File size: 20,133 Bytes
96c0ca2 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 |
#!/usr/bin/env python
# encoding: utf-8
'''
@license: (C) Copyright 2021, Hey.
@author: Hey
@email: sanyuan.**@**.com
@tel: 137****6540
@datetime: 2022/11/26 21:05
@project: LucaOne
@file: metrics.py
@desc: metrics for binary classification or multi-class classification
'''
import csv
import numpy as np
import matplotlib.pyplot as plt
plt.rcParams.update({'font.size': 18})
plt.rcParams['axes.unicode_minus'] = False
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, \
average_precision_score, confusion_matrix, mean_absolute_error, mean_squared_error, r2_score
def topk_accuracy_score(targets, probs, k=3):
'''
topk accuracy
:param targets:
:param probs:
:param k:
:return:
'''
# obtain top-k label
max_k_preds = probs.argsort(axis=1)[:, -k:][:, ::-1]
a_real = np.resize(targets, (targets.shape[0], 1))
# obtain the match result
match_array = np.logical_or.reduce(max_k_preds == a_real, axis=1)
topk_acc_score = match_array.sum() / match_array.shape[0]
return topk_acc_score
def multi_class_acc(targets, probs, threshold=0.5):
if targets.ndim == 2:
targets = np.argmax(targets, axis=1)
preds = np.argmax(probs, axis=1)
return accuracy_score(targets, preds)
def multi_class_precision(targets, probs, average='macro'):
if targets.ndim == 2:
targets = np.argmax(targets, axis=1)
preds = np.argmax(probs, axis=1)
return precision_score(targets, preds, average=average)
def multi_class_recall(targets, probs, average='macro'):
if targets.ndim == 2:
targets = np.argmax(targets, axis=1)
preds = np.argmax(probs, axis=1)
return recall_score(targets, preds, average=average)
def multi_class_f1(targets, probs, average='macro'):
if targets.ndim == 2:
targets = np.argmax(targets, axis=1)
preds = np.argmax(probs, axis=1)
return f1_score(targets, preds, average=average)
def multi_class_roc_auc(targets, probs, average='macro'):
if targets.ndim == 2:
targets = np.argmax(targets, axis=1)
return roc_auc_score(targets, probs, average=average, multi_class='ovr')
def multi_class_pr_auc(targets, probs, average='macro'):
if targets.ndim == 2:
targets = np.argmax(targets, axis=1)
z = probs.shape[1]
new_targets = np.eye(z)[targets]
pr_auc = average_precision_score(new_targets, probs, average=average)
return pr_auc
def metrics_multi_class(targets, probs, average="macro"):
'''
metrics of multi-class classification
:param targets: 1d-array class index (n_samples, )
:param probs: 2d-array probability (n_samples, m_classes)
:return:
'''
if targets.ndim == 2 and targets.shape[1] > 1:
targets = np.argmax(targets, axis=1)
elif targets.ndim == 2 and targets.shape[1] == 1:
targets = np.squeeze(targets, axis=1)
preds = np.argmax(probs, axis=1)
acc = accuracy_score(targets, preds)
prec = precision_score(targets, preds, average=average)
recall = recall_score(targets, preds, average=average)
f1 = f1_score(targets, preds, average=average)
result = {
"acc": round(float(acc), 6),
"prec": round(float(prec), 6),
"recall": round(float(recall), 6),
"f1": round(float(f1), 6)
}
result.update({
"top2_acc": round(float(topk_accuracy_score(targets, probs, k=2)), 6),
"top3_acc": round(float(topk_accuracy_score(targets, probs, k=3)), 6),
"top5_acc": round(float(topk_accuracy_score(targets, probs, k=5)), 6),
"top10_acc": round(float(topk_accuracy_score(targets, probs, k=10)), 6)
})
try:
roc_auc = roc_auc_score(targets, probs, average=average, multi_class='ovr')
result.update({
"roc_auc": round(float(roc_auc), 6)
})
except Exception as e:
pass
try:
z = probs.shape[1]
new_targets = np.eye(z)[targets]
pr_auc = average_precision_score(new_targets, probs, average=average)
result.update({
"pr_auc": round(float(pr_auc), 6),
})
except Exception as e:
pass
return result
def metrics_multi_class_for_pred(targets, preds, probs=None, average="macro", savepath=None):
'''
metrcis for multi-class classification
:param targets: 1d-array class index (n_samples, )
:param preds: 1d-array class index (n_samples, )
:return:
'''
if targets.ndim == 2 and targets.shape[1] > 1:
targets = np.argmax(targets, axis=1)
elif targets.ndim == 2 and targets.shape[1] == 1:
targets = np.squeeze(targets, axis=1)
acc = accuracy_score(targets, preds)
prec = precision_score(targets, preds, average=average)
recall = recall_score(targets, preds, average=average)
f1 = f1_score(y_true=targets, y_pred=preds, average=average)
result = {
"acc": round(float(acc), 6),
"prec": round(float(prec), 6),
"recall": round(float(recall), 6),
"f1": round(float(f1), 6)
}
try:
roc_auc = roc_auc_score(targets, probs, average=average, multi_class='ovr')
result.update({
"roc_auc": round(float(roc_auc), 6)
})
except Exception as e:
pass
try:
z = probs.shape[1]
new_targets = np.eye(z)[targets]
pr_auc = average_precision_score(new_targets, probs, average=average)
result.update({
"pr_auc": round(float(pr_auc), 6),
})
except Exception as e:
pass
return result
def metrics_regression(targets, preds):
'''
metrcis for regression
:param targets: 1d-array class index (n_samples, )
:param preds: 1d-array class index (n_samples, )
:return:
'''
mae = mean_absolute_error(targets, preds)
mse = mean_squared_error(targets, preds)
r2 = r2_score(targets, preds)
return {
"mae": round(float(mae), 6),
"mse": round(float(mse), 6),
"r2": round(float(r2), 6)
}
def transform(targets, probs, threshold):
'''
metrics of binary classification
:param targets: 1d-array class index (n_samples, )
:param probs: 1d-array larger class probability (n_samples, )
:param threshold: 0-1 prob threshokd
:return:
'''
if targets.ndim == 2:
if targets.shape[1] == 2: # [[0, 1], [1, 0]]
targets = np.argmax(targets, axis=1)
else: # [[1], [0]]
targets = targets.flatten()
if probs.ndim == 2:
if probs.shape[1] == 2: # [[0.1, 0.9], [0.9, 0.1]]
preds = np.argmax(probs, axis=1)
probs = probs[:, 1].flatten()
else: # [[0.9], [0.1]]
preds = (probs >= threshold).astype(int).flatten()
probs = probs.flatten()
else:
preds = (probs >= threshold).astype(int)
return targets, probs, preds
def binary_acc(targets, probs, threshold=0.5):
targets, probs, preds = transform(targets, probs, threshold)
return accuracy_score(targets, preds)
def binary_precision(targets, probs, threshold=0.5, average='binary'):
targets, probs, preds = transform(targets, probs, threshold)
return precision_score(targets, preds, average=average)
def binary_recall(targets, probs, threshold=0.5, average='binary'):
targets, probs, preds = transform(targets, probs, threshold)
return recall_score(targets, preds, average=average)
def binary_f1(targets, probs, threshold=0.5, average='binary'):
targets, probs, preds = transform(targets, probs, threshold)
return f1_score(targets, preds, average=average)
def binary_roc_auc(targets, probs, threshold=0.5, average='macro'):
targets, probs, preds = transform(targets, probs, threshold)
return roc_auc_score(targets, probs, average=average)
def binary_pr_auc(targets, probs, threshold=0.5, average='macro'):
targets, probs, preds = transform(targets, probs, threshold)
return average_precision_score(targets, probs, average=average)
def binary_confusion_matrix(targets, probs, threshold=0.5, savepath=None):
targets, probs, preds = transform(targets, probs, threshold)
cm_obj = confusion_matrix(targets, preds, labels=[0, 1])
plot_confusion_matrix_for_binary_class(targets, preds, cm=cm_obj, savepath=savepath)
tn, fp, fn, tp = cm_obj.ravel()
cm = {"tn": int(tn), "fp": int(fp), "fn": int(fn), "tp": int(tp)}
return cm
def metrics_binary(targets, probs, threshold=0.5, average="binary", savepath=None):
'''
metrics for binary classification
:param targets: 1d-array class index (n_samples, )
:param probs: 1d-array larger class probability (n_samples, )
:param threshold: 0-1 prob threshold
:return:
'''
if targets.ndim == 2:
if targets.shape[1] == 2: # [[0, 1], [1, 0]]
targets = np.argmax(targets, axis=1)
else: # [[1], [0]]
targets = targets.flatten()
if probs.ndim == 2:
if probs.shape[1] == 2: # [[0.1, 0.9], [0.9, 0.1]]
preds = np.argmax(probs, axis=1)
probs = probs[:, 1].flatten()
else: # [[0.9], [0.1]]
preds = (probs >= threshold).astype(int).flatten()
probs = probs.flatten()
else:
preds = (probs >= threshold).astype(int)
acc = accuracy_score(targets, preds)
prec = precision_score(targets, preds, average=average)
recall = recall_score(targets, preds, average=average)
f1 = f1_score(targets, preds, average=average)
result = {
"acc": round(float(acc), 6),
"prec": round(float(prec), 6),
"recall": round(float(recall), 6),
"f1": round(float(f1), 6)
}
try:
roc_auc = roc_auc_score(targets, probs, average="macro")
result.update({
"roc_auc": round(float(roc_auc), 6)
})
except Exception as e:
pass
try:
pr_auc = average_precision_score(targets, probs, average="macro")
result.update({
"pr_auc": round(float(pr_auc), 6)
})
except Exception as e:
pass
try:
cm_obj = confusion_matrix(targets, preds, labels=[0, 1])
plot_confusion_matrix_for_binary_class(targets, preds, cm=cm_obj, savepath=savepath)
tn, fp, fn, tp = cm_obj.ravel()
cm = {"tn": int(tn), "fp": int(fp), "fn": int(fn), "tp": int(tp)}
result.update({
"confusion_matrix": cm
})
except Exception as e:
pass
# add mcc
try:
tn, fp, fn, tp = cm["tn"], cm["fp"], cm["fn"], cm["tp"]
mcc = (tn*tp - fp*fn) / (((tp+fp)*(tp+fn)*(tn+fp)*(tn+fn)) ** 0.5)
result.update({
"mcc": round(mcc, 6)
})
except Exception as e:
pass
return result
def metrics_binary_for_pred(targets, preds, probs=None, average="binary", savepath=None):
'''
metrics for binary classification
:param targets: 1d-array class index (n_samples, )
:param preds: 1d-array larger class index (n_samples, )
:return:
'''
if targets.ndim == 2:
if targets.shape[1] == 2: # [[1, 0], [0, 1]
targets = np.argmax(targets, axis=1)
else: # [[1], [0]]
targets = targets.flatten()
if preds.ndim == 2:
if preds.shape[1] == 2: # [[0.9, 0.1], [0.1, 0.9]]
preds = np.argmax(preds, axis=1)
else: # [[0], [1]]
preds = preds.flatten()
cm_obj = confusion_matrix(targets, preds, labels=[0, 1])
plot_confusion_matrix_for_binary_class(targets, preds, cm=cm_obj, savepath=savepath)
tn, fp, fn, tp = cm_obj.ravel()
cm = {"tn": int(tn), "fp": int(fp), "fn": int(fn), "tp": int(tp)}
if len(np.unique(targets)) > 1:
acc = accuracy_score(targets, preds)
prec = precision_score(targets, preds, average=average)
recall = recall_score(targets, preds, average=average)
f1 = f1_score(y_true=targets, y_pred=preds, average=average)
result = {
"acc": round(float(acc), 6),
"prec": round(float(prec), 6),
"recall": round(float(recall), 6),
"f1": round(float(f1), 6)
}
else:
result = {
"acc": round(float((cm["tp"] + cm["tn"]) / (cm["tp"] + cm["tn"] + cm["fp"] + cm["fn"])), 6),
"prec": round(float(cm["tp"]/(cm["tp"] + cm["fp"]) if cm["tp"] + cm["fp"] > 0 else 1.0), 6),
"recall": round(float(cm["tp"]/(cm["tp"] + cm["fn"]) if cm["tp"] + cm["fn"] > 0 else 1.0), 6),
}
result["f1"] = 2 * result["prec"] * result["recall"] / (result["prec"] + result["recall"])
try:
pr_auc = average_precision_score(targets, probs, average="macro")
result.update({
"pr_auc": round(float(pr_auc), 6)
})
except Exception as e:
pass
try:
roc_auc = roc_auc_score(targets, probs, average="macro")
result.update({
"roc_auc": round(float(roc_auc), 6)
})
except Exception as e:
pass
try:
tn, fp, fn, tp = cm["tn"], cm["fp"], cm["fn"], cm["tp"]
mcc = (tn*tp - fp*fn) / (((tp+fp)*(tp+fn)*(tn+fp)*(tn+fn)) ** 0.5)
result.update({
"mcc": round(mcc, 6)
})
except Exception as e:
pass
result.update({
"confusion_matrix": cm
})
return result
def write_error_samples_multi_class(filepath, samples, input_indexs, input_id_2_names, output_id_2_name, targets, probs,
use_other_diags=False, use_other_operas=False, use_checkin_department=False):
'''
write the bad cases of multi-class classification
:param filepath:
:param samples:
:param input_indexs:
:param input_id_2_names:
:param output_id_2_name:
:param targets:
:param probs:
:param use_other_diags:
:param use_other_operas:
:param use_checkin_department:
:return:
'''
targets = np.argmax(targets, axis=1)
preds = np.argmax(probs, axis=1)
with open(filepath, "w") as fp:
writer = csv.writer(fp)
writer.writerow(["score", "y_true", "y_pred", "inputs"])
for i in range(len(targets)):
target = targets[i]
pred = preds[i]
score = 1
if target != pred:
score = 0
if output_id_2_name:
target_label = output_id_2_name[target]
pred_label = output_id_2_name[pred]
else:
target_label = target
pred_label = pred
sample = samples[i]
if input_id_2_names:
new_sample = []
for idx, input_index in enumerate(input_indexs):
if input_index == 3 and not use_checkin_department:
input_index = 12
new_sample.append([input_id_2_names[idx][v] for v in sample[input_index]])
if (input_index == 6 and use_other_diags) or (input_index == 8 and use_other_operas) or (input_index == 10 and use_other_diags):
new_sample.append([input_id_2_names[idx][v] for v in sample[input_index + 1]])
else:
new_sample = sample
row = [score, target_label, pred_label, new_sample]
writer.writerow(row)
def write_error_samples_binary(filepath, samples, input_indexs, input_id_2_names, targets, probs, threshold=0.5,
use_other_diags=False, use_other_operas=False, use_checkin_department=False):
'''
write bad cases of binary classification
:param filepath:
:param samples:
:param input_indexs:
:param input_id_2_names:
:param targets:
:param probs:
:param threshold:
:param use_other_diags:
:param use_other_operas:
:param use_checkin_department:
:return:
'''
with open(filepath, "w") as fp:
writer = csv.writer(fp)
writer.writerow(["score", "y_true", "y_pred", "inputs"])
for i in range(len(targets)):
target = targets[i][0]
if target != 1:
target = 1
prob = probs[i][0]
if prob >= threshold:
pred = 1
else:
pred = 0
score = 1
if target != pred:
score = 0
target_label = "True" if target == 1 else "False"
pred_label = "True" if target == 1 else "False"
sample = samples[i]
if input_id_2_names:
new_sample = []
for idx, input_index in enumerate(input_indexs):
if input_index == 3 and not use_checkin_department:
input_index = 12
new_sample.append([input_id_2_names[idx][v] for v in sample[input_index]])
if (input_index == 6 and use_other_diags) or (input_index == 8 and use_other_operas) or (input_index == 10 and use_other_diags):
new_sample.append([input_id_2_names[idx][v] for v in sample[input_index + 1]])
else:
new_sample = sample
row = [score, target_label, pred_label, new_sample]
writer.writerow(row)
def plot_confusion_matrix_for_binary_class(targets, preds, cm=None, savepath=None):
'''
:param targets: ground truth
:param preds: prediction probs
:param cm: confusion matrix
:param savepath: confusion matrix picture savepth
'''
plt.figure(figsize=(40, 20), dpi=100)
if cm is None:
cm = confusion_matrix(targets, preds, labels=[0, 1])
plt.matshow(cm, cmap=plt.cm.Oranges)
plt.colorbar()
for x in range(len(cm)):
for y in range(len(cm)):
plt.annotate(cm[x, y], xy=(y, x), verticalalignment='center', horizontalalignment='center')
plt.ylabel('True')
plt.xlabel('Prediction')
if savepath:
plt.savefig(savepath, dpi=100)
else:
plt.show()
plt.close("all")
if __name__ == "__main__":
'''multi_class'''
targets = np.array([0, 1, 2, 1, 3])
probs = np.array([[0.9, 0.05, 0.05, 0], [0.5, 0.45, 0.05, 0], [0.4, 0.05, 0.55, 0], [0.1, 0.55, 0.25, 0.1], [0.4, 0.25, 0.35, 0]])
print(metrics_multi_class(targets, probs))
targets = np.array([0, 1, 2, 3, 3])
probs = np.array([[0.9, 0.05, 0.05, 0], [0.5, 0.45, 0.05, 0], [0.4, 0.05, 0.55, 0], [0.1, 0.25, 0.25, 0.4], [0.1, 0.25, 0.25, 0.4]])
print(metrics_multi_class(targets, probs))
targets = np.array([[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1], [0, 0, 0, 1]])
probs = np.array([[0.9, 0.05, 0.05, 0], [0.5, 0.45, 0.05, 0], [0.4, 0.05, 0.55, 0], [0.1, 0.25, 0.25, 0.4], [0.1, 0.25, 0.25, 0.4]])
print(metrics_multi_class(targets, probs))
'''binary'''
targets = np.array([0, 0, 1, 1])
probs = np.array([[0.1], [0.1], [0.1], [0.9]])
print(metrics_binary(targets, probs))
targets = np.array([[0], [0], [1], [1]])
probs = np.array([[0.1], [0.1], [0.1], [0.9]])
print(metrics_binary(targets, probs))
targets = np.array([0, 0, 1, 1])
probs = np.array([[0.1, 0.1, 0.1, 0.9]])
print(metrics_binary(targets, probs))
targets = np.array([0, 0, 1, 1])
probs = np.array([0.1, 0.1, 0.1, 0.9])
print(metrics_binary(targets, probs))
targets = np.array([0, 1, 2, 1, 3])
probs = np.array([[0.9, 0.05, 0.05, 0], [0.5, 0.45, 0.05, 0], [0.4, 0.05, 0.55, 0], [0.1, 0.55, 0.25, 0.1], [0.4, 0.25, 0.25, 0.1]])
z = probs.shape[1]
# print(z)
print(np.eye(z))
new_targets = np.eye(z)[targets]
print(new_targets)
|