How can I make a predict with a lstm-ctc checkpoint in mxnet?

I following example/warpctc/lstm_ocr.py to training a model. Now I had saved a checkpoint mymodel-0100.params and mymodel-symbol.json.
So, How can I make a predict with this checkpoint use only one image?

I had tired to use Predictor interface, code below:

# Load the pre-trained model
symbol_file = "mymodel-symbol.json"
param_file = "mymodel-0100.params"
predictor = Predictor(open(symbol_file).read(),
    open(param_file).read(),
    {'data':(80, 30)})

But data shape always raise error and I don't know how to set this value. Anybody help me thank you.

However, I also tried another way: add one line code in the end of mxnet/example/warpctc/lstm_ocr.py:

model = mx.model.FeedForward(ctx=contexts,
                             symbol=symbol,
                             num_epoch=num_epoch,
                             learning_rate=learning_rate,
                             momentum=momentum,
                             wd=0.00001,
                             initializer=mx.init.Xavier(factor_type="in", magnitude=2.34))

model.fit(X=data_train, eval_data=data_val,
          eval_metric = mx.metric.np(Accuracy),
          batch_end_callback=mx.callback.Speedometer(BATCH_SIZE, 50),)

model.save("ocr")

# add new line for predict
model.predict(data_val)

But it is always error output:

Traceback (most recent call last):
File "lstm_ctc_ocr.py", line 211, in <module>
training_all()
File "lstm_ctc_ocr.py", line 188, in training_all
model.predict(data_val)
File "/home/bobliu/Work/code/DL/mxnet/python/mxnet/model.py", line 618, in predict
self._init_predictor(data_shapes, type_dict)
File "/home/bobliu/Work/code/DL/mxnet/python/mxnet/model.py", line 541, in _init_predictor
self.ctx[0], grad_req='null', type_dict=type_dict, **dict(input_shapes))
File "/home/bobliu/Work/code/DL/mxnet/python/mxnet/symbol.py", line 685, in simple_bind
arg_types, _, aux_types = self.infer_type(**type_dict)
File "/home/bobliu/Work/code/DL/mxnet/python/mxnet/symbol.py", line 417, in infer_type
ctypes.byref(complete)))
File "/home/bobliu/Work/code/DL/mxnet/python/mxnet/base.py", line 77, in check_call
raise MXNetError(py_str(_LIB.MXGetLastError()))
mxnet.base.MXNetError: InferType Error in reshape0: [21:39:03] src/operator/./reshape-inl.h:345: Check failed: (dtype) != (-1) First input must have specified type

Solution

After reading the source code. I know what's wrong with me now.

1. Regrading to predictor() function.

We have to set data shape with a full symbol.
In this example:

    init_c = [('l%d_init_c'%l, (batch_size, num_hidden)) for l in range(num_lstm_layer)]
    init_h = [('l%d_init_h'%l, (batch_size, num_hidden)) for l in range(num_lstm_layer)]
    init_states = init_c + init_h

    all_shapes = [('data', (batch_size, 100 * 30))] + init_states + [('label', (batch_size, num_label))]

So, You have to set for data with all shapes.

2. Regrading to model.predict

I think a bug appear in source code, But i am not sure if i was wrong.
Please check predict function in model.py file.
change below code(Take care to X.provide_label):

    data_shapes = X.provide_data + X.provide_label
    data_names = [x[0] for x in data_shapes]
    type_dict = dict((key, value.dtype) for (key, value) in self.arg_params.items())
    for x in X.provide_data + X.provide_label:
        if isinstance(x, DataDesc):
            type_dict[x.name] = x.dtype
        else:
            type_dict[x[0]] = mx_real_t

and this point:

        for o_list, o_nd in zip(output_list, self._pred_exec.outputs[0:real_size]):
            o_list.append(o_nd.asnumpy())

Now, everything working well with me.