Issues encountered while parsing CIF

[YuanbinLiu]

Dear authors,

When I run main.py, I got a UserWarning "Issues encountered while parsing CIF: Some fractional co-ordinates rounded to ideal values to avoid issues with finite precision". Do this warning have any effect on the results of CGCNN?

[Andrew-S-Rosen]

Hi @YuanbinLiu.

The warning you received is from Pymatgen. Specifically, the warning is from this section of Pymatgen's CIF parser. In short, this should have no noticeable effect on the CGCNN training or evaluation. The rounding, should it occur, is extraordinarily small.

[YuanbinLiu]

Got it, thanks a lot.

[SANTKJD]

Dear authors，

When I run main.py，I got a warning "warnings.warn("Issues encountered while parsing CIF: " + "\n".join(self.warnings))".After running I just got two IDs in test_results.csv.Can you tell how to deal with it?

[Andrew-S-Rosen]

The Pymatgen-generated warning mentioned above is not an issue. If you are only receiving two entries in your testing set, have you made sure to download the dataset of 20k+ MOFs? By default, this repo only ships with a toy example. You need to download the data from Figshare.

[SANTKJD]

I have trained 29000+ cif files.All the cifs are downloaded from materials project by API.

[Andrew-S-Rosen]

Without further details, I can't help debug your issue for you.

I should mention that the CGCNN code used here is almost identical to the original (https://github.com/txie-93/cgcnn). If you have issues with CGCNN, you may wish to consider posting an issue there.

On Sun, Jul 10, 2022, 11:22 AM SANTKJD @.***> wrote:

I have trained 29000+ cif files

— Reply to this email directly, view it on GitHub https://github.com/arosen93/QMOF/issues/19#issuecomment-1179775934, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACCFWGCRBSDM3VFBHCU4I53VTMIHVANCNFSM46NM2ZMQ . You are receiving this because you modified the open/close state.Message ID: @.***>

[SANTKJD]

[SANTKJD]

Here is the test_results.

[Andrew-S-Rosen]

According to your image above, you specifically requested a test set size of 2. As you requested, you can see there are only two entries in the testing set CSV.

[SANTKJD]

But the value I got seems much different from the actual value.

[Andrew-S-Rosen]

Your training set size is only 8. That's orders of magnitude smaller than what you need for deep learning. It's no surprise your testing set results are not what you expect them to be. This is not an issue with the code, so please direct your more fundamental comments to either the parent CGCNN repo or a machine learning forum.

[SANTKJD]

OK，I will try again，thanks a lot .

[SANTKJD]

Dear author， I found that the number I set affects whether it reports errors or not.