-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathget_faitfhul_diag.py
213 lines (159 loc) · 8.96 KB
/
get_faitfhul_diag.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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
from pickle import NONE
from re import T
import pandas as pd
import json
import glob
import os
import argparse
import logging
import numpy as np
import datetime
import gc
date_time = str(datetime.date.today()) + "_" + ":".join(str(datetime.datetime.now()).split()[1].split(":")[:2])
parser = argparse.ArgumentParser()
parser.add_argument(
"--dataset",
type = str,
help = "select dataset / task",
default = "agnews",
#choices = ["sst", "evinf", "agnews", "multirc"]
)
parser.add_argument(
"--data_dir",
type = str,
help = "directory of saved processed data",
default = "datasets/"
)
parser.add_argument(
"--model_dir",
type = str,
help = "directory to save models",
default = "trained_models/"
)
parser.add_argument(
"--evaluation_dir",
type = str,
help = "directory to save faithfulness results",
default = "posthoc_results/"
)
user_args = vars(parser.parse_args())
faithful_result = user_args['evaluation_dir']
dataset = user_args['dataset']
pwd = os.getcwd()
topk_scores_file = os.path.join(pwd, 'posthoc_results', str(dataset), 'topk-faithfulness-scores-detailed.npy')
NOISE_scores_file = os.path.join(pwd, 'posthoc_results', str(dataset), 'NOISElimit-faithfulness-scores-detailed.npy')
ATTENTION_scores_file = os.path.join(pwd, 'posthoc_results', str(dataset), 'ATTENTIONlimit-faithfulness-scores-detailed.npy')
ZEROOUT_scores_file = os.path.join(pwd, 'posthoc_results', str(dataset), 'ZEROOUTlimit-faithfulness-scores-detailed.npy')
TOPk_scores = np.load(topk_scores_file, allow_pickle=True).item()
ZEROOUT_scores = np.load(ZEROOUT_scores_file, allow_pickle=True).item()
ATTENTION_scores = np.load(ATTENTION_scores_file, allow_pickle=True).item()
NOISE_scores = np.load(NOISE_scores_file, allow_pickle=True).item() # key feature_ suff/comp @
data_id_list = TOPk_scores.keys()
fea_list = ['attention', "scaled attention", "gradients", "ig", "deeplift"] # "gradientshap",
rationale_ratios = [0.01, 0.02, 0.05, 0.1, 0.2, 0.5, 1.0]
suff_or_comp = 'comprehensiveness' # sufficiencies or comprehensiveness
def open_file(file_path):
topk_faith = pd.read_json(file_path, orient ='index')
topk_faith.rename(columns = {'AOPC - sufficiency':'AOPC_sufficiency', 'AOPC - comprehensiveness':'AOPC_comprehensiveness'}, inplace = True)
return topk_faith
def generate_table(suff_or_comp, ratio, include_feature_name=True):
D_TOP_Suff = []
D_ATTENTION_Suff = []
D_ZEROOUT_Suff = []
D_NOISE_Suff = []
ABS_D_TOP_Suff = []
ABS_D_ATTENTION_Suff = []
ABS_D_ZEROOUT_Suff = []
ABS_D_NOISE_Suff = []
for FA in fea_list:
Diag_TOP_attention = 0
Diag_ATTENTION_attention = 0
Diag_ZEROOUT_attention = 0
Diag_NOISE_attention = 0
ABS_Diag_TOP_attention = 0
ABS_Diag_ATTENTION_attention = 0
ABS_Diag_ZEROOUT_attention = 0
ABS_Diag_NOISE_attention = 0
for i, data_id in enumerate(data_id_list):
top_random_suff_score = TOPk_scores.get(data_id).get('random').get(f'{suff_or_comp} @ {str(ratio)}')
NOISE_random_suff_score = NOISE_scores.get(data_id).get('random').get(f'{suff_or_comp} @ {str(ratio)}')
ZEROOUT_random_suff_score = ZEROOUT_scores.get(data_id).get('random').get(f'{suff_or_comp} @ {str(ratio)}')
ATTENTION_random_suff_score = ATTENTION_scores.get(data_id).get('random').get(f'{suff_or_comp} @ {str(ratio)}')
top_suff_score = TOPk_scores.get(data_id).get(FA).get(f'{suff_or_comp} @ {str(ratio)}')
if top_suff_score >= top_random_suff_score:
Diag_TOP_attention += 1
ABS_Diag_TOP_attention += top_suff_score
else: pass
NOISE_suff_score = NOISE_scores.get(data_id).get(FA).get(f'{suff_or_comp} @ {str(ratio)}')
if NOISE_suff_score >= NOISE_random_suff_score:
Diag_NOISE_attention += 1
ABS_Diag_ATTENTION_attention += NOISE_suff_score
else: pass
ZEROOUT_suff_score = ZEROOUT_scores.get(data_id).get(FA).get(f'{suff_or_comp} @ {str(ratio)}')
if ZEROOUT_suff_score >= ZEROOUT_random_suff_score:
Diag_ZEROOUT_attention += 1
ABS_Diag_ZEROOUT_attention += ZEROOUT_suff_score
else: pass
ATTENTION_suff_score = ATTENTION_scores.get(data_id).get(FA).get(f'{suff_or_comp} @ {str(ratio)}')
if ATTENTION_suff_score >= ATTENTION_random_suff_score:
Diag_ATTENTION_attention += 1
ABS_Diag_NOISE_attention += ATTENTION_suff_score
else: pass
D_TOP = Diag_TOP_attention/len(data_id_list)
D_TOP_Suff.append(D_TOP)
ABS_D_TOP = ABS_Diag_TOP_attention/len(data_id_list)
ABS_D_TOP_Suff.append(ABS_D_TOP)
D_ATTENTION = Diag_ATTENTION_attention/len(data_id_list)
D_ATTENTION_Suff.append(D_ATTENTION)
ABS_D_ATTENTION = ABS_Diag_ATTENTION_attention/len(data_id_list)
ABS_D_ATTENTION_Suff.append(ABS_D_ATTENTION)
D_ZEROOUT = Diag_ZEROOUT_attention/len(data_id_list)
D_ZEROOUT_Suff.append(D_ZEROOUT)
ABS_D_ZEROOUT = ABS_Diag_ZEROOUT_attention/len(data_id_list)
ABS_D_ZEROOUT_Suff.append(ABS_D_ZEROOUT)
D_NOISE= Diag_NOISE_attention/len(data_id_list)
D_NOISE_Suff.append(D_NOISE)
ABS_D_NOISE= ABS_Diag_NOISE_attention/len(data_id_list)
ABS_D_NOISE_Suff.append(ABS_D_NOISE)
## get faith scores from description json
topk_scores_file = os.path.join(pwd, 'posthoc_results', str(dataset), 'topk-faithfulness-scores-average-description.json')
NOISE_scores_file = os.path.join(pwd, 'posthoc_results', str(dataset), 'NOISElimit-faithfulness-scores-description.json')
ATTENTION_scores_file = os.path.join(pwd, 'posthoc_results', str(dataset), 'ATTENTIONlimit-faithfulness-scores-description.json')
ZEROOUT_scores_file = os.path.join(pwd, 'posthoc_results', str(dataset), 'ZEROOUTlimit-faithfulness-scores-description.json')
topk_df = open_file(topk_scores_file)
print(topk_df)
noise_df = open_file(NOISE_scores_file)
attention_df = open_file(ATTENTION_scores_file)
zeroout_df = open_file(ZEROOUT_scores_file)
topk_suff_or_comp_mean = []
noise_suff_or_comp_mean = []
attention_suff_or_comp_mean = []
zeroout_suff_or_comp_mean = []
if suff_or_comp == 'sufficiency': suff_or_comp = 'sufficiencies'
else: pass
topk_random_faitfhul_score = topk_df.loc['random'][f'{suff_or_comp} @ {ratio}'].get('mean')
noise_random_faitfhul_score = noise_df.loc['random'][f'{suff_or_comp} @ {ratio}'].get('mean')
attention_random_faitfhul_score = attention_df.loc['random'][f'{suff_or_comp} @ {ratio}'].get('mean')
zeroout_random_faitfhul_score = zeroout_df.loc['random'][f'{suff_or_comp} @ {ratio}'].get('mean')
#print(noise_random_faitfhul_score, attention_random_faitfhul_score, zeroout_random_faitfhul_score)
for FA in fea_list:
# print(topk_df.loc[FA][f'{suff_or_comp} @ {ratio}'].get('mean'))
# print(topk_random_faitfhul_score)
topk_suff_or_comp_mean.append((topk_df.loc[FA][f'{suff_or_comp} @ {ratio}'].get('mean'))/(topk_random_faitfhul_score))
#print(noise_df.loc[FA][f'{suff_or_comp} @ {ratio}'].get('mean'))
noise_suff_or_comp_mean.append((noise_df.loc[FA][f'{suff_or_comp} @ {ratio}'].get('mean'))/(noise_random_faitfhul_score))
attention_suff_or_comp_mean.append((attention_df.loc[FA][f'{suff_or_comp} @ {ratio}'].get('mean'))/(attention_random_faitfhul_score))
zeroout_suff_or_comp_mean.append((zeroout_df.loc[FA][f'{suff_or_comp} @ {ratio}'].get('mean'))/(zeroout_random_faitfhul_score))
final_big_df = pd.DataFrame(list(zip(fea_list, D_TOP_Suff, ABS_D_TOP_Suff, topk_suff_or_comp_mean,
D_ZEROOUT_Suff, ABS_D_ZEROOUT_Suff, zeroout_suff_or_comp_mean,
D_NOISE_Suff, ABS_D_NOISE_Suff, noise_suff_or_comp_mean,
D_ATTENTION_Suff, ABS_D_ATTENTION_Suff, attention_suff_or_comp_mean)),
columns =['Feature', f'D - TopK {suff_or_comp} @ {str(ratio)}', f'ABS D - TopK {suff_or_comp} @ {str(ratio)}', f'TopK {suff_or_comp} @ {str(ratio)}',
f'D - Soff(ZEROOUT) {suff_or_comp} @ {str(ratio)}', f'ABS D - Soff(ZEROOUT) {suff_or_comp} @ {str(ratio)}', f'Soff(ZEROOUT) {suff_or_comp} @ {str(ratio)}',
f'D - Soff(NOISE) {suff_or_comp} @ {str(ratio)}', f'ABS D - Soff(NOISE) {suff_or_comp} @ {str(ratio)}', f'Soff(NOISE) {suff_or_comp} @ {str(ratio)}',
f'D - Soff(ATTENTION) {suff_or_comp} @ {str(ratio)}', f'ABS D - Soff(ATTENTION) {suff_or_comp} @ {str(ratio)}', f'Soff(ATTENTION) {suff_or_comp} @ {str(ratio)}'])
fname = os.path.join(pwd, 'Diagnosticity', str(dataset), f'Soft_{suff_or_comp}.csv')
os.makedirs(os.path.join(pwd, 'Diagnosticity', str(dataset)), exist_ok=True)
final_big_df.to_csv(fname)
#return final_big_df
generate_table(suff_or_comp, 0.2, include_feature_name=True)