/
k_neighbours.py
executable file
·135 lines (109 loc) · 4.67 KB
/
k_neighbours.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
from TlseHypDataSet.tlse_hyp_data_set import TlseHypDataSet
from TlseHypDataSet.utils.dataset import DisjointDataSplit
from sklearn.metrics import accuracy_score, f1_score
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import RandomizedSearchCV
from models.load_model import load_model
import numpy as np
import torch
import torch.nn as nn
import json
import pickle as pkl
import os
import pdb
import sys
from tqdm import tqdm
dataset = TlseHypDataSet(
'/path/to/dataset',
pred_mode='pixel',
patch_size=1,
in_h5py=True,
data_on_gpu=True,
)
base_folder = sys.argv[1]
splits = [DisjointDataSplit(dataset, split=default_split) for default_split in dataset.standard_splits]
batch_size = 1024
raw = sys.argv[2]
for split_id, split in enumerate(splits):
if raw == 'raw':
config = {'model': 'raw'}
else:
folder = os.path.join(base_folder, 'split_{}'.format(split_id+1))
with open(os.path.join(folder, 'config.json'), 'r') as f:
config = json.load(f)
model = load_model(config)
model.eval()
checkpoint = torch.load(os.path.join(folder, 'best_model.pth.tar'), map_location=config['device'])
checkpoint = checkpoint['state_dict']
model.load_state_dict(checkpoint)
labeled_loader = torch.utils.data.DataLoader(split.sets_['train'], batch_size=batch_size, pin_memory=True)
val_loader = torch.utils.data.DataLoader(split.sets_['validation'], batch_size=batch_size, pin_memory=True)
test_loader = torch.utils.data.DataLoader(split.sets_['test'], batch_size=batch_size, pin_memory=True)
train_data = []
train_labels = []
val_data = []
val_labels = []
test_data = []
test_labels = []
for data, labels in tqdm(labeled_loader):
with torch.no_grad():
if config['model'] == 'MAE':
_, _, _, data = model.forward(data.view(data.shape[0], data.shape[-1]), mask_ratio=0)
elif config['model'] == 'AE':
data, _ = model.forward(data.view(data.shape[0], data.shape[-1]))
else:
data = data.view(data.shape[0], data.shape[-1])
train_data.append(data.view(data.shape[0], -1))
train_labels.append(labels.view(-1))
for data, labels in tqdm(val_loader):
with torch.no_grad():
if config['model'] == 'MAE':
_, _, _, data = model.forward(data.view(data.shape[0], data.shape[-1]), mask_ratio=0)
elif config['model'] == 'AE':
data, _ = model.forward(data.view(data.shape[0], data.shape[-1]))
else:
data = data.view(data.shape[0], data.shape[-1])
val_data.append(data.view(data.shape[0], -1))
val_labels.append(labels.view(-1))
for data, labels in tqdm(test_loader):
with torch.no_grad():
if config['model'] == 'MAE':
_, _, _, data = model.forward(data.view(data.shape[0], data.shape[-1]), mask_ratio=0)
elif config['model'] == 'AE':
data, _ = model.forward(data.view(data.shape[0], data.shape[-1]))
else:
data = data.view(data.shape[0], data.shape[-1])
test_data.append(data.view(data.shape[0], -1))
test_labels.append(labels.view(-1))
train_data = torch.cat(train_data, dim=0).numpy()
train_labels = torch.cat(train_labels, dim=0).numpy()
val_data = torch.cat(val_data, dim=0).numpy()
val_labels = torch.cat(val_labels, dim=0).numpy()
test_data = torch.cat(test_data, dim=0).numpy()
test_labels = torch.cat(test_labels, dim=0).numpy()
data = np.concatenate((train_data, val_data), axis=0)
labels = np.concatenate((train_labels, val_labels), axis=0)
train_indices = np.arange(len(train_labels))
val_indices = np.arange(len(train_labels), len(train_labels) + len(val_labels))
cv = [(train_indices, val_indices)]
estimator = KNeighborsClassifier()
params = {
'n_neighbors': [3, 5, 10],
'leaf_size': [20, 30, 40],
'p': [1, 2],
'weights': ['uniform', 'distance']
}
clf = RandomizedSearchCV(estimator=estimator, param_distributions=params, cv=cv, n_iter=20, verbose=1)
search = clf.fit(data, labels)
print(search.best_params_)
pred = clf.predict(test_data)
OA = accuracy_score(pred, test_labels)
F1 = f1_score(pred, test_labels, average=None)
avg_F1 = f1_score(pred, test_labels, average='macro')
test_metrics = {
'OA': OA,
'f1_score': list(F1),
'avg F1': avg_F1
}
with open(os.path.join(base_folder, 'KNN_split_{}_test_metrics.json'.format(split_id)), 'w') as f:
json.dump(test_metrics, f, indent=4)