Spektral 1.3.1 GCN Call Exception

[jm6898]

from spektral.layers import GCNConv
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Dropout
from spektral.datasets import citation

#Spektral package version == 1.3.1

#Trying to run a GCN model but following exception error showed up when run e.g.

GCNConv(16, 'relu')([X_in, A_in])

---

RuntimeError: Exception encountered when calling GCNConv.call().

Could not automatically infer the output shape / dtype of 'gcn_conv_3' (of type GCNConv). Either the GCNConv.call() method is incorrect, or you need to implement the GCNConv.compute_output_spec() / compute_output_shape() method. Error encountered:

Tried to convert 'y' to a tensor and failed. Error: None values not supported.

[ByeongJu-Kang]

I also get the same error..

I use ARMAConv layer.
The code that was working well before suddenly produces the same error with you.

---

ValueError Traceback (most recent call last)
in <cell line: 1>()
----> 1 graph_conv_1 = spektral.layers.ARMAConv(channels = 2000, activation = 'elu')([H_in, A_in])
2
3 graph_conv_2 = spektral.layers.ARMAConv(channels = 200,
4 activation = 'elu')([graph_conv_1, A_in])
5

2 frames
/usr/local/lib/python3.10/dist-packages/spektral/layers/convolutional/arma_conv.py in call(self, inputs, mask)
142
143 if mask is not None:
--> 144 output *= mask[0]
145 output = self.activation(output)
146

ValueError: Exception encountered when calling ARMAConv.call().

**Could not automatically infer the output shape / dtype of 'arma_conv' (of type ARMAConv). Either the ARMAConv.call() method is incorrect, or you need to implement the ARMAConv.compute_output_spec() / compute_output_shape() method. Error encountered:

Tried to convert 'y' to a tensor and failed. Error: None values not supported.**

Arguments received by ARMAConv.call():
• args=(['<KerasTensor shape=(None, 5), dtype=float32, sparse=None, name=keras_tensor_1>', '<KerasTensor shape=(None, 2808), dtype=float32, sparse=None, name=keras_tensor>'],)
• kwargs={'mask': ['None', 'None']}

When using spektral version 1.0.3, it works as before, but is very slow.
Therefore, I think there is currently a problem with spektral 1.3.0 and 1.3.1
(I also tried 1.3.0 but got the same error)

I'm waiting for error resolution.

[pppxiyu]

I have the same issue when using GCNConv.