-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathFSLTask.py
166 lines (137 loc) · 6.81 KB
/
FSLTask.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
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
import os
import pickle
import numpy as np
import torch
from pathlib import Path
class FSLTaskMaker:
def __init__(self):
# ========================================================
# Module internal functions and variables
self._min_examples = -1
self._randStates = None
self._rsCfg = None
self.data = None
self.labels = None
self.dsName = None
self.np_random = None
self._maxRuns = 10000
def reset_global_vars(self):
self._min_examples = -1
self._randStates = None
self._rsCfg = None
self.data = None
self.labels = None
self.dsName = None
self.np_random = np.random.RandomState(seed=0)
# Note: The seed here does not matter for reproducibility, because the object
# calls self.np_random.set_state() before using self.np_random in every
# method. There is only one exception: If you call
# self.GenerateRun(iRun, cfg, regenRState=True) without setting
# the np_random state, you may get non-deterministic behavior!
# self.setRandomStates calls self.GenerateRun(iRun, cfg, regenRState=True).
# However, it sets the self.np_random state carefully before
# making this call to ensure reproducibility.
def _load_pickle(self, file):
with open(file, 'rb') as f:
data = pickle.load(f)
labels = [np.full(shape=len(data[key]), fill_value=key)
for key in data]
data = [features for key in data for features in data[key]]
dataset = dict()
dataset['data'] = torch.FloatTensor(np.stack(data, axis=0))
dataset['labels'] = torch.LongTensor(np.concatenate(labels))
return dataset
# =========================================================
# Callable variables and functions from outside the module
def loadDataSet(self, dsname, features_dir):
self.dsName = dsname # Example: self.dsName = 'mini2CUB_novel'
self._randStates = None
self._rsCfg = None
assert os.path.exists(features_dir), f'{features_dir} does not exist'
# Loading data from files on computer
dataset = self._load_pickle(f"{features_dir}/{self.dsName}.pkl")
# Computing the number of items per class in the dataset
self._min_examples = dataset["labels"].shape[0]
for i in range(dataset["labels"].shape[0]):
if torch.where(dataset["labels"] == dataset["labels"][i])[0].shape[0] > 0:
self._min_examples = min(self._min_examples, torch.where(
dataset["labels"] == dataset["labels"][i])[0].shape[0])
print("Guaranteed number of items per class: {:d}\n".format(self._min_examples))
# Generating data tensors
self.data = torch.zeros((0, self._min_examples, dataset["data"].shape[1]))
self.labels = dataset["labels"].clone()
while self.labels.shape[0] > 0:
indices = torch.where(dataset["labels"] == self.labels[0])[0]
self.data = torch.cat([self.data, dataset["data"][indices, :]
[:self._min_examples].view(1, self._min_examples, -1)], dim=0)
indices = torch.where(self.labels != self.labels[0])[0]
self.labels = self.labels[indices]
print("Total of {:d} classes, {:d} elements each, with dimension {:d}\n".format(
self.data.shape[0], self.data.shape[1], self.data.shape[2]))
def GenerateRun(self, iRun, cfg, regenRState=False, generate=True):
if not regenRState:
self.np_random.set_state(self._randStates[iRun])
classes = self.np_random.permutation(np.arange(self.data.shape[0]))[:cfg["n_ways"]]
shuffle_indices = np.arange(self._min_examples)
dataset = None
if generate:
dataset = torch.zeros(
(cfg['n_ways'], cfg['n_shots']+cfg['n_query'], self.data.shape[2]))
for i in range(cfg['n_ways']):
shuffle_indices = self.np_random.permutation(shuffle_indices)
if generate:
dataset[i] = self.data[classes[i], shuffle_indices, :][:cfg['n_shots']+cfg['n_query']]
return dataset
def ClassesInRun(self, iRun, cfg):
self.np_random.set_state(self._randStates[iRun])
classes = self.np_random.permutation(np.arange(self.data.shape[0]))[:cfg["n_ways"]]
return classes
def setRandomStates(self, cfg, cache_dir):
if self._rsCfg == cfg:
return
assert os.path.exists(cache_dir), f'{cache_dir} does not exist'
rsFile = os.path.join(cache_dir, "RandStates_{}_s{}_q{}_w{}_s{}".format(
self.dsName, cfg['n_shots'], cfg['n_query'], cfg['n_ways'], cfg['seed']))
if not os.path.exists(rsFile):
print("{} does not exist, regenerating it...".format(rsFile))
self.np_random.seed(cfg['seed'])
self._randStates = []
for iRun in range(self._maxRuns):
self._randStates.append(self.np_random.get_state())
self.GenerateRun(iRun, cfg, regenRState=True, generate=False)
torch.save(self._randStates, rsFile)
else:
print("reloading random states from file....")
self._randStates = torch.load(rsFile)
self._rsCfg = cfg
def GenerateRunSet(self, start=None, end=None, cfg=None, cache_dir=None):
if start is None:
start = 0
if end is None:
end = self._maxRuns
if cfg is None:
cfg = {"n_shots": 1, "n_ways": 5, "n_query": 15, "seed": 0}
self.setRandomStates(cfg, cache_dir=cache_dir)
print("generating task from {} to {}".format(start, end))
dataset = torch.zeros((end-start, cfg['n_ways'], cfg['n_shots']+cfg['n_query'], self.data.shape[2]))
for iRun in range(end-start):
dataset[iRun] = self.GenerateRun(start+iRun, cfg)
return dataset
# define a main code to test this module
if __name__ == "__main__":
taskmaker = FSLTaskMaker()
print("Testing Task loader for Few Shot Learning")
features_dir = './features/WideResNet_28_10_S2M2_R'
cache_dir = './cache/WideResNet_28_10_S2M2_R'
Path(features_dir).mkdir(parents=True, exist_ok=True)
Path(cache_dir).mkdir(parents=True, exist_ok=True)
taskmaker.loadDataSet('mini2CUB_novel', features_dir=features_dir)
cfg = {"n_shots": 1, "n_ways": 5, "n_query": 15, "seed": 0}
taskmaker.setRandomStates(cfg, cache_dir=cache_dir)
run10 = taskmaker.GenerateRun(10, cfg)
print("First call:", run10[:2, :2, :2])
run10 = taskmaker.GenerateRun(10, cfg)
print("Second call:", run10[:2, :2, :2])
ds = taskmaker.GenerateRunSet(start=2, end=12, cfg=cfg, cache_dir=cache_dir)
print("Third call:", ds[8, :2, :2, :2])
print(ds.size())