File size: 8,126 Bytes
55da406 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 |
# mappo.py
import torch
import torch.nn as nn
import random
import numpy as np
from torch.distributions import Normal
def set_global_seed(seed: int):
random.seed(seed) # Python
np.random.seed(seed) # NumPy
torch.manual_seed(seed) # PyTorch CPU
if torch.cuda.is_available():
torch.cuda.manual_seed_all(seed) # PyTorch GPU
# make CuDNN deterministic (may slow you down a bit):
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
# Universal device selection
if torch.cuda.is_available():
device = torch.device("cuda")
print("Using CUDA (NVIDIA GPU)")
# elif torch.backends.mps.is_available():
# device = torch.device("mps")
# print("Using MPS (Apple Silicon GPU)")
else:
device = torch.device("cpu")
print("Using CPU")
# fix EVERYTHING
SEED = 42
set_global_seed(SEED)
class MLP(nn.Module):
def __init__(self, input_dim, hidden_dims, output_dim):
super().__init__()
layers = []
last_dim = input_dim
for h in hidden_dims:
layers += [nn.Linear(last_dim, h), nn.ReLU()]
last_dim = h
layers.append(nn.Linear(last_dim, output_dim))
self.net = nn.Sequential(*layers)
def forward(self, x):
return self.net(x)
class Actor(nn.Module):
def __init__(self, obs_dim, act_dim, hidden=(64,64)):
super().__init__()
self.net = MLP(obs_dim, hidden, act_dim)
self.log_std = nn.Parameter(torch.zeros(act_dim))
def forward(self, x):
mean = self.net(x)
std = torch.exp(self.log_std)
return mean, std
class Critic(nn.Module):
def __init__(self, state_dim, hidden=(128,128)):
super().__init__()
self.net = MLP(state_dim, hidden, 1)
def forward(self, x):
return self.net(x).squeeze(-1)
class MAPPO:
def __init__(
self,
n_agents,
local_dim,
global_dim,
act_dim,
lr=3e-4,
gamma=0.99,
lam=0.95,
clip_eps=0.2,
k_epochs=10,
batch_size=1024
):
self.n_agents = n_agents
self.gamma = gamma
self.lam = lam
self.clip_eps = clip_eps
self.k_epochs = k_epochs
self.batch_size = batch_size
self.actor = Actor(local_dim, act_dim).to(device)
self.critic = Critic(global_dim).to(device)
self.opt_a = torch.optim.Adam(self.actor.parameters(), lr=lr)
self.opt_c = torch.optim.Adam(self.critic.parameters(), lr=lr)
self.local_dim = local_dim
self.global_dim = global_dim
self.act_dim = act_dim
self.clear_buffer()
def clear_buffer(self):
self.ls = [] # local observations
self.gs = [] # global observations
self.ac = [] # actions
self.lp = [] # log-probs
self.rw = [] # rewards
self.done = [] # done flags
self.next_gs = [] # next global observations
@torch.no_grad()
def select_action(self, local_obs, global_obs):
l = torch.FloatTensor(local_obs).to(device)
mean, std = self.actor(l)
dist = Normal(mean, std)
a = dist.sample()
return a.cpu().numpy(), dist.log_prob(a).sum(-1).cpu().numpy()
def store(self, local_obs, global_obs, action, logp, reward, done, next_global_obs):
self.ls.append(local_obs)
self.gs.append(global_obs)
self.ac.append(action)
self.lp.append(logp)
self.rw.append(reward)
self.done.append(done)
self.next_gs.append(next_global_obs)
def compute_gae(self, values):
"""
values: torch.Tensor shape [T] (one central V(s) per timestep)
returns:
adv_flat: torch.Tensor shape [T * n_agents]
ret_flat: torch.Tensor shape [T * n_agents]
"""
# 1) get raw arrays
vals_1d = values.cpu().numpy() # [T]
T = len(vals_1d)
N = self.n_agents
# 2) broadcast to per-agent
# vals_agent[t,i] = V(state_t)
vals_agent = np.tile(vals_1d[:,None], (1, N)) # [T,N]
# 3) build next_vals likewise
next_vals = np.zeros_like(vals_agent) # [T,N]
next_vals[:-1] = vals_agent[1:]
# if episode didn’t end at final step, bootstrap last:
if not self.done[-1]:
with torch.no_grad():
v_last = self.critic(
torch.FloatTensor(self.next_gs[-1]).to(device)
).cpu().item()
next_vals[-1, :] = v_last
# 4) GAE loop over (T,N)
adv = np.zeros_like(vals_agent, dtype=np.float32)
prev_adv = np.zeros(N, dtype=np.float32)
for t in reversed(range(T)):
mask = 1.0 - float(self.done[t]) # scalar 0/1
rew_t = np.array(self.rw[t], dtype=np.float32) # [N]
delta = rew_t + self.gamma * next_vals[t] * mask - vals_agent[t]
prev_adv = delta + self.gamma * self.lam * mask * prev_adv
adv[t] = prev_adv
# 5) compute returns & flatten
ret = adv + vals_agent # [T,N]
adv_flat = torch.from_numpy(adv.flatten()).to(device)
ret_flat = torch.from_numpy(ret.flatten()).to(device)
return adv_flat, ret_flat
def update(self):
# 1) Raw global states tensor [T, G]
raw_gs = torch.FloatTensor(self.gs).to(device) # [T, G]
# 2) Compute one value V(s_t) per timestep
with torch.no_grad():
vals = self.critic(raw_gs).cpu() # [T]
# 3) Compute advantages and returns using GAE (returns flattened [T*N])
adv_flat, ret_flat = self.compute_gae(vals) # both shape [T * N]
# 4) Prepare per-agent flattened training tensors
# Local states [T*N, local_dim]
ls = torch.FloatTensor(self.ls).view(-1, self.local_dim).to(device)
# Actions [T*N, act_dim]
ac = torch.FloatTensor(self.ac).view(-1, self.act_dim).to(device)
# Old log-probs [T*N]
old_lp = torch.FloatTensor(self.lp).view(-1).to(device)
# Broadcast global states to per-agent: [T, G] -> [T, N, G] -> [T*N, G]
gs = raw_gs.unsqueeze(1).expand(-1, self.n_agents, -1) # [T, N, G]
gs = gs.reshape(-1, self.global_dim).to(device) # [T*N, G]
# Create dataset and loader
dataset = torch.utils.data.TensorDataset(
ls, gs, ac, old_lp, adv_flat, ret_flat
)
gen = torch.Generator()
gen.manual_seed(SEED)
loader = torch.utils.data.DataLoader(
dataset,
batch_size=self.batch_size,
shuffle=True,
num_workers=0,
generator=gen
)
# 5) PPO update loop
for _ in range(self.k_epochs):
for b_ls, b_gs, b_ac, b_lp, b_adv, b_ret in loader:
# Actor update
mean, std = self.actor(b_ls)
dist = Normal(mean, std)
lp_new = dist.log_prob(b_ac).sum(-1)
ratio = torch.exp(lp_new - b_lp)
surr1 = ratio * b_adv
surr2 = torch.clamp(ratio, 1 - self.clip_eps, 1 + self.clip_eps) * b_adv
actor_loss = -torch.min(surr1, surr2).mean()
self.opt_a.zero_grad()
actor_loss.backward()
self.opt_a.step()
# Critic update
val_pred = self.critic(b_gs)
critic_loss = nn.MSELoss()(val_pred, b_ret)
self.opt_c.zero_grad()
critic_loss.backward()
self.opt_c.step()
# 6) Clear buffers for next rollout
self.clear_buffer()
def save(self, path):
torch.save({'actor': self.actor.state_dict(),
'critic': self.critic.state_dict()}, path)
def load(self, path):
data = torch.load(path, map_location=device)
self.actor.load_state_dict(data['actor'])
self.critic.load_state_dict(data['critic'])
|