-
Notifications
You must be signed in to change notification settings - Fork 1
/
metrics_multiple.py
184 lines (168 loc) · 8.91 KB
/
metrics_multiple.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
import sys
import os
import os.path as osp
import glob
from collections import OrderedDict
from collections.abc import Iterable
import json
import subprocess
import pickle
import logging
import h5py
import math
import operator
import pathlib
import pandas as pd
from tqdm import tqdm
import numpy as np
import torch
import numpy as np
class EmptyResdirError(ValueError):
pass
def allkeys(obj, keys=[]):
"""Recursively find all leaf keys in h5. """
keys = []
for key in obj.keys():
if isinstance(obj[key], h5py.Group):
keys += [f'{key}/{el}' for el in allkeys(obj[key])]
else:
keys.append(key)
return keys
def gen_load_resfiles(resdir):
resfiles = glob.glob(osp.join(resdir, '*.pth'))
if len(resfiles) == 0:
resfiles = glob.glob(osp.join(resdir, '*.h5'))
if len(resfiles) == 0:
raise EmptyResdirError(f'Didnt find any resfiles in {resdir}')
for resfile in resfiles:
if resfile.endswith('.pth'):
output_dict = {
key: val.numpy() if torch.torch.is_tensor(val) else val
for key, val in torch.load(resfile).items()
}
else:
output_dict = {}
with h5py.File(resfile, 'r') as fin:
for key in allkeys(fin):
try:
output_dict[key] = fin[key][()]
except AttributeError as err:
# Happens for the string keys... need to figure what
# to do here
logging.warning('Unable to load %s (%s)', key, err)
yield output_dict
def read_results(resdir):
data = next(gen_load_resfiles(resdir))
# TODO allow to read only certain keys, eg some times we only need logits
# which would be faster to read
res_per_layer = {
key: OrderedDict()
for key in data if key not in ['epoch']
}
if len(res_per_layer) == 0:
raise ValueError('No logits found in the output. Note that code was '
'changed Aug 26 2020 that renames "output" to '
'"logits" etc. So might need to rerun testing.')
logging.info('Reading from resfiles')
for data in gen_load_resfiles(resdir):
for i, idx in enumerate(data['idx']):
idx = int(idx)
for key in res_per_layer:
if idx not in res_per_layer[key]:
res_per_layer[key][idx] = []
res_per_layer[key][idx].append(data[key][i])
# Mean over all the multiple predictions per key
final_res = {}
for key in res_per_layer:
if len(res_per_layer[key]) == 0:
continue
max_idx = max(res_per_layer[key].keys())
key_output = np.zeros([
max_idx + 1,
] + list(res_per_layer[key][0][0].shape))
for idx in res_per_layer[key]:
key_output[idx] = np.mean(np.stack(res_per_layer[key][idx]),
axis=0)
final_res[key] = key_output
return final_res
def get_pred_for_uid(data, uid):
data_by_uid = {}
ind = np.where(data['uid'] == uid)
data_by_uid['logits'] = data['logits/action'][ind]
data_by_uid['target'] = data['target/action'][ind]
data_by_uid['pred_classes'] = np.flip(np.argsort(data['logits/action'][ind]))
return data_by_uid
def extract_preds(data):
preds = {
'uid': data['uid'],
'logits': data['logits/action'],
'target': data['target/action']
}
return preds
if __name__ == "__main__":
# filename = "./0.h5"
# with h5py.File(filename, "r") as f:
# # List all groups
# keys = f.keys()
# print("Keys: %s" % keys)
# data = [list(f[k]) for k in keys]
#
# print(len(data))
# results_dirs = ['/Users/tanviaggarwal/Desktop/SOP/model_ek55/ens_kldiv_top50_6may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/notpretrained_ens_mean_kldiv_top50_13may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/pretrained_ens_attention_nh4_kldiv_top50_13may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/pretrained_ens_mean_kldiv_top50_13may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/notpretrained_alberta_kldiv_top50_13may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/notpretrained_roberta_kldiv_top50_13may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/notpretrained_electra_kldiv_top50_13may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/notpretrained_bert_kldiv_top50_13may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/notpretrained_distillbert_kldiv_top50_13may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/pretrained_alberta_kldiv_top50_9may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/pretrained_roberta_kldiv_top50_9may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/pretrained_electra_kldiv_top50_9may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/pretrained_bert_kldiv_top50_9may_wt20_t5/results',
# '/Users/tanviaggarwal/Desktop/SOP/model_ek55/pretrained_distillbert_kldiv_top50_9may_wt20_t5/results'
# ]
# preds_paths = ['./ek55_preds/notpretrained_ens_attention_nh4_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/notpretrained_ens_mean_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/pretrained_ens_attention_nh4_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/pretrained_ens_mean_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/notpretrained_alberta_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/notpretrained_roberta_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/notpretrained_electra_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/notpretrained_bert_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/notpretrained_distillbert_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/pretrained_alberta_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/pretrained_roberta_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/pretrained_electra_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/pretrained_bert_kldiv_top50_wt20_t5_preds.pickle',
# './ek55_preds/pretrained_distillbert_kldiv_top50_wt20_t5_preds.pickle'
# ]
results_dirs = ['/Users/tanviaggarwal/Desktop/SOP/model_egtea/notpretrained_ens_attention_nh4_kldiv_14may_wt150_t10/results',
'/Users/tanviaggarwal/Desktop/SOP/model_egtea/notpretrained_ens_mean_kldiv_14may_wt150_t10/results',
'/Users/tanviaggarwal/Desktop/SOP/model_egtea/pretrained_ens_attention_nh4_kldiv_14may_wt150_t10/results',
'/Users/tanviaggarwal/Desktop/SOP/model_egtea/pretrained_ens_mean_kldiv_14may_wt150_t10/results',
'/Users/tanviaggarwal/Desktop/SOP/model_egtea/pretrained_alberta_kldiv_11may_wt150_t10/results',
'/Users/tanviaggarwal/Desktop/SOP/model_egtea/pretrained_roberta_kldiv_11may_wt150_t10/results',
'/Users/tanviaggarwal/Desktop/SOP/model_egtea/pretrained_electra_kldiv_11may_wt150_t10/results',
'/Users/tanviaggarwal/Desktop/SOP/model_egtea/pretrained_bert_kldiv_11may_wt150_t10/results',
'/Users/tanviaggarwal/Desktop/SOP/model_egtea/pretrained_distillbert_kldiv_11may_wt150_t10/results',
]
preds_paths = ['./egtea_preds/notpretrained_ens_attention_nh4_kldiv_wt150_t10_preds.pickle',
'./egtea_preds/notpretrained_ens_mean_kldiv_wt150_t10_preds.pickle',
'./egtea_preds/pretrained_ens_attention_nh4_kldiv_wt150_t10_preds_preds.pickle',
'./egtea_preds/pretrained_ens_mean_kldiv_wt150_t10_preds_preds.pickle',
'./egtea_preds/pretrained_alberta_kldiv_wt150_t10_preds.pickle',
'./egtea_preds/pretrained_roberta_kldiv_wt150_t10_preds.pickle',
'./egtea_preds/pretrained_electra_kldiv_wt150_t10_preds.pickle',
'./egtea_preds/pretrained_bert_kldiv_wt150_t10_preds.pickle',
'./egtea_preds/pretrained_distillbert_kldiv_wt150_t10_preds.pickle'
]
for i in range(len(results_dirs)):
data = read_results(results_dirs[i])
preds = extract_preds(data)
with open(preds_paths[i], 'wb') as handle:
pickle.dump(preds, handle)
print("processed")
# data_uid_2 = get_pred_for_uid(data, 19)
# print(len(data_uid_2))