zjowowen's picture
init space
079c32c
raw
history blame
12.5 kB
from typing import Dict, Union
import torch
import torch.nn as nn
from functools import reduce
from ding.torch_utils import one_hot, MLP
from ding.utils import squeeze, list_split, MODEL_REGISTRY, SequenceType
from .q_learning import DRQN
class COMAActorNetwork(nn.Module):
"""
Overview:
Decentralized actor network in COMA algorithm.
Interface:
``__init__``, ``forward``
"""
def __init__(
self,
obs_shape: int,
action_shape: int,
hidden_size_list: SequenceType = [128, 128, 64],
):
"""
Overview:
Initialize COMA actor network
Arguments:
- obs_shape (:obj:`int`): the dimension of each agent's observation state
- action_shape (:obj:`int`): the dimension of action shape
- hidden_size_list (:obj:`list`): the list of hidden size, default to [128, 128, 64]
"""
super(COMAActorNetwork, self).__init__()
self.main = DRQN(obs_shape, action_shape, hidden_size_list)
def forward(self, inputs: Dict) -> Dict:
"""
Overview:
The forward computation graph of COMA actor network
Arguments:
- inputs (:obj:`dict`): input data dict with keys ['obs', 'prev_state']
- agent_state (:obj:`torch.Tensor`): each agent local state(obs)
- action_mask (:obj:`torch.Tensor`): the masked action
- prev_state (:obj:`torch.Tensor`): the previous hidden state
Returns:
- output (:obj:`dict`): output data dict with keys ['logit', 'next_state', 'action_mask']
ArgumentsKeys:
- necessary: ``obs`` { ``agent_state``, ``action_mask`` }, ``prev_state``
ReturnsKeys:
- necessary: ``logit``, ``next_state``, ``action_mask``
Examples:
>>> T, B, A, N = 4, 8, 3, 32
>>> embedding_dim = 64
>>> action_dim = 6
>>> data = torch.randn(T, B, A, N)
>>> model = COMAActorNetwork((N, ), action_dim, [128, embedding_dim])
>>> prev_state = [[None for _ in range(A)] for _ in range(B)]
>>> for t in range(T):
>>> inputs = {'obs': {'agent_state': data[t], 'action_mask': None}, 'prev_state': prev_state}
>>> outputs = model(inputs)
>>> logit, prev_state = outputs['logit'], outputs['next_state']
"""
agent_state = inputs['obs']['agent_state']
prev_state = inputs['prev_state']
if len(agent_state.shape) == 3: # B, A, N
agent_state = agent_state.unsqueeze(0)
unsqueeze_flag = True
else:
unsqueeze_flag = False
T, B, A = agent_state.shape[:3]
agent_state = agent_state.reshape(T, -1, *agent_state.shape[3:])
prev_state = reduce(lambda x, y: x + y, prev_state)
output = self.main({'obs': agent_state, 'prev_state': prev_state, 'enable_fast_timestep': True})
logit, next_state = output['logit'], output['next_state']
next_state, _ = list_split(next_state, step=A)
logit = logit.reshape(T, B, A, -1)
if unsqueeze_flag:
logit = logit.squeeze(0)
return {'logit': logit, 'next_state': next_state, 'action_mask': inputs['obs']['action_mask']}
class COMACriticNetwork(nn.Module):
"""
Overview:
Centralized critic network in COMA algorithm.
Interface:
``__init__``, ``forward``
"""
def __init__(
self,
input_size: int,
action_shape: int,
hidden_size: int = 128,
):
"""
Overview:
initialize COMA critic network
Arguments:
- input_size (:obj:`int`): the size of input global observation
- action_shape (:obj:`int`): the dimension of action shape
- hidden_size_list (:obj:`list`): the list of hidden size, default to 128
Returns:
- output (:obj:`dict`): output data dict with keys ['q_value']
Shapes:
- obs (:obj:`dict`): ``agent_state``: :math:`(T, B, A, N, D)`, ``action_mask``: :math:`(T, B, A, N, A)`
- prev_state (:obj:`list`): :math:`[[[h, c] for _ in range(A)] for _ in range(B)]`
- logit (:obj:`torch.Tensor`): :math:`(T, B, A, N, A)`
- next_state (:obj:`list`): :math:`[[[h, c] for _ in range(A)] for _ in range(B)]`
- action_mask (:obj:`torch.Tensor`): :math:`(T, B, A, N, A)`
"""
super(COMACriticNetwork, self).__init__()
self.action_shape = action_shape
self.act = nn.ReLU()
self.mlp = nn.Sequential(
MLP(input_size, hidden_size, hidden_size, 2, activation=self.act), nn.Linear(hidden_size, action_shape)
)
def forward(self, data: Dict) -> Dict:
"""
Overview:
forward computation graph of qmix network
Arguments:
- data (:obj:`dict`): input data dict with keys ['obs', 'prev_state', 'action']
- agent_state (:obj:`torch.Tensor`): each agent local state(obs)
- global_state (:obj:`torch.Tensor`): global state(obs)
- action (:obj:`torch.Tensor`): the masked action
ArgumentsKeys:
- necessary: ``obs`` { ``agent_state``, ``global_state`` }, ``action``, ``prev_state``
ReturnsKeys:
- necessary: ``q_value``
Examples:
>>> agent_num, bs, T = 4, 3, 8
>>> obs_dim, global_obs_dim, action_dim = 32, 32 * 4, 9
>>> coma_model = COMACriticNetwork(
>>> obs_dim - action_dim + global_obs_dim + 2 * action_dim * agent_num, action_dim)
>>> data = {
>>> 'obs': {
>>> 'agent_state': torch.randn(T, bs, agent_num, obs_dim),
>>> 'global_state': torch.randn(T, bs, global_obs_dim),
>>> },
>>> 'action': torch.randint(0, action_dim, size=(T, bs, agent_num)),
>>> }
>>> output = coma_model(data)
"""
x = self._preprocess_data(data)
q = self.mlp(x)
return {'q_value': q}
def _preprocess_data(self, data: Dict) -> torch.Tensor:
"""
Overview:
preprocess data to make it can be used by MLP net
Arguments:
- data (:obj:`dict`): input data dict with keys ['obs', 'prev_state', 'action']
- agent_state (:obj:`torch.Tensor`): each agent local state(obs)
- global_state (:obj:`torch.Tensor`): global state(obs)
- action (:obj:`torch.Tensor`): the masked action
ArgumentsKeys:
- necessary: ``obs`` { ``agent_state``, ``global_state``} , ``action``, ``prev_state``
Return:
- x (:obj:`torch.Tensor`): the data can be used by MLP net, including \
``global_state``, ``agent_state``, ``last_action``, ``action``, ``agent_id``
"""
t_size, batch_size, agent_num = data['obs']['agent_state'].shape[:3]
agent_state_ori, global_state = data['obs']['agent_state'], data['obs']['global_state']
# splite obs, last_action and agent_id
agent_state = agent_state_ori[..., :-self.action_shape - agent_num]
last_action = agent_state_ori[..., -self.action_shape - agent_num:-agent_num]
last_action = last_action.reshape(t_size, batch_size, 1, -1).repeat(1, 1, agent_num, 1)
agent_id = agent_state_ori[..., -agent_num:]
action = one_hot(data['action'], self.action_shape) # T, B, A,N
action = action.reshape(t_size, batch_size, -1, agent_num * self.action_shape).repeat(1, 1, agent_num, 1)
action_mask = (1 - torch.eye(agent_num).to(action.device))
action_mask = action_mask.view(-1, 1).repeat(1, self.action_shape).view(agent_num, -1) # A, A*N
action = (action_mask.unsqueeze(0).unsqueeze(0)) * action # T, B, A, A*N
global_state = global_state.unsqueeze(2).repeat(1, 1, agent_num, 1)
x = torch.cat([global_state, agent_state, last_action, action, agent_id], -1)
return x
@MODEL_REGISTRY.register('coma')
class COMA(nn.Module):
"""
Overview:
The network of COMA algorithm, which is QAC-type actor-critic.
Interface:
``__init__``, ``forward``
Properties:
- mode (:obj:`list`): The list of forward mode, including ``compute_actor`` and ``compute_critic``
"""
mode = ['compute_actor', 'compute_critic']
def __init__(
self, agent_num: int, obs_shape: Dict, action_shape: Union[int, SequenceType],
actor_hidden_size_list: SequenceType
) -> None:
"""
Overview:
initialize COMA network
Arguments:
- agent_num (:obj:`int`): the number of agent
- obs_shape (:obj:`Dict`): the observation information, including agent_state and \
global_state
- action_shape (:obj:`Union[int, SequenceType]`): the dimension of action shape
- actor_hidden_size_list (:obj:`SequenceType`): the list of hidden size
"""
super(COMA, self).__init__()
action_shape = squeeze(action_shape)
actor_input_size = squeeze(obs_shape['agent_state'])
critic_input_size = squeeze(obs_shape['agent_state']) + squeeze(obs_shape['global_state']) + \
agent_num * action_shape + (agent_num - 1) * action_shape
critic_hidden_size = actor_hidden_size_list[-1]
self.actor = COMAActorNetwork(actor_input_size, action_shape, actor_hidden_size_list)
self.critic = COMACriticNetwork(critic_input_size, action_shape, critic_hidden_size)
def forward(self, inputs: Dict, mode: str) -> Dict:
"""
Overview:
forward computation graph of COMA network
Arguments:
- inputs (:obj:`dict`): input data dict with keys ['obs', 'prev_state', 'action']
- agent_state (:obj:`torch.Tensor`): each agent local state(obs)
- global_state (:obj:`torch.Tensor`): global state(obs)
- action (:obj:`torch.Tensor`): the masked action
ArgumentsKeys:
- necessary: ``obs`` { ``agent_state``, ``global_state``, ``action_mask`` }, ``action``, ``prev_state``
ReturnsKeys:
- necessary:
- compute_critic: ``q_value``
- compute_actor: ``logit``, ``next_state``, ``action_mask``
Shapes:
- obs (:obj:`dict`): ``agent_state``: :math:`(T, B, A, N, D)`, ``action_mask``: :math:`(T, B, A, N, A)`
- prev_state (:obj:`list`): :math:`[[[h, c] for _ in range(A)] for _ in range(B)]`
- logit (:obj:`torch.Tensor`): :math:`(T, B, A, N, A)`
- next_state (:obj:`list`): :math:`[[[h, c] for _ in range(A)] for _ in range(B)]`
- action_mask (:obj:`torch.Tensor`): :math:`(T, B, A, N, A)`
- q_value (:obj:`torch.Tensor`): :math:`(T, B, A, N, A)`
Examples:
>>> agent_num, bs, T = 4, 3, 8
>>> agent_num, bs, T = 4, 3, 8
>>> obs_dim, global_obs_dim, action_dim = 32, 32 * 4, 9
>>> coma_model = COMA(
>>> agent_num=agent_num,
>>> obs_shape=dict(agent_state=(obs_dim, ), global_state=(global_obs_dim, )),
>>> action_shape=action_dim,
>>> actor_hidden_size_list=[128, 64],
>>> )
>>> prev_state = [[None for _ in range(agent_num)] for _ in range(bs)]
>>> data = {
>>> 'obs': {
>>> 'agent_state': torch.randn(T, bs, agent_num, obs_dim),
>>> 'action_mask': None,
>>> },
>>> 'prev_state': prev_state,
>>> }
>>> output = coma_model(data, mode='compute_actor')
>>> data= {
>>> 'obs': {
>>> 'agent_state': torch.randn(T, bs, agent_num, obs_dim),
>>> 'global_state': torch.randn(T, bs, global_obs_dim),
>>> },
>>> 'action': torch.randint(0, action_dim, size=(T, bs, agent_num)),
>>> }
>>> output = coma_model(data, mode='compute_critic')
"""
assert mode in self.mode, "not support forward mode: {}/{}".format(mode, self.mode)
if mode == 'compute_actor':
return self.actor(inputs)
elif mode == 'compute_critic':
return self.critic(inputs)