Alternative implementation of the DCGRU recurrent cell in Tensorflow 2
Diffusion Convolutional Gated Recurrent Unit (DCGRU) is the recurrent cell adopted in the Diffusion Convolutional Recurrent Neural Network (DCRNN), a deep learning architecture introduced by Y. Li et al. [1] to tackle spatiotemporal forecasting tasks, where data is represented as a graph signal. The DCGRU cell is a modification of the standard GRU, having each matrix multiplication operation replaced by the the diffusion convolution operation (see [1]).
(Since I'm not part of the authors of the original DCRNN paper [1], this should be treated as an unofficial implementation)
Define the dcgru cell:
from tensorflow import keras
from dcgru_cell_tf2 import DCGRUCell
dcgru_cell = DCGRUCell(units=20,adj_mx=G_adj_mx, K_diffusion=2,
num_nodes=num_nodes,filter_type="random_walk")Wrap the dcgru cell in a keras RNN layer:
dcgru_layer = keras.layers.RNN(dcgru_cell)A Jupyter notebook showing a basic usage of a DCGRU layer (DCGRU_example.ipynb)
https://colab.research.google.com/drive/1sq0s8j3gGpBvSq058aBcrvSX2wLuAE-t?usp=sharing
A Jupyter notebook with comparison of the DCGRU with other common neural networks (DCGRU_testing.ipynb)
https://colab.research.google.com/drive/1-Em1U6jJZO7yOL8Uimgm5o0vkLIPP0dE?usp=sharing
[1] Li, Y., Yu, R., Shahabi, C., Liu, Y.: Diffusion convolutional recurrent neural net-work: Data-driven traffic forecasting. arXiv preprint arXiv:1707.01926 (2017)