API tools faq
paste
Advertisement
Guest User

Untitled

a guest
Jun 19th, 2019
89
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 15.76 KB | None | 0 0
raw download clone embed print report
  1. Traceback (most recent call last):
  2. File "/current_project/robo-planet/torch_planet/main.py", line 117, in <module>
  3. transition_model = TransitionModel(args.belief_size, args.state_size, env.action_size, args.hidden_size, args.embedding_size, args.activation_function).to(device=args.device)
  4. File "/home/maximecb/.local/lib/python3.6/site-packages/torch/jit/__init__.py", line 952, in init_then_register
  5. _create_methods_from_stubs(self, methods)
  6. File "/home/maximecb/.local/lib/python3.6/site-packages/torch/jit/__init__.py", line 913, in _create_methods_from_stubs
  7. self._create_methods(defs, rcbs, defaults)
  8. RuntimeError:
  9. arguments for call are not valid:
  10.  
  11. for operator aten::mul(Tensor self, Tensor other) -> Tensor:
  12. expected a value of type Tensor for argument 'self' but found Tensor[]
  13. @jit.script_method
  14. def forward(self, prev_state:torch.Tensor, actions:torch.Tensor, prev_belief:torch.Tensor, observations:Optional[torch.Tensor]=None, nonterminals:Optional[torch.Tensor]=None) -> List[torch.Tensor]:
  15. # Create lists for hidden states (cannot use single tensor as buffer because autograd won't work with inplace writes)
  16. T = actions.size(0) + 1
  17. beliefs, prior_states, prior_means, prior_std_devs, posterior_states, posterior_means, posterior_std_devs = [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T
  18. ~~~~~~~~~~~~~~ <--- HERE
  19. beliefs[0], prior_states[0], posterior_states[0] = prev_belief, prev_state, prev_state
  20. # Loop over time sequence
  21. for t in range(T - 1):
  22. _state = prior_states[t] if observations is None else posterior_states[t] # Select appropriate previous state
  23. _state = _state if nonterminals is None else _state * nonterminals[t] # Mask if previous transition was terminal
  24. # Compute belief (deterministic hidden state)
  25. hidden = self.act_fn(self.fc_embed_state_action(torch.cat([_state, actions[t]], dim=1)))
  26. beliefs[t + 1] = self.rnn(hidden, beliefs[t])
  27. # Compute state prior by applying transition dynamics
  28. for operator aten::mul(Tensor self, Scalar other) -> Tensor:
  29. expected a value of type Tensor for argument 'self' but found Tensor[]
  30. @jit.script_method
  31. def forward(self, prev_state:torch.Tensor, actions:torch.Tensor, prev_belief:torch.Tensor, observations:Optional[torch.Tensor]=None, nonterminals:Optional[torch.Tensor]=None) -> List[torch.Tensor]:
  32. # Create lists for hidden states (cannot use single tensor as buffer because autograd won't work with inplace writes)
  33. T = actions.size(0) + 1
  34. beliefs, prior_states, prior_means, prior_std_devs, posterior_states, posterior_means, posterior_std_devs = [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T
  35. ~~~~~~~~~~~~~~ <--- HERE
  36. beliefs[0], prior_states[0], posterior_states[0] = prev_belief, prev_state, prev_state
  37. # Loop over time sequence
  38. for t in range(T - 1):
  39. _state = prior_states[t] if observations is None else posterior_states[t] # Select appropriate previous state
  40. _state = _state if nonterminals is None else _state * nonterminals[t] # Mask if previous transition was terminal
  41. # Compute belief (deterministic hidden state)
  42. hidden = self.act_fn(self.fc_embed_state_action(torch.cat([_state, actions[t]], dim=1)))
  43. beliefs[t + 1] = self.rnn(hidden, beliefs[t])
  44. # Compute state prior by applying transition dynamics
  45. for operator aten::mul(Tensor self, Tensor other, *, Tensor out) -> Tensor:
  46. expected a value of type Tensor for argument 'self' but found Tensor[]
  47. @jit.script_method
  48. def forward(self, prev_state:torch.Tensor, actions:torch.Tensor, prev_belief:torch.Tensor, observations:Optional[torch.Tensor]=None, nonterminals:Optional[torch.Tensor]=None) -> List[torch.Tensor]:
  49. # Create lists for hidden states (cannot use single tensor as buffer because autograd won't work with inplace writes)
  50. T = actions.size(0) + 1
  51. beliefs, prior_states, prior_means, prior_std_devs, posterior_states, posterior_means, posterior_std_devs = [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T
  52. ~~~~~~~~~~~~~~ <--- HERE
  53. beliefs[0], prior_states[0], posterior_states[0] = prev_belief, prev_state, prev_state
  54. # Loop over time sequence
  55. for t in range(T - 1):
  56. _state = prior_states[t] if observations is None else posterior_states[t] # Select appropriate previous state
  57. _state = _state if nonterminals is None else _state * nonterminals[t] # Mask if previous transition was terminal
  58. # Compute belief (deterministic hidden state)
  59. hidden = self.act_fn(self.fc_embed_state_action(torch.cat([_state, actions[t]], dim=1)))
  60. beliefs[t + 1] = self.rnn(hidden, beliefs[t])
  61. # Compute state prior by applying transition dynamics
  62. for operator aten::mul(int a, int b) -> int:
  63. expected a value of type int for argument 'a' but found Tensor[]
  64. @jit.script_method
  65. def forward(self, prev_state:torch.Tensor, actions:torch.Tensor, prev_belief:torch.Tensor, observations:Optional[torch.Tensor]=None, nonterminals:Optional[torch.Tensor]=None) -> List[torch.Tensor]:
  66. # Create lists for hidden states (cannot use single tensor as buffer because autograd won't work with inplace writes)
  67. T = actions.size(0) + 1
  68. beliefs, prior_states, prior_means, prior_std_devs, posterior_states, posterior_means, posterior_std_devs = [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T
  69. ~~~~~~~~~~~~~~ <--- HERE
  70. beliefs[0], prior_states[0], posterior_states[0] = prev_belief, prev_state, prev_state
  71. # Loop over time sequence
  72. for t in range(T - 1):
  73. _state = prior_states[t] if observations is None else posterior_states[t] # Select appropriate previous state
  74. _state = _state if nonterminals is None else _state * nonterminals[t] # Mask if previous transition was terminal
  75. # Compute belief (deterministic hidden state)
  76. hidden = self.act_fn(self.fc_embed_state_action(torch.cat([_state, actions[t]], dim=1)))
  77. beliefs[t + 1] = self.rnn(hidden, beliefs[t])
  78. # Compute state prior by applying transition dynamics
  79. for operator aten::mul(float a, float b) -> float:
  80. expected a value of type float for argument 'a' but found Tensor[]
  81. @jit.script_method
  82. def forward(self, prev_state:torch.Tensor, actions:torch.Tensor, prev_belief:torch.Tensor, observations:Optional[torch.Tensor]=None, nonterminals:Optional[torch.Tensor]=None) -> List[torch.Tensor]:
  83. # Create lists for hidden states (cannot use single tensor as buffer because autograd won't work with inplace writes)
  84. T = actions.size(0) + 1
  85. beliefs, prior_states, prior_means, prior_std_devs, posterior_states, posterior_means, posterior_std_devs = [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T
  86. ~~~~~~~~~~~~~~ <--- HERE
  87. beliefs[0], prior_states[0], posterior_states[0] = prev_belief, prev_state, prev_state
  88. # Loop over time sequence
  89. for t in range(T - 1):
  90. _state = prior_states[t] if observations is None else posterior_states[t] # Select appropriate previous state
  91. _state = _state if nonterminals is None else _state * nonterminals[t] # Mask if previous transition was terminal
  92. # Compute belief (deterministic hidden state)
  93. hidden = self.act_fn(self.fc_embed_state_action(torch.cat([_state, actions[t]], dim=1)))
  94. beliefs[t + 1] = self.rnn(hidden, beliefs[t])
  95. # Compute state prior by applying transition dynamics
  96. for operator aten::mul(int a, float b) -> float:
  97. expected a value of type int for argument 'a' but found Tensor[]
  98. @jit.script_method
  99. def forward(self, prev_state:torch.Tensor, actions:torch.Tensor, prev_belief:torch.Tensor, observations:Optional[torch.Tensor]=None, nonterminals:Optional[torch.Tensor]=None) -> List[torch.Tensor]:
  100. # Create lists for hidden states (cannot use single tensor as buffer because autograd won't work with inplace writes)
  101. T = actions.size(0) + 1
  102. beliefs, prior_states, prior_means, prior_std_devs, posterior_states, posterior_means, posterior_std_devs = [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T
  103. ~~~~~~~~~~~~~~ <--- HERE
  104. beliefs[0], prior_states[0], posterior_states[0] = prev_belief, prev_state, prev_state
  105. # Loop over time sequence
  106. for t in range(T - 1):
  107. _state = prior_states[t] if observations is None else posterior_states[t] # Select appropriate previous state
  108. _state = _state if nonterminals is None else _state * nonterminals[t] # Mask if previous transition was terminal
  109. # Compute belief (deterministic hidden state)
  110. hidden = self.act_fn(self.fc_embed_state_action(torch.cat([_state, actions[t]], dim=1)))
  111. beliefs[t + 1] = self.rnn(hidden, beliefs[t])
  112. # Compute state prior by applying transition dynamics
  113. for operator aten::mul(float a, int b) -> float:
  114. expected a value of type float for argument 'a' but found Tensor[]
  115. @jit.script_method
  116. def forward(self, prev_state:torch.Tensor, actions:torch.Tensor, prev_belief:torch.Tensor, observations:Optional[torch.Tensor]=None, nonterminals:Optional[torch.Tensor]=None) -> List[torch.Tensor]:
  117. # Create lists for hidden states (cannot use single tensor as buffer because autograd won't work with inplace writes)
  118. T = actions.size(0) + 1
  119. beliefs, prior_states, prior_means, prior_std_devs, posterior_states, posterior_means, posterior_std_devs = [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T
  120. ~~~~~~~~~~~~~~ <--- HERE
  121. beliefs[0], prior_states[0], posterior_states[0] = prev_belief, prev_state, prev_state
  122. # Loop over time sequence
  123. for t in range(T - 1):
  124. _state = prior_states[t] if observations is None else posterior_states[t] # Select appropriate previous state
  125. _state = _state if nonterminals is None else _state * nonterminals[t] # Mask if previous transition was terminal
  126. # Compute belief (deterministic hidden state)
  127. hidden = self.act_fn(self.fc_embed_state_action(torch.cat([_state, actions[t]], dim=1)))
  128. beliefs[t + 1] = self.rnn(hidden, beliefs[t])
  129. # Compute state prior by applying transition dynamics
  130. for operator mul(float a, Tensor b) -> Tensor:
  131. expected a value of type float for argument 'a' but found Tensor[]
  132. @jit.script_method
  133. def forward(self, prev_state:torch.Tensor, actions:torch.Tensor, prev_belief:torch.Tensor, observations:Optional[torch.Tensor]=None, nonterminals:Optional[torch.Tensor]=None) -> List[torch.Tensor]:
  134. # Create lists for hidden states (cannot use single tensor as buffer because autograd won't work with inplace writes)
  135. T = actions.size(0) + 1
  136. beliefs, prior_states, prior_means, prior_std_devs, posterior_states, posterior_means, posterior_std_devs = [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T
  137. ~~~~~~~~~~~~~~ <--- HERE
  138. beliefs[0], prior_states[0], posterior_states[0] = prev_belief, prev_state, prev_state
  139. # Loop over time sequence
  140. for t in range(T - 1):
  141. _state = prior_states[t] if observations is None else posterior_states[t] # Select appropriate previous state
  142. _state = _state if nonterminals is None else _state * nonterminals[t] # Mask if previous transition was terminal
  143. # Compute belief (deterministic hidden state)
  144. hidden = self.act_fn(self.fc_embed_state_action(torch.cat([_state, actions[t]], dim=1)))
  145. beliefs[t + 1] = self.rnn(hidden, beliefs[t])
  146. # Compute state prior by applying transition dynamics
  147. for operator mul(int a, Tensor b) -> Tensor:
  148. expected a value of type int for argument 'a' but found Tensor[]
  149. @jit.script_method
  150. def forward(self, prev_state:torch.Tensor, actions:torch.Tensor, prev_belief:torch.Tensor, observations:Optional[torch.Tensor]=None, nonterminals:Optional[torch.Tensor]=None) -> List[torch.Tensor]:
  151. # Create lists for hidden states (cannot use single tensor as buffer because autograd won't work with inplace writes)
  152. T = actions.size(0) + 1
  153. beliefs, prior_states, prior_means, prior_std_devs, posterior_states, posterior_means, posterior_std_devs = [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T
  154. ~~~~~~~~~~~~~~ <--- HERE
  155. beliefs[0], prior_states[0], posterior_states[0] = prev_belief, prev_state, prev_state
  156. # Loop over time sequence
  157. for t in range(T - 1):
  158. _state = prior_states[t] if observations is None else posterior_states[t] # Select appropriate previous state
  159. _state = _state if nonterminals is None else _state * nonterminals[t] # Mask if previous transition was terminal
  160. # Compute belief (deterministic hidden state)
  161. hidden = self.act_fn(self.fc_embed_state_action(torch.cat([_state, actions[t]], dim=1)))
  162. beliefs[t + 1] = self.rnn(hidden, beliefs[t])
  163. # Compute state prior by applying transition dynamicsfor call at:
  164. @jit.script_method
  165. def forward(self, prev_state:torch.Tensor, actions:torch.Tensor, prev_belief:torch.Tensor, observations:Optional[torch.Tensor]=None, nonterminals:Optional[torch.Tensor]=None) -> List[torch.Tensor]:
  166. # Create lists for hidden states (cannot use single tensor as buffer because autograd won't work with inplace writes)
  167. T = actions.size(0) + 1
  168. beliefs, prior_states, prior_means, prior_std_devs, posterior_states, posterior_means, posterior_std_devs = [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T, [torch.empty(0)] * T
  169. ~~~~~~~~~~~~~~~~~~~~ <--- HERE
  170. beliefs[0], prior_states[0], posterior_states[0] = prev_belief, prev_state, prev_state
  171. # Loop over time sequence
  172. for t in range(T - 1):
  173. _state = prior_states[t] if observations is None else posterior_states[t] # Select appropriate previous state
  174. _state = _state if nonterminals is None else _state * nonterminals[t] # Mask if previous transition was terminal
  175. # Compute belief (deterministic hidden state)
  176. hidden = self.act_fn(self.fc_embed_state_action(torch.cat([_state, actions[t]], dim=1)))
  177. beliefs[t + 1] = self.rnn(hidden, beliefs[t])
  178. # Compute state prior by applying transition dynamics
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!