Type Error while Training a new MEGNetModel from structures about megnet HOT 9 CLOSED

@kdmsit can you give a minimal example to reproduce your error? From the error, It can be related to the data you use too

from megnet.

Data is being used from megnet.data.crystal import CrystalGraph it seems !!

from megnet.

CrystalGraph is just a way to convert your crystal structures (pymatgen Structure objects) into a graph. Please see my previous comment

from megnet.

I didnot get you. I have not read any data yet.
Crsytal Graph structures (pymatgen Structure objects) I am getting from "from megnet.data.crystal import CrystalGraph". Later that is converted to graph using graph_converter = CrystalGraph(cutoff=r_cutoff).
Then when I am initialising the model by model = MEGNetModel(graph_converter=graph_converter, centers=gaussian_centers, width=gaussian_width), I am getting the error. I am getting this error to initialise the model.

from megnet.

@kdmsit I copied and pasted the code, but it runs fine on my laptop. Please try to update megnet to latest github version and also update tensorflow.

from megnet.

Thanks a lot!!.. it worked. TensorFlow version issue was there.

One more thing I just want to ask. To train it for large data like around 30K what should be the suitable epochs and nfeat_bond ?

from megnet.

It really depends on the type of the data, maybe you can try the nfeat_bond 100 and epoch 1000.

from megnet.

I have run the model for Band-Gap with 20%(7300 Materials) I used to train and rest 80% test (30K). I got around MAE as 1.43. Can you comments on this?
I think this is far more than other baselines?

from megnet.

@kdmsit yes that’s too high in general, but is possible depending on the data you used. For MP data, this is way too high

from megnet.