Created by: Termset

https://github.com/PaddlePaddle/PARL/issues/91 这个issue也提到了构建ｌｓｔｍ的问题

我简单尝试了，在IMPALA的model中加入lstm网络出现了错误

class AtariModel(parl.Model):
    def __init__(self, act_dim):

        self.conv1 = layers.conv2d(
            num_filters=16, filter_size=4, stride=2, padding=1, act='relu')
        self.conv2 = layers.conv2d(
            num_filters=32, filter_size=4, stride=2, padding=2, act='relu')
        self.conv3 = layers.conv2d(
            num_filters=256, filter_size=11, stride=1, padding=0, act='relu')
        self.lstm_cell = layers.LSTMCell(hidden_size=256)
        self.policy_conv = layers.conv2d(
            num_filters=act_dim,
            filter_size=1,
            stride=1,
            padding=0,
            act=None,
            param_attr=ParamAttr(initializer=fluid.initializer.Normal()))
        self.fc = layers.fc(size=256, act='relu')
        self.action_fc = layers.fc(size=act_dim)
        self.value_fc = layers.fc(
            size=1,
            param_attr=ParamAttr(initializer=fluid.initializer.Normal()))

    def policy(self, obs):
        """
        Args:
            obs: A float32 tensor of shape [B, C, H, W]
        Returns:
            policy_logits: B * ACT_DIM
        """
        obs = obs / 255.0
        conv1 = self.conv1(obs)
        conv2 = self.conv2(conv1)
        conv3 = self.conv3(conv2)  # (-1, 256, 1, 1)
        flattened = layers.flatten(conv3, axis=1)  # (-1, 256)
        flattened = layers.unsqueeze(flattened, axes=-1)
        lstm_out = layers.rnn(cell=self.lstm_cell, inputs=flattened)
        policy_fc = self.fc(lstm_out)
        policy_logits = self.action_fc(policy_fc)
        # policy_conv = self.policy_conv(conv3)  # (-1, 6, 1, 1)
        # policy_logits = layers.flatten(policy_conv, axis=1)  # (-1, 6)
        
        return policy_logits

Traceback (most recent call last):
  File "train.py", line 255, in <module>
    learner = Learner(config)
  File "train.py", line 56, in __init__
    self.learn_data_provider)
  File "/home/tianqi/PARL/examples/IMPALA/atari_agent.py", line 29, in __init__
    super(AtariAgent, self).__init__(algorithm)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/parl/core/fluid/agent.py", line 79, in __init__
    self.build_program()
  File "/home/tianqi/PARL/examples/IMPALA/atari_agent.py", line 43, in build_program
    self.sample_actions, self.behaviour_logits = self.alg.sample(obs)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/parl/algorithms/fluid/impala/impala.py", line 209, in sample
    logits = self.model.policy(obs)
  File "/home/tianqi/PARL/examples/IMPALA/atari_model.py", line 57, in policy
    lstm_out = layers.rnn(cell=self.lstm_cell, inputs=flattened)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/layers/rnn.py", line 551, in rnn
    outputs, new_states = cell.call(inputs, copy_states, **kwargs)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/layers/rnn.py", line 416, in call
    new_hidden, new_cell = self.lstm_unit(inputs, pre_hidden, pre_cell)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/dygraph/layers.py", line 461, in __call__
    outputs = self.forward(*inputs, **kwargs)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/contrib/layers/rnn_impl.py", line 820, in forward
    layers.sigmoid(layers.elementwise_add(f, self._forget_bias))),
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/layers/nn.py", line 12416, in elementwise_add
    return _elementwise_op(LayerHelper('elementwise_add', **locals()))
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/layers/nn.py", line 12237, in _elementwise_op
    'use_mkldnn': use_mkldnn})
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/layer_helper.py", line 43, in append_op
    return self.main_program.current_block().append_op(*args, **kwargs)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/framework.py", line 2610, in append_op
    attrs=kwargs.get("attrs", None))
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/framework.py", line 1971, in __init__
    self.desc.infer_shape(self.block.desc)
paddle.fluid.core_avx.EnforceNotMet: 

--------------------------------------------
C++ Call Stacks (More useful to developers):
--------------------------------------------
0   std::string paddle::platform::GetTraceBackString<std::string const&>(std::string const&, char const*, int)
1   paddle::platform::EnforceNotMet::EnforceNotMet(std::string const&, char const*, int)
2   paddle::operators::ElementwiseOp::InferShape(paddle::framework::InferShapeContext*) const
3   paddle::framework::OpDesc::InferShape(paddle::framework::BlockDesc const&) const

------------------------------------------
Python Call Stacks (More useful to users):
------------------------------------------
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/framework.py", line 2610, in append_op
    attrs=kwargs.get("attrs", None))
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/layer_helper.py", line 43, in append_op
    return self.main_program.current_block().append_op(*args, **kwargs)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/layers/nn.py", line 12237, in _elementwise_op
    'use_mkldnn': use_mkldnn})
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/layers/nn.py", line 12416, in elementwise_add
    return _elementwise_op(LayerHelper('elementwise_add', **locals()))
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/contrib/layers/rnn_impl.py", line 820, in forward
    layers.sigmoid(layers.elementwise_add(f, self._forget_bias))),
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/dygraph/layers.py", line 461, in __call__
    outputs = self.forward(*inputs, **kwargs)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/layers/rnn.py", line 416, in call
    new_hidden, new_cell = self.lstm_unit(inputs, pre_hidden, pre_cell)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/paddle/fluid/layers/rnn.py", line 551, in rnn
    outputs, new_states = cell.call(inputs, copy_states, **kwargs)
  File "/home/tianqi/PARL/examples/IMPALA/atari_model.py", line 57, in policy
    lstm_out = layers.rnn(cell=self.lstm_cell, inputs=flattened)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/parl/algorithms/fluid/impala/impala.py", line 209, in sample
    logits = self.model.policy(obs)
  File "/home/tianqi/PARL/examples/IMPALA/atari_agent.py", line 43, in build_program
    self.sample_actions, self.behaviour_logits = self.alg.sample(obs)
  File "/home/tianqi/anaconda3/lib/python3.6/site-packages/parl/core/fluid/agent.py", line 79, in __init__
    self.build_program()
  File "/home/tianqi/PARL/examples/IMPALA/atari_agent.py", line 29, in __init__
    super(AtariAgent, self).__init__(algorithm)
  File "train.py", line 56, in __init__
    self.learn_data_provider)
  File "train.py", line 255, in <module>
    learner = Learner(config)

----------------------
Error Message Summary:
----------------------
NotFoundError: No Input(Y) found for ElementwiseOp operator.
  [Hint: Expected ctx->HasInput("Y") == true, but received ctx->HasInput("Y"):0 != true:1.] at (/paddle/paddle/fluid/operators/elementwise/elementwise_op.h:42)
  [operator < elementwise_add > error]