forked from carpedm20/simulated-unsupervised-tensorflow
-
Notifications
You must be signed in to change notification settings - Fork 0
/
buffer.py
33 lines (28 loc) · 960 Bytes
/
buffer.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
import numpy as np
class Buffer(object):
def __init__(self, config, rng):
self.rng = rng
self.buffer_size = config.buffer_size
self.batch_size = config.batch_size
image_dims = [
config.input_height,
config.input_width,
config.input_channel,
]
self.idx = 0
self.data = np.zeros([self.buffer_size] + image_dims)
def push(self, batches):
batch_size = len(batches)
if self.idx + batch_size > self.buffer_size:
random_idx1 = self.rng.choice(self.idx, self.batch_size/2)
random_idx2 = self.rng.choice(batch_size, self.batch_size/2)
self.data[random_idx1] = batches[random_idx2]
else:
self.data[self.idx:self.idx+batch_size] = batches
self.idx += batch_size
def sample(self, n=None):
assert self.idx > n, "not enough data is pushed"
if n is None:
n = self.batch_size/2
random_idx = self.rng.choice(self.idx, n)
return self.data[random_idx]