search

hazdzz / STGCN

The PyTorch implementation of STGCN.
GNU Lesser General Public License v2.1
484 stars 106 forks source link

Could you explain this constraint? #5

Closed jeongwhanchoi closed 3 years ago

jeongwhanchoi commented 3 years ago

https://github.com/hazdzz/STGCN/blob/17f6dd00d381dcf1edbd8bd3ff13f795501bb166/main_road_traffic.py#L89-L90

I'd like to use other datasets, but it seems to need to understand the n_hist argument. I wonder about the relation among n_hist, K_t and etc.

jeongwhanchoi commented 3 years ago

I'm gonna use the dataset with the 60mins interval data. (cf. 5mins interval in PeMSD dataset) How should I config the arguments in that dataset environment?

Thank you

hazdzz commented 3 years ago

https://github.com/hazdzz/STGCN/blob/17f6dd00d381dcf1edbd8bd3ff13f795501bb166/main_road_traffic.py#L89-L90

I'd like to use other datasets, but it seems to need to understand the n_hist argument.

I wonder about the relation among n_hist, K_t and etc.

n_his must be greater than n_pred or equals to n_pred. Kt means kernel number of temporal convolution. n_his means how many time intervals for training, n_pred means which one time interval the model should output. For example, n_his = 12, time intervals = 5 mins, n_pred = 3, that means give model 12 5 = 60 mins data, the model should output the next 3 5 = 15 mins data.

hazdzz commented 3 years ago

I'm gonna use the dataset with the 60mins interval data. (cf. 5mins interval in PeMSD dataset)

How should I config the arguments in that dataset environment?

Thank you

You could refer to config files that I write for different datasets.

jeongwhanchoi commented 3 years ago

https://github.com/hazdzz/STGCN/blob/17f6dd00d381dcf1edbd8bd3ff13f795501bb166/main_road_traffic.py#L89-L90

I'd like to use other datasets, but it seems to need to understand the n_hist argument. I wonder about the relation among n_hist, K_t and etc.

n_his must be greater than n_pred or equals to n_pred. Kt means kernel number of temporal convolution. n_his means how many time intervals for training, n_pred means which one time interval the model should output. For example, n_his = 12, time intervals = 5 mins, n_pred = 3, that means give model 12 5 = 60 mins data, the model should output the next 3 5 = 15 mins data.

Thank you for your explanation.