NotImplementedError: Can't modify this type of data

[sahooora]

Below there is the snippet of code I use for import and conversion:

import onnx
from onnx2keras import onnx_to_keras

onnx_model = onxx.load('UCI_simplification_1.onnx')
k_model = onnx_to_keras(onnx_model, ['input_1'])

When I run I get:

INFO:onnx2keras:Converter is called.
DEBUG:onnx2keras:List input shapes:
DEBUG:onnx2keras:None
DEBUG:onnx2keras:List inputs:
DEBUG:onnx2keras:Input 0 -> input_1.
DEBUG:onnx2keras:List outputs:
DEBUG:onnx2keras:Output 0 -> dense.
DEBUG:onnx2keras:Gathering weights to dictionary.
DEBUG:onnx2keras:Found weight dense/kernel:0 with shape (128, 6).
DEBUG:onnx2keras:Found weight dense/bias:0 with shape (6,).
DEBUG:onnx2keras:Found weight lstm_W with shape (1, 512, 3).
DEBUG:onnx2keras:Found weight lstm_R with shape (1, 512, 128).
DEBUG:onnx2keras:Found weight lstm_B with shape (1, 1024).
DEBUG:onnx2keras:Found input input_1 with shape [128, 3]
DEBUG:onnx2keras:######
DEBUG:onnx2keras:...
DEBUG:onnx2keras:Converting ONNX operation
DEBUG:onnx2keras:type: Transpose
DEBUG:onnx2keras:node_name: lstm_X
DEBUG:onnx2keras:node_params: {'perm': [1, 0, 2], 'change_ordering': False, 'name_policy': None}
DEBUG:onnx2keras:...
DEBUG:onnx2keras:Check if all inputs are available:
DEBUG:onnx2keras:Check input 0 (name input_1).
DEBUG:onnx2keras:... found all, continue
WARNING:onnx2keras:transpose:Can't permute batch dimension. Result may be wrong.
Traceback (most recent call last):
  File "digits.py", line 31, in <module>
    model_k = onnx_to_keras(onnx_model, ['input_1'])
  File "/usr/local/lib/python3.8/dist-packages/onnx2keras/converter.py", line 175, in onnx_to_keras
    AVAILABLE_CONVERTERS[node_type](
  File "/usr/local/lib/python3.8/dist-packages/onnx2keras/reshape_layers.py", line 27, in convert_transpose
    raise NotImplementedError('Can\'t modify this type of data')
NotImplementedError: Can't modify this type of data

Thanks in advance

[Andredance]

Hi. Please, share your ONNX model

[sahooora]

Hi. Please, share your ONNX model

Here it is.

UCI.tar.gz

Thanks in advance

[Andredance]

I checked your model and got some:

• Current implementation doesn't support Transpose operation, if it's input, isn't a NumPy array (in your case Transpose is used immediately after Input layer, which is TF type).
• I think it's possible is some way to make it work with TF type but It will require much time and will not solve your problem totally because after that Transpose layer you have LSTM layer which is not supported by the library at all.

Sorry for answering for a long time. Maybe in the future, we gonna support LSTM but I can say exactly when. Also, you can try to make it work by yourself if you want. Good luck!