Pytorch - AttributeError: 'tuple' object has no attribute 'dim'

Question

I am trying to use this architecture:

class Net(BaseFeaturesExtractor):

    def __init__(self, observation_space: gym.spaces.Box, features_dim: int = 512):
        super(Net, self).__init__(observation_space, features_dim)

        self.conv1 = nn.Conv2d(1, 64, 3, stride=1, padding=1)
        self.conv2 = nn.Conv2d(64, 64, 3, stride=1, padding=1)
        self.conv3 = nn.Conv2d(64, 64, 3, stride=1)
        self.conv4 = nn.Conv2d(64, 64, 3, stride=1)

        self.bn1 = nn.BatchNorm2d(64)
        self.bn2 = nn.BatchNorm2d(64)
        self.bn3 = nn.BatchNorm2d(64)
        self.bn4 = nn.BatchNorm2d(64)

        self.fc1 = nn.Linear(64 * (7 - 4) * (6 - 4), 128)
        self.fc_bn1 = nn.BatchNorm1d(128)

        self.fc2 = nn.Linear(128, 64)
        self.fc_bn2 = nn.BatchNorm1d(64)

        self.fc3 = nn.Linear(64, 7)

        self.fc4 = nn.Linear(64, 1)


    def forward(self, s):
        #                                                            s: batch_size x board_x x board_y
        s = s.view(-1, 1, 7, 6)  # batch_size x 1 x board_x x board_y
        s = F.relu(self.bn1(self.conv1(s)))  # batch_size x num_channels x board_x x board_y
        s = F.relu(self.bn2(self.conv2(s)))  # batch_size x num_channels x board_x x board_y
        s = F.relu(self.bn3(self.conv3(s)))  # batch_size x num_channels x (board_x-2) x (board_y-2)
        s = F.relu(self.bn4(self.conv4(s)))  # batch_size x num_channels x (board_x-4) x (board_y-4)
        s = s.view(-1,64 * (7 - 4) * (6 - 4))

        s = F.dropout(
            F.relu(self.fc_bn1(self.fc1(s))),
            p=0.3,
            training=self.training)  # batch_size x 128
        s = F.dropout(
            F.relu(self.fc_bn2(self.fc2(s))),
            p=0.3,
            training=self.training)  # batch_size x 64

        pi = self.fc3(s)  # batch_size x action_size
        v = self.fc4(s)  # batch_size x 1

        return F.log_softmax(pi, dim=1), th.tanh(v)

When I am trying to use this architecture, I am getting following error:

Traceback (most recent call last):
  File "/Users/joe/Documents/JUPYTER/ConnectX/training3.py", line 130, in <module>
    learner.learn(total_timesteps=iterations, callback=eval_callback)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/stable_baselines3/ppo/ppo.py", line 264, in learn
    reset_num_timesteps=reset_num_timesteps,
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/stable_baselines3/common/on_policy_algorithm.py", line 222, in learn
    continue_training = self.collect_rollouts(self.env, callback, self.rollout_buffer, n_rollout_steps=self.n_steps)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/stable_baselines3/common/on_policy_algorithm.py", line 154, in collect_rollouts
    actions, values, log_probs = self.policy.forward(obs_tensor)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/stable_baselines3/common/policies.py", line 545, in forward
    latent_pi, latent_vf, latent_sde = self._get_latent(obs)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/stable_baselines3/common/policies.py", line 564, in _get_latent
    latent_pi, latent_vf = self.mlp_extractor(features)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/stable_baselines3/common/torch_layers.py", line 220, in forward
    shared_latent = self.shared_net(features)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/torch/nn/modules/container.py", line 117, in forward
    input = module(input)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/torch/nn/modules/linear.py", line 93, in forward
    return F.linear(input, self.weight, self.bias)
  File "/Library/Frameworks/Python.framework/Versions/3.7/lib/python3.7/site-packages/torch/nn/functional.py", line 1688, in linear
    if input.dim() == 2 and bias is not None:
AttributeError: 'tuple' object has no attribute 'dim'

How this problem can be fixed?

Answer 1

I tried to reproduce a small working code based on the class definitions given by you and I was able to get the outputs from the model. Here is the following code:

# BaseFeaturesExtractor class
import gym
import torch as th
from torch import nn

class BaseFeaturesExtractor(nn.Module):
    """
    Base class that represents a features extractor.
    :param observation_space:
    :param features_dim: Number of features extracted.
    """

    def __init__(self, observation_space: gym.Space, features_dim: int = 0):
        super(BaseFeaturesExtractor, self).__init__()
        assert features_dim > 0
        self._observation_space = observation_space
        self._features_dim = features_dim

    @property
    def features_dim(self) -> int:
        return self._features_dim

    def forward(self, observations: th.Tensor) -> th.Tensor:
        raise NotImplementedError()

# Net class
class Net(BaseFeaturesExtractor):

    def __init__(self, observation_space: gym.spaces.Box, features_dim: int = 512):
        super(Net, self).__init__(observation_space, features_dim)

        self.conv1 = nn.Conv2d(1, 64, 3, stride=1, padding=1)
        self.conv2 = nn.Conv2d(64, 64, 3, stride=1, padding=1)
        self.conv3 = nn.Conv2d(64, 64, 3, stride=1)
        self.conv4 = nn.Conv2d(64, 64, 3, stride=1)

        self.bn1 = nn.BatchNorm2d(64)
        self.bn2 = nn.BatchNorm2d(64)
        self.bn3 = nn.BatchNorm2d(64)
        self.bn4 = nn.BatchNorm2d(64)

        self.fc1 = nn.Linear(64 * (7 - 4) * (6 - 4), 128)
        self.fc_bn1 = nn.BatchNorm1d(128)

        self.fc2 = nn.Linear(128, 64)
        self.fc_bn2 = nn.BatchNorm1d(64)

        self.fc3 = nn.Linear(64, 7)

        self.fc4 = nn.Linear(64, 1)


    def forward(self, s):
        #                                                            s: batch_size x board_x x board_y
        s = s.view(-1, 1, 7, 6)  # batch_size x 1 x board_x x board_y
        s = F.relu(self.bn1(self.conv1(s)))  # batch_size x num_channels x board_x x board_y
        s = F.relu(self.bn2(self.conv2(s)))  # batch_size x num_channels x board_x x board_y
        s = F.relu(self.bn3(self.conv3(s)))  # batch_size x num_channels x (board_x-2) x (board_y-2)
        s = F.relu(self.bn4(self.conv4(s)))  # batch_size x num_channels x (board_x-4) x (board_y-4)
        s = s.view(-1,64 * (7 - 4) * (6 - 4))

        s = F.dropout(
            F.relu(self.fc_bn1(self.fc1(s))),
            p=0.3,
            training=self.training)  # batch_size x 128
        s = F.dropout(
            F.relu(self.fc_bn2(self.fc2(s))),
            p=0.3,
            training=self.training)  # batch_size x 64

        pi = self.fc3(s)  # batch_size x action_size
        v = self.fc4(s)  # batch_size x 1

        return F.log_softmax(pi, dim=1), th.tanh(v)

# Minimal code to reproduce a forward pass

import numpy as np
import torch
import torch.nn.functional as F

params = gym.spaces.Box(np.array([-1,0,0]), np.array([+1,+1,+1]))
model = Net(params)
inputs = torch.randn(2, 1, 7, 6)
outputs = model(inputs)
print(outputs[0].shape, outputs[1].shape) # prints (torch.Size([2, 7]), torch.Size([2, 1]))