-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_baseline_sk.py
332 lines (276 loc) · 11.8 KB
/
run_baseline_sk.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
#!/usr/bin/env python3
# Author: Armit
# Create Time: 2023/05/05
import pickle as pkl
from pathlib import Path
from argparse import ArgumentParser
from traceback import print_exc
import warnings ; warnings.simplefilter("ignore")
import jieba
import numpy as np
import matplotlib.pyplot as plt
try:
from sklearn.decomposition import PCA, KernelPCA
from sklearn.manifold import TSNE
try:
# not all components are safely patchable, leave them unhacked
from sklearnex import patch_sklearn ; patch_sklearn()
except:
print_exc()
print('>> sklearnex not installed, performance may be slow')
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.neighbors import KNeighborsClassifier
from sklearn.tree import DecisionTreeClassifier, ExtraTreeClassifier
from sklearn.ensemble import RandomForestClassifier, ExtraTreesClassifier, VotingClassifier, BaggingClassifier, AdaBoostClassifier, GradientBoostingClassifier, HistGradientBoostingClassifier
from sklearn.naive_bayes import BernoulliNB, GaussianNB, MultinomialNB, ComplementNB, CategoricalNB
from sklearn.linear_model import RidgeClassifierCV, LogisticRegressionCV, PassiveAggressiveClassifier, SGDClassifier
from sklearn.svm import SVC, NuSVC, LinearSVC
from sklearn.neural_network import MLPClassifier, BernoulliRBM
from sklearn.metrics import precision_recall_fscore_support, confusion_matrix
except:
print_exc()
print('>> sklearn not installed, some of the features may not work')
try:
import fasttext.util
from fasttext.FastText import _FastText as FastText
except:
print_exc()
print('>> fasttext not installed, some of the features may not work')
from utils import *
from mk_vocab import make_tokenizer
FASTTEXT_CKPT_PATH = DATA_PATH / 'cc.zh.300.bin'
FEATURES = [
'tfidf',
'fasttext',
]
ANALYZERS = [
'char',
'word',
'sent',
'2gram',
'3gram',
'kgram',
]
MODELS = {
'knn-5': lambda: KNeighborsClassifier(n_neighbors=5),
'knn-7': lambda: KNeighborsClassifier(n_neighbors=7),
'knn-10': lambda: KNeighborsClassifier(n_neighbors=10),
'knn-20': lambda: KNeighborsClassifier(n_neighbors=20),
'dt': lambda: DecisionTreeClassifier(max_depth=8),
'et': lambda: ExtraTreeClassifier(max_depth=8),
'rf': lambda: RandomForestClassifier(n_estimators=16),
'ets': lambda: ExtraTreesClassifier(n_estimators=16, max_depth=3),
# 'vote': lambda: VotingClassifier(),
# 'bag': lambda: BaggingClassifier(),
'adabst': lambda: AdaBoostClassifier(n_estimators=16),
'gbst': lambda: GradientBoostingClassifier(n_estimators=16, max_depth=3),
'hgbst': lambda: HistGradientBoostingClassifier(max_depth=3),
'hgbst': lambda: BernoulliNB(),
'gnb': lambda: GaussianNB(),
'mnb': lambda: MultinomialNB(),
'cnb': lambda: ComplementNB(),
# 'catnb': lambda: CategoricalNB(),
'ridge': lambda: RidgeClassifierCV(alphas=[1e-3, 1e-2, 1e-1, 1]),
'log': lambda: LogisticRegressionCV(),
'paclf': lambda: PassiveAggressiveClassifier(),
'sgdclf': lambda: SGDClassifier(),
'svc-l': lambda: SVC(kernel='linear'),
'svc-p': lambda: SVC(kernel='poly'),
'svc-r': lambda: SVC(kernel='rbf'),
'svc-s': lambda: SVC(kernel='sigmoid'),
'nusvc-l': lambda: NuSVC(kernel='linear'),
'nusvc-p': lambda: NuSVC(kernel='poly'),
'nusvc-r': lambda: NuSVC(kernel='rbf'),
'nusvc-s': lambda: NuSVC(kernel='sigmoid'),
'mlp-d64-s50': lambda: MLPClassifier(hidden_layer_sizes=[64], max_iter=50),
'mlp-d64-s100': lambda: MLPClassifier(hidden_layer_sizes=[64], max_iter=100),
'mlp-d16-s50': lambda: MLPClassifier(hidden_layer_sizes=[16], max_iter=50),
'mlp-d16-s100': lambda: MLPClassifier(hidden_layer_sizes=[16], max_iter=100),
# 'rbm': lambda: BernoulliRBM(n_components=256),
}
def run_tfidf(analyzer:str) -> Datasets:
''' This should be more like syntaxical feature '''
assert analyzer in ['char', 'word'] or analyzer.endswith('gram')
def process_data(split:str, tfidfvec:TfidfVectorizer) -> Tuple[NDArray, NDArray]:
T, Y = load_dataset(split)
tfidf = tfidfvec.fit_transform(T) if split == 'train' else tfidfvec.transform(T)
X = tfidf.todense() # [N=1600, K=3386]
if isinstance(X, np.matrix): X = X.A
return X, Y
if analyzer == 'char':
tokenizer = None
stop_words = STOP_WORDS_CHAR
elif analyzer == 'word':
tokenizer = jieba.lcut_for_search
stop_words = STOP_WORDS_WORD
elif analyzer.endswith('gram'):
tokenizer = make_tokenizer(LOG_PATH / analyzer / 'vocab.txt')
stop_words = None
analyzer = 'word' # NOTE: overrides
tfidfvec = TfidfVectorizer(analyzer=analyzer, tokenizer=tokenizer, stop_words=stop_words)
X_train, Y_train = process_data('train', tfidfvec)
X_test, Y_test = process_data('test', tfidfvec)
X_valid, Y_valid = process_data('valid', tfidfvec)
return (X_train, Y_train), (X_test, Y_test), (X_valid, Y_valid)
def run_fasttext(analyzer:str) -> Datasets:
''' This should be more like semantical feature '''
assert analyzer in ['char', 'word', 'sent'] or analyzer.endswith('gram')
if not FASTTEXT_CKPT_PATH.exists():
import shutil
fasttext.util.download_model('zh', if_exists='ignore')
BASE_PATH = Path(__file__).absolute()
shutil.move(BASE_PATH / 'cc.zh.300.bin', DATA_PATH)
shutil.move(BASE_PATH / 'cc.zh.300.bin.gz', DATA_PATH)
embed: FastText = fasttext.load_model(str(FASTTEXT_CKPT_PATH))
def process_data(split:str) -> Dataset:
T, Y = load_dataset(split)
if analyzer == 'char':
X = [np.stack([embed.get_word_vector(w) for w in list(t) if w in embed and w not in STOP_WORDS_CHAR], axis=0).mean(axis=0) for t in T]
elif analyzer == 'word':
X = [np.stack([embed.get_word_vector(w) for w in jieba.cut_for_search(t) if w in embed and w not in STOP_WORDS_WORD], axis=0).mean(axis=0) for t in T]
elif analyzer == 'sent':
X = [embed.get_sentence_vector(t) for t in T]
elif analyzer.endswith('gram'):
tokenizer = make_tokenizer(LOG_PATH / analyzer / 'vocab.txt')
X = [np.stack([(embed.get_word_vector(w) if w in embed else embed.get_sentence_vector(w)) for w in tokenizer(t)], axis=0).mean(axis=0) for t in T]
return np.stack(X, axis=0), Y
X_train, Y_train = process_data('train')
X_test, Y_test = process_data('test')
X_valid, Y_valid = process_data('valid')
return (X_train, Y_train), (X_test, Y_test), (X_valid, Y_valid)
def run_visualize(datasets:Datasets, name:str, out_dp:Path):
(X_train, Y_train), (X_test, Y_test), (X_valid, Y_valid) = datasets
def save_plot(fp:Path, z_train, z_test, z_valid, s=1):
nonlocal Y_train, Y_test, Y_valid
plt.subplot(221)
plt.scatter(z_train[:, 0], z_train[:, 1], s, c=Y_train, marker='o', alpha=0.7, label='train')
plt.scatter(z_test [:, 0], z_test [:, 1], s, c=Y_test, marker='x', alpha=0.7, label='test')
plt.scatter(z_valid[:, 0], z_valid[:, 1], s, c=Y_valid, marker='*', alpha=0.7, label='valid')
plt.axis('off')
plt.title('all')
plt.subplot(222)
plt.scatter(z_valid[:, 0], z_valid[:, 1], s, c=Y_valid, marker='*', alpha=0.7, label='valid')
plt.axis('off')
plt.title('valid')
plt.subplot(223)
plt.scatter(z_train[:, 0], z_train[:, 1], s, c=Y_train, marker='o', alpha=0.7, label='train')
plt.axis('off')
plt.title('train')
plt.subplot(224)
plt.scatter(z_test [:, 0], z_test [:, 1], s, c=Y_test, marker='x', alpha=0.7, label='test')
plt.axis('off')
plt.title('test')
plt.suptitle(fp.stem)
plt.tight_layout()
plt.subplots_adjust()
plt.savefig(fp, dpi=600)
if 'pca':
pca = PCA(n_components=2)
z_train = pca.fit_transform(X_train)
z_test = pca. transform(X_test)
z_valid = pca. transform(X_valid)
save_plot(out_dp / f'pca_{name}.png', z_train, z_test, z_valid)
if 'kpca': # NOTE: just alike pca, not giving anything new
for k in ['linear', 'poly', 'rbf', 'sigmoid', 'cosine']:
kpca = KernelPCA(n_components=2, kernel=k)
z_train = kpca.fit_transform(X_train)
z_test = kpca. transform(X_test)
z_valid = kpca. transform(X_valid)
save_plot(out_dp / f'kpca-{k}_{name}.png', z_train, z_test, z_valid)
if 'tsne':
tsne = TSNE(n_components=2)
z_all = tsne.fit_transform(np.concatenate([X_train, X_test, X_valid], axis=0))
cp = len(X_train)
cp2 = cp + len(X_test)
save_plot(out_dp / f'tsne_{name}.png', z_all[:cp, :], z_all[cp:cp2, :], z_all[cp2:, :])
def run_model(name, model, datasets:Datasets, logger:Logger) -> Scores:
(X_train, Y_train), (X_test, Y_test), (X_valid, Y_valid) = datasets
model.fit(X_train, Y_train)
precs, recalls, f1s, cmats = [], [], [], []
logger.info(f'[{name}]')
for split in SPLITS:
logger.info(f'<{split}>')
X_split = locals().get(f'X_{split}')
Y_split = locals().get(f'Y_{split}')
Y_pred = model.predict(X_split)
prec, recall, f1, _ = precision_recall_fscore_support(Y_split, Y_pred, average=None)
cmat = confusion_matrix(Y_split, Y_pred)
precs .append(prec)
recalls.append(recall)
f1s .append(f1)
cmats .append(cmat)
logger.info(f'>> prec: {prec}')
logger.info(f'>> recall: {recall}')
logger.info(f'>> f1: {f1}')
logger.info('>> confusion_matrix:')
logger.info(cmat)
logger.info('-' * 32)
logger.info('\n')
return precs, recalls, f1s, cmats
def go_train(args):
out_dp: Path = LOG_PATH / args.analyzer / args.feature
out_dp.mkdir(exist_ok=True, parents=True)
datasets: Datasets = globals()[f'run_{args.feature}'](args.analyzer)
run_visualize(datasets, f'{args.feature}-{args.analyzer}', out_dp)
logger = get_logger(out_dp / 'run.log', mode='w')
result = { }
for name, model_fn in MODELS.items():
print(f'<< running {name}...')
try:
logger.info(f'exp: {args.feature}-{args.analyzer}-{name}')
precs, recalls, f1s, cmats = run_model(name, model_fn(), datasets, logger)
result[name] = {
'prec': precs,
'recall': recalls,
'f1': f1s,
'cmat': cmats,
}
except: print_exc()
with open(out_dp / 'result.pkl', 'wb') as fh:
pkl.dump(result, fh)
def go_eval(args):
for feature in FEATURES:
for analyzer in ANALYZERS:
out_dp = LOG_PATH / analyzer / feature
if not out_dp.exists(): continue
with open(out_dp / 'result.pkl', 'rb') as fh:
result = pkl.load(fh)
names, f1s = [], []
for name, scores in result.items():
names.append(name)
f1s.append(np.stack(scores['f1'], axis=-1))
f1s = np.stack(f1s, axis=0) # [n_model=30, n_cls=4, n_split=3]
plt.clf()
plt.figure(figsize=(6, 8))
n_fig = f1s.shape[-1]
for i in range(n_fig):
plt.subplot(n_fig, 1, i+1)
for j in range(f1s.shape[-2]): # each class
plt.plot(f1s[:, j, i], label=j)
plt.title(SPLITS[i])
plt.legend(loc=4, prop={'size': 6})
plt.xticks(ticks=range(len(names)), labels=names, rotation=90, ha='right')
plt.suptitle(f'f1-score {feature}-{analyzer}')
plt.tight_layout()
savefig(LOG_PATH / f'scores_{feature}_{analyzer}.png')
plt.clf()
plt.figure(figsize=(6, 8))
n_fig = f1s.shape[-1]
for i in range(n_fig):
plt.subplot(n_fig, 1, i+1)
plt.plot(f1s[:, :, i].mean(axis=1), label=j)
plt.title(SPLITS[i])
plt.xticks(ticks=range(len(names)), labels=names, rotation=90, ha='right')
plt.suptitle(f'f1-score avg. {feature}-{analyzer}')
plt.tight_layout()
savefig(LOG_PATH / f'scores_{feature}_{analyzer}-avg.png')
if __name__ == '__main__':
parser = ArgumentParser()
parser.add_argument('-L', '--analyzer', choices=ANALYZERS, help='tokenize level')
parser.add_argument('-F', '--feature', choices=FEATURES, help='input feature')
parser.add_argument('--eval', action='store_true', help='compare result scores')
args = parser.parse_args()
if args.eval:
go_eval(args)
exit(0)
go_train(args)