[GitHub] shadowleaves opened a new issue #4045: inferred shape error with FullyConnected layer

[GitBox <gi...@apache.org>]

shadowleaves opened a new issue #4045: inferred shape error with FullyConnected layer
URL: https://github.com/apache/incubator-mxnet/issues/4045
 
 
   very simple program that starts with two input nodes (batch_size=100, n_inputs=2) and then maps it to a layer of 3 hidden nodes via a 2x3 weight matrix. got the following error. any idea what's going on? how could the inferred shape be (3,2)? 
   
   ```
   mxnet.base.MXNetError: InferShape Error in fullyconnected0's weight argument
   Shape inconsistent, Provided=(2,3), inferred shape=(3,2)Corresponding keyword of symbol: wx
   Shape inconsistent, Provided=(2,3), inferred shape=(3,2)
   ```
   
   ```
   #!/usr/bin/env python
   
   import mxnet as mx
   import numpy as np
   
   
   if __name__ == '__main__':
   
       # Create dataset
       nb_train = 2000  # Number of training samples
       batch_size = 100
   
       X_train = np.random.sample((nb_train, 2))
       T_train = np.random.sample((nb_train, ))
   
       train_dataiter = mx.io.NDArrayIter(data=X_train,
                                          label=T_train,
                                          batch_size=batch_size,
                                          label_name='softmax_label',
                                          shuffle=True)
   
       net = mx.symbol.Variable('data')
       wx = mx.sym.Variable('wx', shape=(2, 3))
       net = mx.symbol.FullyConnected(data=net, num_hidden=3,
                                      weight=wx,
                                      no_bias=True)
       net = mx.symbol.SoftmaxOutput(data=net, name='softmax')
   
       mod = mx.mod.Module(net)
       mod.bind(data_shapes=train_dataiter.provide_data,
                label_shapes=train_dataiter.provide_label)
   ```

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services