feedback on the predicted relational graphs.

Data

In this code, the raw data file is a .npz file produced by the savez function of numpy. In the file, a dictionary-like object is saved, where the time series array (shape: seq_len x num_nodes x num_features) can be indexed by the key 'x' and the adjacency matrix array (shape: num_nodes x num_nodes) can be indexed by the key 'adj'. Please save your data in this format when adding new data if possible.

Usage

Run model training:

python -m script_train -s /path/to/save/results

Run model editing:

python -m script_edit -mp /path/to/trained/model -s /path/to/save/results