AtTimeStepSplitter -- code not align

[pigu163]

For splitter - AtTimeStepSplitter , code not align, which one is correct ?

Code from paper Scalable Spatiotemporal Graph Neural Networks -- AAAI 2023

else:
    val_idx = np.setdiff1d(val_idx, test_idx)
    train_idx = np.setdiff1d(np.arange(len(dataset)), test_idx)
    train_idx = np.setdiff1d(train_idx, val_idx)
return self.set_indices(train_idx, val_idx, test_idx)

Code from TSL

else:
    val_idx = np.setdiff1d(val_idx, test_idx)
    train_idx = np.setdiff1d(np.arange(len(dataset)), test_idx)
train_idx = np.setdiff1d(train_idx, val_idx)
return self.set_indices(train_idx, val_idx, test_idx)

train_idx will change the size of training set

[marshka]

Hi, the code from tsl is the most up-to-date, please consider using it. In this example, both else branches have the same outcome. In tsl snippet we also ensure that in all cases validation samples are excluded from the training set.