Where is main.py?

[liushiliushi]

I couldn't find main.py as mentioned in README

[yangzhao1230]

Thank you for your interest in our work. Indeed, there have been significant changes in both the overall model and the codebase since the initial release. As you can see from the commit history, I have made corrections to the code. However, I have not had the opportunity to update the README file yet.

The main.py you mentioned has now been replaced with main_shuffle_wo_smiles.py. You can refer to this new code in the meantime. I apologize for any inconvenience this may have caused.

I am currently quite busy, but I plan to update the README once things settle down.

[liushiliushi]

I'm very interested in your work. Thanks for your prompt assistance in resolving the issue!

[liushiliushi]

By the way, could you please tell me how to extract pairs of (Graph, Text) data from KV-PLM into data/kv_data? Can I simply put the txt file under into data/kv_data? Thank you very much!

[yangzhao1230]

I am providing you with access to the well-formatted data through the following link: https://drive.google.com/file/d/1XTUwZYnyly5x0dZB7AprJw_hAQk7pHAr/view?usp=drive_link

[liushiliushi]

Thanks for your help!

I encountered another issue while running main_shuffle_wo_smiles.py. When the code reaches line 64, specifically, "for batch in dataloader", I encountered the following error:

_pickle.PicklingError: Can't pickle <class 'torch_geometric.data.batch.Batch'>: it's not the same object as torch_geometric.data.batch.Batch.

Do you know the reason for this issue? I suspect it might be related to the version of torch_geometric. Could you please provide the version of torch_geometric that you are using?

[yangzhao1230]

Please refer to https://github.com/ddz16/MoleculePrediction/tree/2dd496a71a6f05ae8322bab3fc3ffc44245891e9#environment

[liushiliushi]

Thank you for your kind help! I managed to resolve the issue by upgrading torch_geometric from version 2.0.0 to 2.1.0. Just to note, if the torch_geometric version is 2.4.0, it will also result in an error. So the version of torch_geometric is indeed crucial. I recommend that you consider mentioning the required torch_geometric version in the README when you are available.

I have another question. I encounterd this text in the terminal while running. Is this normal?

Some weights of the model checkpoint at bert_pretrained/ were not used when initializing BertModel: ['cls.predictions.decoder.weight', 'cls.predictions.bias', 'cls.seq_relationship.weight', 'cls.seq_relationship.bias', 'cls.predictions.transform.dense.bias', 'cls.predictions.decoder.bias', 'cls.predictions.transform.LayerNorm.bias', 'cls.predictions.transform.LayerNorm.weight', 'cls.predictions.transform.dense.weight']
- This IS expected if you are initializing BertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing BertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).

[yangzhao1230]

In just a warning, which doesn't matter. The original SciBERT further incorporates the CLS token into some downstream tasks, but we only use the features and do not require the weights for these downstream tasks.

[liushiliushi]

Ok, thanks for your time and assistance. I'm truly grateful for your help.

[qiansirjun]

May I ask if there is only zero sample cross modal retrieval task in the code? If I want to input only one text and retrieve the corresponding image, how should I modify the code？