[GitHub] dbsxdbsx opened a new issue #9068: How to use sym.contrib.ctc_loss instead of baidu WARPCTC

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

dbsxdbsx opened a new issue #9068: How to use sym.contrib.ctc_loss instead of baidu WARPCTC
URL: https://github.com/apache/incubator-mxnet/issues/9068
 
 
   win10, mxnet 0.12,python 2.7
   For the official example WARPCTC, I am trying to use sym.contrib.ctc_loss instead of baidu WARPCTC, as I am with windows. So I modified `lstm.py`. 
   ```
   def lstm_unroll(num_lstm_layer, seq_len,
                   num_hidden, num_label):
       param_cells = []
       last_states = []
       for i in range(num_lstm_layer):
           param_cells.append(LSTMParam(i2h_weight=mx.sym.Variable("l%d_i2h_weight" % i),
                                        i2h_bias=mx.sym.Variable("l%d_i2h_bias" % i),
                                        h2h_weight=mx.sym.Variable("l%d_h2h_weight" % i),
                                        h2h_bias=mx.sym.Variable("l%d_h2h_bias" % i)))
           state = LSTMState(c=mx.sym.Variable("l%d_init_c" % i),
                             h=mx.sym.Variable("l%d_init_h" % i))
           last_states.append(state)
       assert(len(last_states) == num_lstm_layer)
   
       # embeding layer
       data = mx.sym.Variable('data')
       label = mx.sym.Variable('label')
       wordvec = mx.sym.SliceChannel(data=data, num_outputs=seq_len, squeeze_axis=1)
   
       hidden_all = []
       for seqidx in range(seq_len):
           hidden = wordvec[seqidx]
           for i in range(num_lstm_layer):
               next_state = lstm(num_hidden, indata=hidden,
                                 prev_state=last_states[i],
                                 param=param_cells[i],
                                 seqidx=seqidx, layeridx=i)
               hidden = next_state.h
               last_states[i] = next_state
           hidden_all.append(hidden)
   
       hidden_concat = mx.sym.Concat(*hidden_all, dim=0)
       pred = mx.sym.FullyConnected(data=hidden_concat, num_hidden=11)
   
       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 = seq_len)
       sm =mx.sym.contrib.ctc_loss(data=pred, label=label, label_lengths = num_label, data_lengths  = seq_len,use_data_lengths =True,use_label_lengths=True)
       return sm
   ``` 
   Noitice that ` sm =mx.sym.contrib.ctc_loss(data=pred, label=label, label_lengths = num_label, data_lengths  = seq_len,use_data_lengths =True,use_label_lengths=True)` is what I modified. Then running `toy_ctc.py`,  I get error said: `AssertionError: Argument data_lengths must be Symbol instances, but got 80`
   
   I wonder how to fix it.

----------------------------------------------------------------
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