-
Notifications
You must be signed in to change notification settings - Fork 3
/
vis_utils.py
178 lines (154 loc) · 7.86 KB
/
vis_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
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
167
168
169
170
171
172
173
174
175
176
177
"""
Created on 13:59 at 15/10/2021/
@author: bo
"""
import matplotlib.pyplot as plt
from matplotlib.ticker import FuncFormatter, StrMethodFormatter, NullFormatter
import os
import numpy as np
import math
from scipy.special import softmax
def give_avg_std(perf, print_info=False):
perf = np.array(perf)
avg = np.mean(perf, axis=0)
std = np.std(perf, axis=0) * 1.95 / np.sqrt(len(perf))
if print_info:
for i in range(5):
print("Top %d, accuracy %.2f with 95 confidence interval %.2f" % (i + 1, avg[i, 1],
std[i, 1]))
return avg, std
def weight_prediction_fast(prediction_original, group_index, tr_label=[], tt_label=[]):
prediction = prediction_original.copy()
prediction = softmax(prediction, axis=-1)
upper_bound = [upper_bound_multiple_spectra(prediction[:, v]) for v in group_index]
upper_bound = np.transpose(upper_bound, (1, 0))
weights = upper_bound / np.max(upper_bound, axis=-1, keepdims=True)
weights_repeat = [np.repeat(weights[:, i:i+1], len(q), axis=-1) for i, q in enumerate(group_index)]
weights = np.concatenate(weights_repeat, axis=-1)
prediction = prediction * weights
if len(tr_label) > 0:
original_accu = np.sum(tr_label[np.argmax(prediction_original, axis=-1)] == tt_label) / len(tt_label) * 100
updated_accu = np.sum(tr_label[np.argmax(prediction, axis=-1)] == tt_label) / len(tt_label) * 100
print("original accuracy: %.2f vs updated accuracy: %.2f" % (original_accu, updated_accu))
return prediction, [original_accu, updated_accu]
def majority_voting(prediction, tr_label, tt_label, num):
argsort_index = np.argsort(prediction, axis=-1)[:, ::-1]
predict_label = [tr_label[v[:num]] for v in argsort_index]
predict_single_label = np.zeros_like(tt_label)
for i, s_pred in enumerate(predict_label):
_u_s, _u_c = np.unique(s_pred, return_counts=True)
predict_single_label[i] = _u_s[np.argmax(_u_c)]
accuracy = np.sum([v == q for v, q in zip(predict_single_label, tt_label)]) / len(tt_label)
return accuracy, predict_single_label
def enlarge_feature_maps(feature_map, factor=4):
"""Enlarge feature map by a factor"""
new_feat = np.zeros([len(feature_map) * factor, np.shape(feature_map)[1]])
for i in range(len(feature_map)):
new_feat[i*factor:(i+1)*factor] = feature_map[i:i+1]
return new_feat
def extract_accuracy_from_txt_file_simple(raman_type, path_init):
path2read = path_init + "/" + raman_type + "/tds/"
txt_file_group = [v for v in sorted(os.listdir(path2read)) if '.txt' in v]
output = []
for txt_file in txt_file_group:
txt_file = path2read + txt_file
with open(txt_file, 'r') as f:
content = f.readlines()
performance = [v for v in content if "The accuracy on using ensembled prediction" in v]
performance = [float(v.split("is ")[-1].split("\n")[0]) for v in performance]
ensemble_perf = np.reshape(performance, [-1, 4])
single_performance = [v for v in content if "original accuracy" in v and "updated accuracy" in v]
single_original_perf = [float(v.split(": ")[1].split(" vs")[0]) for v in single_performance]
single_updated_perf = [float(v.split("updated accuracy: ")[1].split("\n")[0]) for v in single_performance]
single_performance = np.concatenate([np.reshape(single_original_perf, [-1, 1]),
np.reshape(single_updated_perf, [-1, 1])], axis=1)[
len(single_original_perf) // 2:]
counting_performance = [v for v in content if "Count" in v and "accuracy" in v]
counting_performance = np.reshape(counting_performance, [-1, 6])
counting_accuracy = np.zeros([len(counting_performance), 4])
counting_count = np.zeros([len(counting_performance), 4])
for i in range(len(counting_performance)):
temp = [v.split(" ") for v in counting_performance[i]]
_count_stat = [int(v[1].split(":")[0]) for v in temp]
_accu_stat = [float(v[-1].split("\n")[0]) for v in temp]
counting_count[i, :] = [_count_stat[j] for j in [1, 2, 4, 5]]
counting_accuracy[i, :] = [_accu_stat[j] for j in [1, 2, 4, 5]]
output.append([single_performance, ensemble_perf, counting_accuracy, counting_count])
return output
def show_top5_prediction(select_index, prediction, matched_index, data_group, wavenumber, top5_index=[]):
norm_score = softmax(prediction, axis=-1)[select_index]
if len(top5_index) == 0:
top5_index = np.argsort(norm_score)[-5:][::-1]
matched_top5 = matched_index[select_index][top5_index]
[tr_spectra, tr_label], [tt_spectra, tt_label], label_name = data_group
if "class" in label_name[0]:
label_name = [v.replace("class", "organic") for v in label_name]
legend_g = ["Input:%s" % label_name[tt_label[select_index]]]
for i in range(len(top5_index)):
legend_g.append("Top-%d: %s (%.2f)" % (i+1, label_name[top5_index[i]], (norm_score[top5_index[i]]*100)))
fig = plt.figure(figsize=(7, 6))
ax_global = ax_global_get(fig)
ax_global.set_xticks([])
ax_global.set_yticks([])
for i in range(len(top5_index)+1):
if i == 0:
c_u = 'r'
else:
if top5_index[i-1] == tt_label[select_index]:
c_u = 'r'
else:
c_u = np.random.random([3])
ax = fig.add_subplot(len(top5_index)+1, 1, i+1)
if i == 0:
ax.plot(wavenumber, tt_spectra[select_index], color=c_u, label=legend_g[i])
else:
ax.plot(wavenumber, tr_spectra[matched_top5[i-1]], color=c_u, label=legend_g[i])
ax.legend(loc='best')
if i != len(top5_index):
ax.xaxis.set_major_formatter(plt.NullFormatter())
ax_global.set_xlabel("\nWavenumber (cm" + r"$^{-1}$" + ")")
ax_global.set_ylabel("Intensity (a.u.) \n")
def compare_accuracy_simple(raman_type, select_step, path_init):
output = extract_accuracy_from_txt_file_simple(raman_type, path_init)
stat_single_step = []
for i in range(4):
_stat = [v[-2][select_step, i] for v in output] # 0.05, 0.1, 0.5, 2
stat_single_step.append(_stat)
return np.array(stat_single_step)
def upper_bound_multiple_spectra(predictions):
q1 = np.quantile(predictions, 0.25, axis=-1)
q3 = np.quantile(predictions, 0.75, axis=-1)
iqr = q3 - q1
upper_bound = q3 + 1.5 * iqr
return upper_bound
def show_twinx(alpha_group, accuracy, set_size, ax):
if not ax:
fig = plt.figure(figsize=(5, 3))
ax = fig.add_subplot(111)
ax.set_ylabel("Empirical coverage", color='r')
ax.set_xlabel("Theoretical coverage (1 -" + r'$\alpha$' + ")")
ax_t = ax.twinx()
ax_t.set_ylabel("Average set size", color='g')
else:
ax_t = ax.twinx()
scale = [100 if np.max(alpha_group) > 1 else 1][0]
ax.plot(scale - alpha_group, accuracy, color='r', marker='.', label="Accuracy")
ax.tick_params(axis='y', labelcolor='r')
if np.max(accuracy) > 1:
ax.set_ylim(top=100.0)
else:
ax.set_ylim(top=1.0)
ax_t.plot(scale - alpha_group, set_size, color='g', marker='.', label="Average set size")
ax_t.tick_params(axis='y', labelcolor='g')
ax_t.set_yscale("symlog")
ax_t.yaxis.set_major_formatter(StrMethodFormatter('{x:.2f}'))
ax_t.yaxis.set_minor_formatter(NullFormatter())
ax_t.grid(False)
def ax_global_get(fig):
ax_global = fig.add_subplot(111, frameon=False)
ax_global.spines['top'].set_color('none')
ax_global.spines['bottom'].set_color('none')
ax_global.spines['left'].set_color('none')
ax_global.spines['right'].set_color('none')
ax_global.tick_params(labelcolor='w', top='off', bottom='off', left='off', right='off')
return ax_global