ValueError: Negative dimension size caused by subtracting 3 from 1 for 'dueling_dqn/conv1/Conv2D' (op: 'Conv2D') with input shapes: [?,1,14,32], [3,3,32,32].

[ouyangzhuzhu]

here I have quesion below when I run dqn for defend the center senario: How can I solve it.... Traceback (most recent call last): File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1567, in _create_c_op c_op = c_api.TF_FinishOperation(op_desc) tensorflow.python.framework.errors_impl.InvalidArgumentError: Negative dimension size caused by subtracting 3 from 1 for 'dueling_dqn/conv1/Conv2D' (op: 'Conv2D') with input shapes: [?,1,14,32], [3,3,32,32].

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "./train_dqn.py", line 6, in train_dqn() File "/home/lab/wsy/deep_rl_vizdoom/_train_test.py", line 74, in train_dqn dqn = DQN(model_savefile=model_savefile, settings) File "/home/lab/wsy/deep_rl_vizdoom/_dqn_algo.py", line 80, in init settings) File "/home/lab/wsy/deep_rl_vizdoom/networks/dqn.py", line 122, in init super(DuelingDQNNet, self).init(*args, kwargs) File "/home/lab/wsy/deep_rl_vizdoom/networks/dqn.py", line 44, in init self._prepare_ops() File "/home/lab/wsy/deep_rl_vizdoom/networks/dqn.py", line 55, in _prepare_ops name_scope=self._name_scope) File "/home/lab/wsy/deep_rl_vizdoom/networks/dqn.py", line 130, in create_architecture fc_input = self.get_input_layers(img_input, misc_input, name_scope) File "/home/lab/wsy/deep_rl_vizdoom/networks/common.py", line 72, in get_input_layers self.conv_strides) File "/home/lab/wsy/deep_rl_vizdoom/networks/common.py", line 20, in create_conv_layers padding="VALID", scope=name_scope + "/conv" + str(i)) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/contrib/framework/python/ops/arg_scope.py", line 183, in func_with_args return func(*args, *current_args) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/contrib/layers/python/layers/layers.py", line 1049, in convolution outputs = layer.apply(inputs) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/layers/base.py", line 828, in apply return self.call(inputs, args, kwargs) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/layers/base.py", line 717, in call outputs = self.call(inputs, *args, **kwargs) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/layers/convolutional.py", line 168, in call outputs = self._convolution_op(inputs, self.kernel) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py", line 868, in call return self.conv_op(inp, filter) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py", line 520, in call return self.call(inp, filter) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/ops/nn_ops.py", line 204, in call name=self.name) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/ops/gen_nn_ops.py", line 956, in conv2d data_format=data_format, dilations=dilations, name=name) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper op_def=op_def) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3392, in create_op op_def=op_def) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1734, in init control_input_ops) File "/home/d3alg/.local/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1570, in _create_c_op raise ValueError(str(e)) ValueError: Negative dimension size caused by subtracting 3 from 1 for 'dueling_dqn/conv1/Conv2D' (op: 'Conv2D') with input shapes: [?,1,14,32], [3,3,32,32].

[ouyangzhuzhu]

here is my command: ./train_dqn.py -s settings/examples/defend_the_center_dqn.yml

and the full context of my yml is: config_file: defend_the_center.cfg scenario_tag: defend_cent

resolution: [40, 30] conv_filters_num: [ 32, 32 ] conv_filters_sizes: [ 4, 3 ] conv_strides: [ 2, 1 ]

General training options

epsilon_decay_steps: 2e05

train_steps_per_epoch: 5000 frozen_steps: 2000

run_tag: dqn_defend_the_center use_misc: false

[Miffyli]

Did you modify the resolution and/or filter sizes/strides? Convolutions do not work with all kinds of combinations of these, and in this case (I think) you try to convolve over smaller image than the size of filter. If these were default settings, then something must have changed internally in Tensorflow that broke the code.

[ouyangzhuzhu]

3ks Miffyli, this problem slovede like this , modifying the yml from : conv_filters_num: [ 32, 32 ] conv_filters_sizes: [ 4, 3 ] conv_strides: [ 2, 1 ]

to: conv_filters_num: [ 32 ] conv_filters_sizes: [ 4 ] conv_strides: [ 2 ]

because in the code there is only one conv-level for dqn , i guss there is some special method when in a3c.

In common.py:

def get_input_layers(self, img_input=None, misc_input=None, name_scope=None): ..... conv_layers = create_conv_layers(img_input, name_scope, self.conv_filters_num, self.conv_filters_sizes, self.conv_strides) if self.use_misc: if misc_input is None: misc_input = self.vars.state_misc fc_input = tf.concat(values=[conv_layers, misc_input], axis=1) else: (the misc is none in my defent_the_center_dqn.yml) fc_input = conv_layers