A Sparse Gating Convolutional Recurrent Network for Traffic Flow Prediction

Table 2

The details of module on TaxiBJ dataset.

Hyperparameter	ConvGRU	SConvGRU	SConvGRU+	ConvLSTM	SConvLSTM	SConvLSTM+
CNN layers	2	2	2	2	2	2
Number of filters in CNN	8 (first layer), 16 (second layer)
Kernel size in CNN	(3, 3)	(3, 3)	(3, 3)	(3, 3)	(3, 3)	(3, 3)
Stride in CNN	1 (first layer), 2 (second layer)
Unit layers	1	1	1	1	1	1
Hidden channels of unit	64	64	64	64	64	64
Kernel size in unit	(3, 3)	(3, 3)	(3, 3)	(3, 3)	(3, 3)	(3, 3)
Input of gating mechanism
DCNN layers	2	2	2	2	2	2
Number of filters in DCNN	8 (first layer), 2 (second layer)
Kernel size in DCNN	(3, 3)	(3, 3)	(3, 3)	(3, 3)	(3, 3)	(3, 3)
Stride in DCNN	2 (first layer), 1 (second layer)
Batch size	16	16	16	16	16	16
Timestep	10	10	10	10	10	10
Epoch	300	300	300	300	300	300
Optimizer	Adam [36]	Adam	Adam	Adam	Adam	Adam
Learning rate	0.0002	0.0002	0.0002	0.0002	0.0002	0.0002
Strategy	Early-stopping	Early-stopping	Early-stopping	Early-stopping	Early-stopping	Early-stopping