forked from BoulderDS/feature-importance
-
Notifications
You must be signed in to change notification settings - Fork 0
/
distribution.py
181 lines (164 loc) · 8.23 KB
/
distribution.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
178
179
180
181
import numpy as np
import pandas as pd
import seaborn as sns
import scipy.stats as ss
import matplotlib.pyplot as plt
import lstm as lc
import analysis as als
import similarity as simi
import scikit_classification as sc
import os
import utils
import torch
from scipy.spatial import distance
from collections import defaultdict, OrderedDict, Counter
'''
START OF ENTROPY FUNCTIONS
'''
def run_entropy(dataset_name, save_dir, k_list, models, feature_types, second=False):
train_tokens, dev_tokens, train_dev_tokens, test_tokens, \
train_labels, dev_labels, train_dev_labels, test_labels = utils.load_data(dataset_name)
y_data, min_vals, max_vals, y_min_val, y_max_val = [], [], [], 0, 0
for model_name in models:
dicts, d_keys = als.create_model_d(save_dir, model_name, test_labels=test_labels)
tmp_y_data = als.get_entropy(test_tokens, dicts, d_keys, k_list)
assert len(tmp_y_data) == len(d_keys)
tmp_min_val, tmp_max_val = als.get_min_max(tmp_y_data)
min_vals.append(tmp_min_val)
max_vals.append(tmp_max_val)
y_data.append(tmp_y_data)
y_min_val = np.min(min_vals) - 0.25
y_min_val = max(0, y_min_val)
y_max_val = np.max(max_vals) + 0.25
simi.show_simi_plot(k_list, y_data, 'Number of important features (k)', 'Entropy', '', \
(13, 12), '', x_min=np.min(k_list)-0.5, x_max=np.max(k_list)+0.5, \
y_min=y_min_val, y_max=y_max_val, if_model=True, second=second, \
if_combi=False, if_builtin_posthoc=True)
'''
END OF ENTROPY FUNCTIONS
'''
'''
START POS WITH BUILT-IN FUNCTIONS
'''
def show_bar_plot(x_data, all_combi_data, x_label, y_label, file_name, \
fig_size, folder_name, y_err=None, x_min=None, x_max=None, \
y_min=None, y_max=None, labels=None, save=False):
fig, ax = plt.subplots(figsize=fig_size)
index = np.arange(len(x_data))
bar_width = 0.3
colors = ['#344b5b', '#356384', '#367bac', '#4892c6', '#69a6d0', '#8abbdb']
for idx, y_data in enumerate(all_combi_data):
bar = plt.bar(index*(bar_width*len(labels)+0.2)+(bar_width*idx), y_data, width=bar_width, \
tick_label=x_data, color=colors[idx], label=labels[idx])
plt.legend(bbox_to_anchor=(1.04,0), loc="lower left", borderaxespad=0)
ax.set_xticks(index*(bar_width*len(labels)+0.2)+bar_width*idx - bar_width*(len(labels)/2))
if x_label != None:
plt.xlabel(x_label)
if y_label != None:
plt.ylabel(y_label)
if x_min != None and x_max != None:
plt.xlim(x_min, x_max)
if y_min != None and y_max != None:
plt.ylim(y_min, y_max)
if save:
path = get_save_path(folder_name, file_name)
plt.show()
plt.close()
def run_pos_percent(dataset_name, data_dir, save_dir, models, feature_types, k):
train_tokens, dev_tokens, train_dev_tokens, test_tokens, \
train_labels, dev_labels, train_dev_labels, test_labels = utils.load_data(dataset_name)
train_pos, dev_pos, train_dev_pos, test_pos = utils.get_pos(dataset_name, data_dir)
pos_types = ['NOUN', 'VERB', 'ADJ', 'ADV', 'PRON', 'DET']
token_pos_d = als.get_token_pos_d(test_tokens, test_pos)
vocab_size = len(test_tokens) * k
min_vals, max_vals, y_min_val, y_max_val = [], [], 0, 0
for idx, feature_type in enumerate(feature_types):
dicts, d_keys = als.create_explainer_d(save_dir, feature_type, len(test_labels), test_labels=test_labels)
tmp_y_data = als.get_combi_pos(d_keys, dicts, test_tokens, k, token_pos_d, vocab_size)
tmp_min_val, tmp_max_val = als.get_min_max(tmp_y_data)
y_min_val = max(tmp_min_val-1, 0)
y_max_val = min(tmp_max_val+1, 100)
y_data = als.format_pos_data(tmp_y_data, pos_types)
assert len(y_data) == len(pos_types)
display_model_names = als.get_explainer_combinations(combi=False)
display_feature_names = als.get_model_combinations(combi=False)
x_data = []
x_data.append('Background')
for model_name in display_model_names:
label = '{}'.format(model_name)
x_data.append(label)
show_bar_plot(x_data, y_data, '', 'Percentage', \
'', (15, 14), '', y_min=y_min_val, \
y_max=y_max_val, labels=pos_types)
'''
END POS WITH BUILT-IN FUNCTIONS
'''
'''
START OF DISTANCE BETWEEN POS & IMPORTANT WORDS FUNCTIONS
'''
def get_jensen_shannon(test_tokens, dicts, d_keys, k_list, var, combinations=None, token_pos_d=None):
data = []
if var == 'word_dist':
background_keys, background_total, background_counter = als.get_keys_total(test_tokens)
k_combi_keys, k_combi_total, k_combi_counter = als.get_combi_keys_total(test_tokens, dicts, d_keys, k_list)
for idx_a, combi_keys in enumerate(k_combi_keys):
tmp = []
for idx_b, combi_k in enumerate(combi_keys):
all_keys = set(combi_k) | set(background_keys)
first_proba = [background_counter.get(k, 0) / background_total for k in all_keys]
cur_counter, cur_total = k_combi_counter[idx_a][idx_b], k_combi_total[idx_a][idx_b]
second_proba = [cur_counter.get(k, 0) / cur_total for k in all_keys]
jensen_shannon = distance.jensenshannon(first_proba, second_proba, base=2.0)
tmp.append(jensen_shannon)
data.append(tmp)
elif var == 'pos':
for combi in combinations:
combi1, combi2 = combi[0], combi[1]
tmp = []
for k in k_list:
combi1_top_k_l = als.get_tokens_top_k(test_tokens, dicts[combi1], k)
combi2_top_k_l = als.get_tokens_top_k(test_tokens, dicts[combi2], k)
assert len(combi1_top_k_l) == len(combi2_top_k_l)
combi1_pos_val = als.get_pos_val(combi1_top_k_l, token_pos_d)
combi2_pos_val = als.get_pos_val(combi2_top_k_l, token_pos_d)
assert len(combi1_pos_val) == len(combi2_pos_val)
jensen_shannon = distance.jensenshannon(combi1_pos_val, combi2_pos_val, base=2.0)
tmp.append(jensen_shannon)
data.append(tmp)
elif var == 'background':
tokens = [row.split() for row in test_tokens]
vocab_size = als.get_vocab_size(test_tokens)
background_pos_val = als.get_pos_val(tokens, token_pos_d, vocab_size)
for key in combinations:
tmp = []
for k in k_list:
combi1_top_k_l = als.get_tokens_top_k(test_tokens, dicts[key], k)
combi1_pos_val = als.get_pos_val(combi1_top_k_l, token_pos_d)
assert len(combi1_pos_val) == len(background_pos_val)
jensen_shannon = distance.jensenshannon(combi1_pos_val, background_pos_val, base=2.0)
tmp.append(jensen_shannon)
data.append(tmp)
return data
def run_js_pos(dataset_name, data_dir, save_dir, models, feature_types, k_list, second=False):
train_tokens, dev_tokens, train_dev_tokens, test_tokens, \
train_labels, dev_labels, train_dev_labels, test_labels = utils.load_data(dataset_name)
train_pos, dev_pos, train_dev_pos, test_pos = utils.get_pos(dataset_name, data_dir)
token_pos_d = als.get_token_pos_d(test_tokens, test_pos)
# compare with background
y_data, min_vals, max_vals, y_min_val, y_max_val = [], [], [], 0, 0
for model_name in models:
dicts, d_keys = als.create_model_d(save_dir, model_name, test_labels=test_labels)
tmp_y_data = get_jensen_shannon(test_tokens, dicts, d_keys, k_list, 'background', \
combinations=d_keys, token_pos_d=token_pos_d)
tmp_min_val, tmp_max_val = als.get_min_max(tmp_y_data)
min_vals.append(tmp_min_val)
max_vals.append(tmp_max_val)
y_data.append(tmp_y_data)
y_min_val = np.min(min_vals) - 0.05
y_max_val = np.max(max_vals) + 0.05
simi.show_simi_plot(k_list, y_data, 'Number of important features (k)', 'Jensen-Shannon Score', '', \
(13, 12), '', y_min=y_min_val, y_max=y_max_val, if_model=True, second=second, \
if_combi=False, if_background=True, if_builtin_posthoc=True)
'''
START OF DISTANCE BETWEEN POS & IMPORTANT WORDS FUNCTIONS
'''