The loss of STSGCN is irregularly large

[SonghuaHu-UMD]

I am currently testing the STSGCN and found the loss is irregularly large (>1e10).
I read the code and found the output is accumulated after the graph convolution:

Bigscity-LibCity/libcity/model/traffic_flow_prediction/STSGCN.py

Line 50 in 85d1351

data = torch.matmul(adj, data)

I checked the paper and code and thought they use the distance inverse as the adjacent matrix, rather than the Gauss kernel (https://github.com/Davidham3/STSGCN/blob/master/utils.py#L105).

    # Fills cells in the matrix with distances.
    with open(distance_df_filename, 'r') as f:
        f.readline()
        reader = csv.reader(f)
        for row in reader:
            if len(row) != 3:
                continue
            i, j, distance = int(row[0]), int(row[1]), float(row[2])
            if type_ == 'connectivity':
                A[i, j] = 1
                A[j, i] = 1
            elif type == 'distance':
                A[i, j] = 1 / distance
                A[j, i] = 1 / distance
            else:
                raise ValueError("type_ error, must be "
                                 "connectivity or distance!")
    return A

I added a new function in traffic_state_dataset to calculate the distance inverse for STSGCN and the loss now looks reasonable
(PEMSD7(M), 12 prediction steps):

MAE	MAPE	MSE	RMSE	masked_MAE	masked_MAPE	masked_MSE	masked_RMSE	R2	EVAR
1.461792	0.033766	6.305071	2.51099	1.461792	0.033766	6.305071	2.51099	0.966612	0.967362
1.886784	0.044624	12.02078	3.4671	1.886784	0.044624	12.02078	3.4671	0.936329	0.937318
2.208205	0.05365	17.84879	4.224782	2.208205	0.05365	17.84879	4.224782	0.90546	0.907067
2.445271	0.060798	23.0774	4.803894	2.445271	0.060798	23.0774	4.803894	0.877856	0.879756
2.636102	0.067311	27.92206	5.284132	2.636102	0.067311	27.92206	5.284132	0.852311	0.854849
2.793431	0.072273	32.20569	5.675005	2.793431	0.072273	32.20569	5.675005	0.829672	0.8325
2.914345	0.076151	35.48109	5.956601	2.914345	0.076151	35.48109	5.956601	0.812326	0.815592
3.017157	0.079508	38.48553	6.203671	3.017157	0.079508	38.48553	6.203671	0.796511	0.799579
3.103143	0.082126	40.83116	6.389926	3.103143	0.082126	40.83116	6.389926	0.784075	0.786962
3.165864	0.084001	42.70291	6.534747	3.165864	0.084001	42.70291	6.534747	0.774234	0.776773
3.236152	0.086049	44.63507	6.680949	3.236152	0.086049	44.63507	6.680949	0.763969	0.766483
3.298688	0.087792	46.06493	6.787116	3.298688	0.087792	46.06493	6.787116	0.756369	0.75932

[aptx1231]

Good job! Thank you!

[aptx1231]

You can merge this PR.

[SonghuaHu-UMD]

You can merge this PR.

Thanks for approving! I don't have the write access so I could not merge this PR.