Megnet & NMPN error using Custom dataset

[thegodone]

Traceback (most recent call last):
File "GCN-keras.py", line 213, in
dataloader = NumpyTensorList(*[getattr(dataset, x['name']) for x in hyper['model']['inputs']])
File "/Users/itetko/anaconda3/lib/python3.8/site-packages/kgcnn-1.1.2-py3.8.egg/kgcnn/io/loader.py", line 34, in init
File "/Users/itetko/anaconda3/lib/python3.8/site-packages/kgcnn-1.1.2-py3.8.egg/kgcnn/io/loader.py", line 37, in _length_test
File "/Users/itetko/anaconda3/lib/python3.8/site-packages/kgcnn-1.1.2-py3.8.egg/kgcnn/io/loader.py", line 37, in
TypeError: object of type 'NoneType' has no len()

I use the last version of the code 1.1.2 Master bug got an error.

I don't see this error with GraphSage, GAT, GATv2 models parameter

[thegodone]

I use this data preparation script

[PatReis]

The error means that the name of the input layer does not have a property set in the dataset. For DimeNetPP it is angle_indices for sure. so you also need
dataset.set_angle()
But I should write something to that also the name is printed in the error message, otherwise we would not know what it was...

[PatReis]

ah for GCN its edge_weights I guess
the can be set for the dataset via normalize_edge_weights_sym()

[thegodone]

Great thanks Envoyé de mon iPhone

…

Le 1 déc. 2021 à 18:04, Patrick Reiser ***@***.***> a écrit :  ah for GCN its edge_weights I guess the can be set for the dataset via normalize_edge_weights_sym() — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[PatReis]

for megnet it is probably the name="graph_number" but that can be replaces with "graph_attributes" for the inputs of the model paramters MegNet needs a graph tensor as state or graph input feature.

[thegodone]

Thanks for the tips, I was able to run GCN and DimenetPP models now. But I still don't know what to do with the Megnet code. Cause I don't see in the repo where to set the graph_attributes in the features of my custom dataset.

[PatReis]

oh sorry, I meant in the hyper parameters for input there is
"inputs": [ {"shape": [None], "name": "node_number", "dtype": "float32", "ragged": True}, {"shape": [None, 3], "name": "node_coordinates", "dtype": "float32", "ragged": True}, {"shape": [None, 2], "name": "range_indices", "dtype": "int64", "ragged": True}, {"shape": [2], "name": "graph_attributes", "dtype": "float32", "ragged": False} ],
And when there is a not a graph_attributes then this could be a problem.

[thegodone]

I see but can we make them on the fly i see that it is mw só not difficult to add a dataset.set_graph_attribute(mw) ? Envoyé de mon iPhone

…

Le 2 déc. 2021 à 14:14, Patrick Reiser ***@***.***> a écrit :  oh sorry, I meant in the hyper parameters for input there is "inputs": [ {"shape": [None], "name": "node_number", "dtype": "float32", "ragged": True}, {"shape": [None, 3], "name": "node_coordinates", "dtype": "float32", "ragged": True}, {"shape": [None, 2], "name": "range_indices", "dtype": "int64", "ragged": True}, {"shape": [2], "name": "graph_attributes", "dtype": "float32", "ragged": False} ], And when there is a not a graph_attributes then this could be a problem. — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

[thegodone]

I remove graph_attributes part from the hyper json but got another error

my settings:

[PatReis]

No No sorry, you can not remove it. MegNet needs graph attributes. I meant if in place of graph_attributes there is something like graph_size or graph_number (which is not set, then there will be a Problem.) But graph_attributes should be set automatically.

[PatReis]

Can you put this before the dataloader part and run again?

for x in hyper['model']['inputs']:
    if not hasattr(dataset, x['name']):
        print("ERROR: Dataset does not has information on %s" % x['name'])

[PatReis]

I will make a method to the dataset base class like process_hyper

[thegodone]

this is really an excellent idea to avoid mistake between data process and architectude needs dependencies.

[thegodone]

by using this part in the "json":
{"shape": [2], "name": "graph_attributes", "dtype": "float32", "ragged": False}
fix the issue, I use version master of the morning before our new modifications.