/
run_svm_with_feedback.py
168 lines (129 loc) · 5.57 KB
/
run_svm_with_feedback.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
from common import *
import svm
import svmutil
import display_image
import sklearn.metrics
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
class RunSVMWithFeedback:
def __init__(self, random_feedback=False):
self.X_train = []
self.y_train = []
self.labels_train = []
self.X_test = None
self.y_test = None
self.labels_test = None
self.svm_model = None
self.all_distances = []
self.num_refinements = CONFIG["svm"]["num_refinements"]
self.num_for_feedback = CONFIG["svm"]["num_feedback"]
self.display_image = None
self.svm_param = CONFIG["svm"]["param"]
self.simulated = CONFIG["svm"]["simulated"]
self.random = random_feedback
self.refinement_results = []
logging.debug("LOADING DATA")
self.all_data = self.load_all_data()
logging.debug("LOADED DATA")
def load_all_data(self):
[all_pos_data, all_neg_data, all_pos_labels, all_neg_labels] = \
np.load(CONFIG["database_config"]["all_data_file"])
self.X_train, self.y_train, self.labels_train = [], [], []
assert len(all_pos_data) + len(all_neg_data) == \
len(all_pos_labels) + len(all_neg_labels)
random_perm_pos = \
np.random.permutation(len(all_pos_data))[:CONFIG["svm"]["num_pos"]]
random_perm_neg = \
np.random.permutation(len(all_neg_data))[:CONFIG["svm"]["num_neg"]]
random_perm = np.append(random_perm_pos, len(all_pos_data) + random_perm_neg)
np.random.shuffle(random_perm)
all_data_list = all_pos_data + all_neg_data
all_y_list = [1]*len(all_pos_data) + [-1]*len(all_neg_data)
all_labels_list = all_pos_labels + all_neg_labels
X_test = [all_data_list[i] for i in random_perm]
self.y_test = [all_y_list[i] for i in random_perm]
self.labels_test = [all_labels_list[i] for i in random_perm]
assert sum(self.y_test) == \
CONFIG["svm"]["num_pos"] - CONFIG["svm"]["num_neg"]
self.X_test = [
dict(zip(range(len(x)), x)) for x in X_test
]
def calc_distance(self, x):
svm_node_array, _ = svm.gen_svm_nodearray(x)
distance = svmutil.svm_distance_from_plane(svm_node_array, self.svm_model)
return abs(distance[0])
def get_ind_for_feedback(self, X_test, use_random_indices):
if use_random_indices:
return np.random.choice(range(len(X_test)), self.num_for_feedback,
replace=False)
all_distances = []
for x in X_test:
all_distances.append(self.calc_distance(x))
sorted_ind = np.argsort(all_distances)
return sorted_ind[:self.num_for_feedback]
def get_result_from_image(self, ind):
if(self.simulated):
return self._running_y_test[ind]
label = self._running_labels_test[ind]
display_img = display_image.DisplayPhotos()
display_img.image_path = get_file_path_from_label(label) + "." + \
CONFIG["database_config"]["image_extension"]
logging.debug("DISPLAYING IMAGE " + display_img.image_path)
display_img.display_window()
return 1 if display_img.result[1] else -1
def run_feedback(self, sorted_indices):
for i in sorted_indices:
res = self.get_result_from_image(i)
self._running_X_train.append(self._running_X_test[i])
self._running_y_train.append(res)
self._running_labels_train.append(self._running_labels_test[i])
for i in sorted(sorted_indices, reverse=True):
self._running_X_test.pop(i)
self._running_y_test.pop(i)
self._running_labels_test.pop(i)
def train_test_svm(self):
logging.debug("TRAINING Samples: " + str(len(self._running_X_train)))
logging.debug("TESTING Samples: " + str(len(self._running_X_test)))
svm_problem = svm.svm_problem(self._running_y_train, self._running_X_train)
self.svm_model = svmutil.svm_train(svm_problem, self.svm_param)
predicted_labels, predicted_mse, predicted_probs = \
svmutil.svm_predict(self._running_y_test,
self._running_X_test, self.svm_model, "-b 1")
res = sklearn.metrics.accuracy_score(self._running_y_test, predicted_labels)
self.refinement_results.append(res)
print("RESULT: " + str(res*100))
def run_svm(self):
self._running_X_train = self.X_train[:]
self._running_y_train = self.y_train[:]
self._running_labels_train = self.labels_train[:]
self._running_X_test = self.X_test[:]
self._running_y_test = self.y_test[:]
self._running_labels_test = self.labels_test[:]
for i in range(self.num_refinements):
sorted_ind = self.get_ind_for_feedback(self._running_X_test,
True if i == 0 else self.random)
self.run_feedback(sorted_ind)
logging.debug("REFINEMENT ROUND: " + str(i+1))
self.train_test_svm()
def main():
svm_feedback_random = RunSVMWithFeedback(random_feedback=True)
svm_feedback_random.run_svm()
svm_feedback_non_random = RunSVMWithFeedback(random_feedback=False)
svm_feedback_non_random.run_svm()
p1 = [x*100 for x in svm_feedback_random.refinement_results]
p2 = [x*100 for x in svm_feedback_non_random.refinement_results]
plt.plot(p1, linewidth=2.0, label="random")
plt.scatter(range(len(p1)), p1)
plt.plot(p2, linewidth=2.0, label="SVM-active")
plt.scatter(range(len(p2)), p2)
plt.xlabel("Refinements")
plt.ylabel("Accuracy")
plt.title("{} feedbacks per refinement and {} refinements".
format(svm_feedback_random.num_for_feedback,
svm_feedback_random.num_refinements))
plt.legend(loc="lower right")
plt.ylim(0, 100)
plt.xlim(0, CONFIG["svm"]["num_refinements"])
plt.savefig(CONFIG["svm"]["plot_file"])
if __name__ == "__main__":
main()