TypeError when trying seq2seq model

[abhimohta]

Hi

I have the following architecture: Given two sequences (first and second) I am trying to predict a sequence of floats (labels) inside the multiple LSTM-based model. The way I have implemented the model is similar to the sequence to sequence model in Tutorial 204. The code calling the model is as under:

firstAxis = Axis.new_unique_dynamic_axis('firstAxis')
secondAxis = Axis.new_unique_dynamic_axis('secondAxis')
labelAxis = Axis.new_unique_dynamic_axis('labelAxis')
FirstSequence = C.sequence.input_variable(len(word_to_idx), sequence_axis=firstAxis, name="first")
SecondSequence = C.sequence.input_variable(len(word_to_idx), sequence_axis=secondAxis, name="second")
LabelSequence = C.sequence.input_variable(1, sequence_axis=labelAxis, name="label")
def create_model_train(s2smodel):
    @C.Function
    def model_train(first: FirstSequence,
                    second: SecondSequence,
                    labels: LabelSequence): # The error is on this line
        past_labels = C.layers.Delay(initial_state=sentence_start)(labels)
        return s2smodel(past_labels, first, second)
    return model_train

However, in the function model_train I get this error:

Traceback (most recent call last):
  File "cntktrain-lstm.py", line 230, in <module>
    logs = train(model, data, args, logs, data['train'].word_to_idx, 1)
  File "cntktrain-lstm.py", line 187, in train
    model_train = create_model_train(s2smodel)
  File "cntktrain-lstm.py", line 150, in create_model_train
    labels: LabelSequence): 
  File "/root/anaconda3/envs/cntk-py35/lib/python3.5/site-packages/cntk/ops/functions.py", line 109, in __new__
    return Function._to_Function(*args, **kwargs)
  File "/root/anaconda3/envs/cntk-py35/lib/python3.5/site-packages/cntk/ops/functions.py", line 163, in _to_Function
    args = [make_arg_variable(arg_name, annotations) for arg_name in arg_names]
  File "/root/anaconda3/envs/cntk-py35/lib/python3.5/site-packages/cntk/ops/functions.py", line 163, in <listcomp>
    args = [make_arg_variable(arg_name, annotations) for arg_name in arg_names]
  File "/root/anaconda3/envs/cntk-py35/lib/python3.5/site-packages/cntk/ops/functions.py", line 139, in make_arg_variable
    var_type = Variable._Type._sanitize(var_type)
  File "/root/anaconda3/envs/cntk-py35/lib/python3.5/site-packages/cntk/variables.py", line 260, in _sanitize
    if the_type == float:
  File "/root/anaconda3/envs/cntk-py35/lib/python3.5/site-packages/cntk/cntk_py.py", line 3761, in <lambda>
    Variable.__eq__ = lambda a,b: (a is not None and b is not None and Variable_eq(a,b)) or (a is None and b is None)
TypeError: in method 'Variable_eq', argument 2 of type 'CNTK::Variable const &'

I couldn't find anything online for this error. Can someone please guide me?

[ke1337]

Please take a look at the debugging manual.