-
Notifications
You must be signed in to change notification settings - Fork 12
/
utils.py
51 lines (42 loc) · 2.02 KB
/
utils.py
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
"""
Based on https://github.com/sfujim/BCQ
"""
import numpy as np
import torch
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
class ReplayBuffer(object):
def __init__(self, state_dim, action_dim, max_size=int(2e6)):
self.max_size = max_size
self.ptr = 0
self.size = 0
self.storage = dict()
self.storage['state'] = np.zeros((max_size, state_dim))
self.storage['action'] = np.zeros((max_size, action_dim))
self.storage['next_state'] = np.zeros((max_size, state_dim))
self.storage['reward'] = np.zeros((max_size, 1))
self.storage['not_done'] = np.zeros((max_size, 1))
def add(self, state, action, next_state, reward, done):
self.storage['state'][self.ptr] = state
self.storage['action'][self.ptr] = action
self.storage['next_state'][self.ptr] = next_state
self.storage['reward'][self.ptr] = reward
self.storage['not_done'][self.ptr] = 1. - done
self.ptr = (self.ptr + 1) % self.max_size
self.size = min(self.size + 1, self.max_size)
def sample(self, batch_size):
ind = np.random.randint(0, self.size, size=batch_size)
return (
torch.FloatTensor(self.storage['state'][ind]).to(device),
torch.FloatTensor(self.storage['action'][ind]).to(device),
torch.FloatTensor(self.storage['next_state'][ind]).to(device),
torch.FloatTensor(self.storage['reward'][ind]).to(device),
torch.FloatTensor(self.storage['not_done'][ind]).to(device)
)
def save(self, filename):
np.save("./buffers/" + filename + ".npy", self.storage)
def load(self, data):
assert('next_observations' in data.keys())
for i in range(data['observations'].shape[0] - 1):
self.add(data['observations'][i], data['actions'][i], data['next_observations'][i],
data['rewards'][i], data['terminals'][i])
print("Dataset size:" + str(self.size))