TypeError: Compose expect `Symbol` as arguments

[poorfriend]

I want to implement CNN_BLSTM_CTC with CUDNN, but an error arise:
Traceback (most recent call last):
File "/home/Documents/MyCode/VLPR/view.py", line 25, in
output = CNN_MultiBLSTM_CTC_CUDNN(num_hidden, num_label, num_lstm_layer)
File "/home/Documents/MyCode/VLPR/blstm.py", line 325, in CNN_MultiBLSTM_CTC_CUDNN
output, _ = cell.unroll(num_length, inputs=hidden, merge_outputs=True, layout='NTC')
File "/home/mxnet/python/mxnet/rnn/rnn_cell.py", line 637, in unroll
inputs = symbol.Concat(inputs, dim=0)
File "/home/mxnet/python/mxnet/_ctypes/symbol.py", line 191, in creator
s._compose(*args, name=name, **symbol_kwargs)
File "/home/mxnet/python/mxnet/symbol.py", line 241, in _compose
raise TypeError('Compose expect Symbol as arguments')
TypeError: Compose expect Symbol as arguments

Here is my code, How can I fix it

`def CNN_MultiBLSTM_CTC_CUDNN(num_hidden, num_label, num_lstm_layer, num_class = 70, num_length = 32):

data = mx.sym.Variable('data')
label = mx.sym.Variable('label')

wordvec = CNN(data)
hidden = [mx.symbol.Flatten(wordvec[i]) for i in range(num_length)]

cell = mx.rnn.FusedRNNCell(num_hidden, num_layers=num_lstm_layer, mode='lstm', bidirectional=True)

output, _ = cell.unroll(num_length, inputs=hidden, merge_outputs=True, layout='NTC')

pred = mx.sym.Reshape(output, shape=(-1, num_hidden))

pred = mx.sym.FullyConnected(data=pred, num_hidden=num_class)

label = mx.sym.Reshape(data=label, shape=(-1,))
label = mx.sym.Cast(data=label, dtype='int32')
sm = mx.sym.WarpCTC(data=pred, label=label, label_length=num_label, input_length=num_length)

return sm`

def CNN3(indata):
"""the net is similar to VGG16
each image size is 64256
"""
# stage 1
body = CNN_Module(data=indata, kernel=(3, 3), pad=(1, 1), num_filter=64, stride=(1, 1), num_module=2, BatchNorm=True)
body = mx.symbol.Pooling(data=body, pool_type="max", kernel=(2, 2), stride=(2, 2)) #32128
# stage 2
body = CNN_Module(data=body, kernel=(3, 3), pad=(1, 1), num_filter=128, stride=(1, 1), num_module=2, BatchNorm=True)
body = mx.symbol.Pooling(data=body, pool_type="max", kernel=(2, 2), stride=(2, 2)) #1664
# stage 3
body = CNN_Module(data=body, kernel=(3, 3), pad=(1, 1), num_filter=256, stride=(1, 1), num_module=2, BatchNorm=True)
body = mx.symbol.Pooling(data=body, pool_type="max", kernel=(2, 2), stride=(2, 2)) #832
# stage 4
body = CNN_Module(data=body, kernel=(3, 3), pad=(1, 1), num_filter=256, stride=(1, 1), num_module=2, BatchNorm=True)
body = mx.symbol.Pooling(data=body, pool_type="max", kernel=(2, 1), stride=(2, 1)) #432
# stage 5
body = CNN_Module(data=body, kernel=(3, 3), pad=(1, 1), num_filter=512, stride=(1, 1), num_module=2, BatchNorm=True)
body = mx.symbol.Pooling(data=body, pool_type="max", kernel=(2, 1), stride=(2, 1)) #232
# stage 5
body = CNN_Module(data=body, kernel=(2, 1), pad=(0, 0), num_filter=512, stride=(1, 1), num_module=1, BatchNorm=True)
body = mx.sym.SliceChannel(data=body, num_outputs=32, axis=3, squeeze_axis=1)

return body

[poorfriend]

@piiswrong

[Godricly]

Maybe updating to latest version will help. #5114

[poorfriend]

@Godricly , Thank ,it works. but when I run

output = CNN_MultiBLSTM_CTC_CUDNN(num_hidden, num_label, num_lstm_layer) mx.viz.plot_network(output,shape = {'data': (5,1,64,256),'label':(5,8)}).view()

another error arise:
infer_shape error. Arguments:
Traceback (most recent call last):
data: (5, 1, 64, 256)
File "/home/Documents/MyCode/VLPR/view.py", line 48, in
label: (5, 8)
mx.viz.plot_network(output,shape = {'data': (5,1,64,256),'label':(5,8)}).view()
File "/home/mxnet/python/mxnet/visualization.py", line 206, in plot_network
_, out_shapes, _ = interals.infer_shape(**shape)
File "/home/mxnet/python/mxnet/symbol.py", line 562, in infer_shape
res = self._infer_shape_impl(False, *args, **kwargs)
File "/home/mxnet/python/mxnet/symbol.py", line 645, in _infer_shape_impl
ctypes.byref(complete)))
File "/home/mxnet/python/mxnet/base.py", line 77, in check_call
raise MXNetError(py_str(_LIB.MXGetLastError()))
mxnet.base.MXNetError: Error in operator warpctc0: Shape inconsistent, Provided=(40,), inferred shape=(80,)

I guess this is caused by bidirectional lstm, but I don't know how to fix it, can you help me

[poorfriend]

@piiswrong, Can I used fusedrnn to struct a multilayer bidirectional lstm

[ysh329]

same problem.

TypeError: Compose expect `Symbol` as arguments

[szha]

This issue is closed due to lack of activity in the last 90 days. Feel free to ping me to reopen if this is still an active issue. Thanks!