Drop Edge with gradient clipping and smaller model

com-youtube

gcn cycles: (4, 4, 6, 6, 8, 8)
conv sizes: (48, 48, 32, 32, 24, 24)
drops: (0.3, 0.3, 0.2, 0.2, 0.1, 0.1)
Largest Linear: 224 x 32
Graph: Data(edge_index=[2, 5975248], x=[1134890, 1], y=[1134890, 1])
{'val_top_1p': 79.10645047585477,
 'val_top_5p': 77.05484280276329,
 'val_top_10p': 78.77767889398972,
 'val_kendal': 59.79420296905783,
 'val_mse': 448.61683670491743,
 'val_max_error': 2120.5370786637122,
 'run_time': 240.21285891532898}

dblp

gcn cycles: (4, 4, 6, 6, 8, 8)
conv sizes: (48, 48, 32, 32, 24, 24)
drops: (0.3, 0.3, 0.2, 0.2, 0.1, 0.1)
Largest Linear: 224 x 32
Graph: Data(edge_index=[2, 17298012], x=[4000148, 1], y=[4000148, 1])
{'val_top_1p': 78.03554911127222,
 'val_top_5p': 70.12404565840195,
 'val_top_10p': 71.69299074532391,
 'val_kendal': 72.43665323132845,
 'val_mse': 370.7392910465112,
 'val_max_error': 471.159696087762,
 'run_time': 982.3556971549988}