Dimenet_QM9 pretrained error

[sudaksh14]

Hi PyG Community,
I was trying to load the pretrained dimenet model from PyG but faced with the below error:

Could you pls suggest what could be missing as I have already installed the latest version of tensorflow in the environment.

[rusty1s]

Which version of tensorflow are you using? tf.train.load_checkpoint should exist, see https://www.tensorflow.org/api_docs/python/tf/train/load_checkpoint.

[sudaksh14]

Hi Matthias and Edison,
I came up with another issue with using the pre-trained dimenet model which is as below:

Here the error seems that the original dimenet was written with tf but pytorch geometric is based upon pytorch. Could you suggest any solution.
Thanks for your help and time!

[rusty1s]

Is it possible for you to print train_idx in nn/models/dimenet.py line 569? What's the output here?

[sudaksh14]

Hi Matthias,
Below is the snapshot of output:

Its strange that the type is torch.tensor, still the error above says that torch.tensor type is acceptable.

[EdisonLeeeee]

torch.int32 is not supported as an index, you need to convert it to torch.long instead, e.g., x.long().

[sudaksh14]

Thnx Edison,
It works now but I downloaded the QM9 data from the pytorch geometric library, so I think it could be updated to implement the long tensor as the data type.