-
Notifications
You must be signed in to change notification settings - Fork 0
/
compute_metrics.py
386 lines (320 loc) · 13.3 KB
/
compute_metrics.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
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
"""
This script should take in a csv file and user selected two columns, and compute some metric
Example: python compute_metrics.py -c test.csv -gt gt_col -pred pred_col
"""
import ast
import argparse
import numpy as np
import pandas as pd
from requests import get
from tqdm.auto import tqdm
from pprint import pprint
import multiprocessing
from sklearn.metrics import accuracy_score
from scipy.spatial.distance import cosine as cosine_similarity
import chromadb
from pathlib import Path
from time import sleep
from utils.parse_arxiv import load_json, save_json
def get_ids_embed_dict(db_res_obj):
ids_embed_dict = {}
for i, (id, embed) in enumerate(zip(db_res_obj['ids'], db_res_obj['embeddings'])):
ids_embed_dict[str(id)] = embed
return ids_embed_dict
def stack_embed_inorder(res_id_list, ids_embed_dict, fill=True, verbose=False):
embed_list = []
if len(res_id_list) != len(ids_embed_dict):
print("size mismatch, {} vs {}".format(len(res_id_list), len(ids_embed_dict)))
if np.abs( len(res_id_list) - len(ids_embed_dict) ) > 1:
# if the size mismatch is too large, return False
return False
for i, id in enumerate(res_id_list):
id = str(id)
if id not in ids_embed_dict:
if verbose: print(i, id)
if fill:
id = res_id_list[i-1]
else:
return False
embed_list.append(ids_embed_dict[id])
return np.array(embed_list)
def compute_distance_metrics(gt, pred, abstract_collection):
gt_results = abstract_collection.get(ids=gt, include=["embeddings"])
pred_results = abstract_collection.get(ids=pred, include=["embeddings"])
# DEBUG
# print(len(gt), len(gt_results['ids']))
# print(len(pred), len(pred_results['ids']))
# print()
# if len(pred) != len(pred_results['ids']):
# for id in pred:
# if id not in pred_results['ids']:
# print(id)
# sleep(1)
# Cosine similarity
gt_id_embed_dict = get_ids_embed_dict(gt_results)
pred_id_embed_dict = get_ids_embed_dict(pred_results)
gt_results = stack_embed_inorder(gt, gt_id_embed_dict)
pred_results = stack_embed_inorder(pred, pred_id_embed_dict)
if gt_results is False:
print("size mismatch for gt")
return None
if pred_results is False:
print("size mismatch for pred")
return None
distances = [cosine_similarity(g, p) for g, p in zip(gt_results, pred_results)]
return np.mean(distances)
def batch_compute_distance_metrics(gt_list, pred_list, abstract_collection):
"""
Compute the percentage of ground truth that is included in the prediction for a batch of data, return a list of percentages
"""
dists = []
for gt, pred in zip(gt_list, pred_list):
if len(gt) != len(pred):
if len(dists)!=0:
print("size mismatch, curr dist not empty, add cuur mean to the list")
dists.append(np.mean(dists))
continue
# abstract_collection sometimes fail to return k results...
mean_dist = compute_distance_metrics(gt, pred, abstract_collection)
if mean_dist is not None:
dists.append(mean_dist)
else:
if len(gt) != len(pred):
print("too many mismatch, add cuur mean to the list")
dists.append(np.mean(dists))
return dists
# return [
# compute_distance_metrics(gt, pred, abstract_collection)
# for gt, pred in zip(gt_list, pred_list)
# ]
def mproc_batch_compute_distance_metrics(args):
"""
Compute the percentage of ground truth that is included in the prediction for a batch of data, multiprocess version
"""
gt_list, pred_list, distance_list, normalize, abstract_collection = args
distance_list.extend(
batch_compute_distance_metrics(gt_list, pred_list, abstract_collection)
)
# TODO: double check compute_percent_include
def compute_percent_include(gt, pred):
"""
Compute the percentage of ground truth that is included in the prediction
"""
counter = 0
for p in pred:
if p in gt:
counter += 1
# print(counter)
# print(len(gt))
return counter / len(gt)
def batch_compute_percent_include(gt_list, pred_list):
"""
Compute the percentage of ground truth that is included in the prediction for a batch of data, return a list of percentages
"""
return [compute_percent_include(gt, pred) for gt, pred in zip(gt_list, pred_list)]
def mproc_batch_compute_percent_include(args):
"""
Compute the percentage of ground truth that is included in the prediction for a batch of data, multiprocess version
"""
gt_list, pred_list, percent_include_list, normalize = args
percent_include_list.extend(batch_compute_percent_include(gt_list, pred_list))
def compute_accuracy(gt, pred, normalize=True):
"""
Compute accuracy
"""
return accuracy_score(gt, pred, normalize=normalize)
def batch_compute_accuracy(gt_list, pred_list, normalize=True):
"""
Compute accuracy for a batch of data, return a list of accuracies
"""
accs = []
for gt, pred in zip(gt_list, pred_list):
if len(gt) != len(pred):
print("size mismatch, add cuur mean to the list")
accs.append(np.mean(accs))
continue
accs.append( accuracy_score(gt, pred, normalize=normalize) )
return accs
def mproc_batch_compute_accuracy(args):
"""
Compute accuracy for a batch of data, multiprocess version
"""
gt_list, pred_list, acc_list, normalize = args
# DEBUG
# conter = 0
# for i in range(len(gt_list)):
# if len(gt_list[i]) != len(pred_list[i]):
# conter += 1
# print(i)
# print(len(gt_list[i]))
# print(len(pred_list[i]))
# print()
# print("counter: ", conter)
acc_list.extend(batch_compute_accuracy(gt_list, pred_list, normalize=normalize))
FILENAME_COL = "filename"
def get_col_name(metric_name, query_name, stat_name):
return f"{metric_name}_{query_name}_{stat_name}"
def append_results(results_df, accuracy_lists, cols_to_eval, metric_name):
cur_col = 1
for accuracy_list in accuracy_lists:
accuracy = np.array(accuracy_list)
query_name = cols_to_eval[cur_col]
row_loc = results_df.index[-1]
results_df.loc[
row_loc, get_col_name(metric_name, query_name, "mean")
] = accuracy.mean()
# results_df.loc[
# row_loc, get_col_name(metric_name, query_name, "std")
# ] = accuracy.std()
# results_df.loc[
# row_loc, get_col_name(metric_name, query_name, "max")
# ] = accuracy.max()
# results_df.loc[
# row_loc, get_col_name(metric_name, query_name, "min")
# ] = accuracy.min()
cur_col += 1
def get_cols_to_eval(args, cfg, res_dict):
potential_args_to_eval = [
cfg["vectorDBGT"]["collection_name"],
cfg["vectorDB"]["collection_name"],
args.hybrid_pred,
args.weighted_hybrid_pred,
]
cols_to_eval = []
for arg in potential_args_to_eval:
if arg is not None and arg in res_dict['0'].keys():
cols_to_eval.append(arg)
return cols_to_eval
def get_metrics_to_eval(args):
metrics_to_eval = []
if args.metrics == "distances":
metrics_to_eval.append("distances")
elif args.metrics == "accuracy":
metrics_to_eval.append("accuracy")
elif args.metrics == "include":
metrics_to_eval.append("percent_include")
else:
metrics_to_eval.append("distances")
metrics_to_eval.append("accuracy")
metrics_to_eval.append("percent_include")
return metrics_to_eval
def process_metric(args, cfg, res_dict, results_df, metric_name):
print(f"Processing {metric_name}")
cols_to_eval = get_cols_to_eval(args, cfg, res_dict)
# compute metrics
manager = multiprocessing.Manager()
lists_to_eval = []
for col in cols_to_eval:
# convert the string to a list
# output_list = df[col].apply(ast.literal_eval).tolist()
# For json
output_list = [v[col] for k, v in res_dict.items()]
lists_to_eval.append(output_list)
# print(len(output_list))
# print(len(lists_to_eval))
# exit()
# 1. Accuracy
acc_normalize = True
accuracy_lists = []
args_lists = []
for i in range(1, len(lists_to_eval)):
accuracy_lists.append(manager.list())
args_lists.append([])
# loop through the data a batch at a time
batch_size = 3000
abstract_collection = None
if metric_name == "distances":
client = chromadb.PersistentClient(path=cfg["vectorDBGT"]["root"])
abstract_collection = client.get_collection(name=cfg["vectorDBGT"]["collection_name"])
for i in tqdm(range(0, len(lists_to_eval[0]), batch_size)):
batches = []
for j in range(len(lists_to_eval)):
batches.append(lists_to_eval[j][i : i + batch_size])
for j in range(len(args_lists)):
if metric_name == "distances":
args_lists[j].append(
(
batches[0], # We assume the first batch is the ground truth
batches[j + 1],
accuracy_lists[j],
acc_normalize,
abstract_collection,
)
)
else:
args_lists[j].append(
(
batches[0], # We assume the first batch is the ground truth
batches[j + 1],
accuracy_lists[j],
acc_normalize,
)
)
if metric_name == "distances":
for args_list in args_lists:
for arg in args_list:
mproc_batch_compute_distance_metrics(arg)
append_results(results_df, accuracy_lists, cols_to_eval, "distances")
elif metric_name == "accuracy":
with multiprocessing.Pool(processes=num_processes) as pool:
for a_list in args_lists:
for arg in a_list:
pool.map(mproc_batch_compute_accuracy, [arg])
append_results(results_df, accuracy_lists, cols_to_eval, "accuracy")
elif metric_name == "percent_include":
with multiprocessing.Pool(processes=num_processes) as pool:
for args_list in args_lists:
for arg in args_list:
pool.map(mproc_batch_compute_percent_include, [arg])
append_results(results_df, accuracy_lists, cols_to_eval, "percent_include")
if __name__ == "__main__":
# get a number from the command line
parser = argparse.ArgumentParser(description="Compute metrics")
parser.add_argument("-c", "--cfg", metavar="", type=str, default=None, help="path to the config file")
parser.add_argument("-f", "--folder", type=str, required=True, help="folder to read csv file from")
parser.add_argument("-p", "--proc", type=int, default=1, help="number of processes to use")
parser.add_argument("-hd", "--hybrid_pred", type=str, default="hybrid", help="the name of the hybrid prediction column")
parser.add_argument("-whd", "--weighted_hybrid_pred", type=str, default="weighted_hybrid",
help="the name of the weighted hybrid prediction column")
parser.add_argument("-m", "--metrics", type=str, default="all",
help="Specify the metrics to use. Currently supporting accuracy, percent_include and distances")
parser.add_argument( "-s", "--save", type=str, default="stats.csv", help="file to save the results to")
args = parser.parse_args()
cfg = load_json(args.cfg)
num_processes = args.proc
metrics_to_eval = get_metrics_to_eval(args)
results_df = pd.DataFrame()
inference_res_folder = Path(args.folder)
# check save path, should be a csv file
if args.save[-4:] != ".csv":
print("Save path should be a csv file")
exit()
for f in inference_res_folder.iterdir():
if f.suffix != ".json":
continue
print(f"Processing {f.name}")
# df = pd.read_csv(f)
res_dict = load_json(f)
# Add the column names
if results_df.shape[1] == 0:
columns = ["filename"]
cols_to_eval = get_cols_to_eval(args, cfg, res_dict)
for metric in metrics_to_eval:
for col in cols_to_eval[1:]:
columns.append(get_col_name(metric, col, "mean"))
# columns.append(get_col_name(metric, col, "std"))
# columns.append(get_col_name(metric, col, "max"))
# columns.append(get_col_name(metric, col, "min"))
results_df = pd.DataFrame(columns=columns)
row = [f.name]
for i in range(results_df.shape[1] - 1):
row.append(np.nan)
# results_df = results_df.append(
# pd.Series(row, index=results_df.columns), ignore_index=True
# )
new_row = pd.Series(row, index=results_df.columns)
results_df = pd.concat([results_df, pd.DataFrame([new_row])], ignore_index=True)
for metric in metrics_to_eval:
process_metric(args, cfg, res_dict, results_df, metric)
# Save the results
results_df.to_csv(args.save, index=False)