-
Notifications
You must be signed in to change notification settings - Fork 0
/
eval.py
317 lines (255 loc) · 12.3 KB
/
eval.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
from scipy.linalg import norm
from sklearn.metrics import confusion_matrix, classification_report, accuracy_score, f1_score
from networks.model import primary_encoder_v2_no_pooler_for_con
import gc
import numpy as np
import pandas as pd
import torch
from transformers import BertTokenizer
import warnings
from util_visualization import get_dataset_labels, clean_tweet, create_labels, plot_tsne_pca
def predictions(df, emb_dataframe, df_lab, hard, nb):
print("calcolo True false")
labels_pred = []
labs_0 = []
labs_grou = []
for i in range(len(df)):
vote_uno = 0
vote_zero = 0
# prendo riga del dataframe riferita al tweet di test di cui voglio fare la predizione
row = emb_dataframe.iloc[[i]]
copy_lab = df_lab.copy()
# Ottengo gli indici dei 3 tweet più simili e l'indice del tweet meno simile
ind_max = np.argmax(row)
ind_min = np.argmin(row)
if copy_lab[ind_min + 1] == 0:
label_reverse = 1
else:
label_reverse = 0
y = np.argsort(row)
ind_max2 = y[0][-2]
ind_max3 = y[0][-3]
# faccio votazione tra i 35 valori più simili
for k in range(1, nb): # 36
if copy_lab[y[0][-k] + 1] == 1:
vote_uno = vote_uno + 1
else:
vote_zero = vote_zero + 1
# Se la differenza di voti è minore di hard, sono in uno stato di indecisione e predico disagreement
# Se no faccio append normalmente
if abs(vote_uno - vote_zero) < hard:
labels_pred.append(0)
elif vote_uno > vote_zero:
labels_pred.append(1)
else:
labels_pred.append(0)
return labels_pred
#Funzione che calcola la matrice di similarità tra ogni istanza di test e tutte le istanze di train restituisce in output poi un datatframe
def cal_distances(df,word_embeddings_train,word_embeddings_test):
print("Calcolo matrice distanze embeddings")
warnings.filterwarnings("ignore")
matrix = []
for i in range(len(df)):
col = []
for j in range(len(word_embeddings_train)):
text1 = word_embeddings_train[j]
text2 = word_embeddings_test[i]
# cosine similarity
distance = np.dot(text1, text2) / (norm(text1) * norm(text2))
#distance = np.linalg.norm(text1 - text2) # euclidean distance
col.append(distance)
matrix.append(col)
emb_dataframe = pd.DataFrame(matrix)
return emb_dataframe
# Funzione che estrae gli embeddings dal modello
def create_embeddings(df,layers):
word_embeddings = []
base_dir = './Disagreement/Dataset'
tokenizer = BertTokenizer.from_pretrained(base_dir + '/TokenzierBert')
for idx in range(1, len(df) + 1):
with (torch.no_grad()):
input_dict = tokenizer(df["text"][idx], return_tensors="pt", padding='max_length', truncation=True,
max_length=128)
input_dict.to(device)
df = df.tail(-1)
"""hidden_states, _ = model.get_cls_features_ptrnsp(input_dict["input_ids"], input_dict["attention_mask"])
del input_dict
word_embeddings.append(hidden_states.cpu().numpy()[0])"""
_,_,hidden_states = model.get_cls_features_ptrnsp(input_dict["input_ids"], input_dict["attention_mask"])
del input_dict
if layers != 1:
hidden_states = hidden_states[-layers:] # Seleziona gli ultimi n layer
layer_average = torch.mean(torch.stack(hidden_states), dim=0) # Media dei layer
sentence_embeddings = torch.mean(layer_average, dim=1).squeeze() # Media dei token
else:
sentence_embeddings = torch.mean(hidden_states[-1], dim=1).squeeze()
sentence_embeddings = sentence_embeddings.cpu().numpy()
word_embeddings.append(sentence_embeddings)
return np.array(word_embeddings)
#funzione che calcola le performance
def prediction_metrics(lab_test, labels_pred):
matrix = confusion_matrix(lab_test, labels_pred)
print("Confusion Matrix:\n", matrix)
print("Classification Report:\n", classification_report(lab_test, labels_pred))
print("Total Accuracy:\n", accuracy_score(lab_test, labels_pred))
print("Equals Only\n")
return accuracy_score(lab_test, labels_pred)
#funzione che crea dataframe delle similarità tra gli embeddings
def create_dataframe_distances(df_tot_train,df_test,file_name,plot_graph,save,layers):
# Creo embeddings
print("Inizio a creare le feature del Test")
word_embeddings_train = create_embeddings(df_tot_train,layers)
word_embeddings_test = create_embeddings(df_test,layers)
col_train = create_labels(df_tot_train["hard_label"])
col_test = create_labels(df_test["hard_label"])
# Faccio plot embeddings sia con label di Odio che di Disagreement
gc.collect()
print("Plotting..............")
if plot_graph:
hid, hid_test = plot_tsne_pca("Total", word_embeddings_train, word_embeddings_test, col_train, col_test)
col_train = create_labels(df_tot_train["disagreement"])
col_test = create_labels(df_test["disagreement"])
if plot_graph:
_, _ = plot_tsne_pca("Total", word_embeddings_train, word_embeddings_test, col_train, col_test)
emb_dataframe = cal_distances(df_test, word_embeddings_train, word_embeddings_test)
if save:
emb_dataframe.to_csv(file_name, sep=',', index=False, encoding='utf-8')
return emb_dataframe
# Carico Modello
model = primary_encoder_v2_no_pooler_for_con(768, 2)
model.load_state_dict(
torch.load('./models/infoNCEBest.pth'), strict=False)
model.eval()
gc.collect()
# Carico e pulisco il dataset
base_dir = './Disagreement/Dataset'
print("Load Dati")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
model.eval()
base_dir = './Disagreement/Dataset'
tokenizer = BertTokenizer.from_pretrained(base_dir + '/TokenzierBert')
# Read csv test
warnings.filterwarnings("ignore")
df_md_test = pd.read_json(base_dir + '/MD-Agreement_test.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_md_test = get_dataset_labels(df_md_test)
df_brexit_test = pd.read_json(base_dir + '/HS-Brexit_test.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_brexit_test = get_dataset_labels(df_brexit_test)
df_armis_test = pd.read_json(base_dir + '/ArMIS_test.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_armis_test = get_dataset_labels(df_armis_test)
df_conv_test = pd.read_json(base_dir + '/ConvAbuse_test.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_conv_test = get_dataset_labels(df_conv_test)
# Unisco in un unico dataset
frames = [df_md_test, df_armis_test, df_conv_test, df_brexit_test]
df_test = pd.concat(frames)
df_test = df_test.reset_index()
df_test = df_test.drop(['soft_label_0', 'soft_label_1'], axis=1)
original_text_test = df_test["text"]
df_brexit_test = clean_tweet(df_brexit_test)
df_md_test = clean_tweet(df_md_test)
df_conv_test = clean_tweet(df_conv_test)
frames = [df_md_test, df_armis_test, df_conv_test, df_brexit_test]
df_test = pd.concat(frames)
df_test = df_test.reset_index()
df_test = df_test.drop(['soft_label_0', 'soft_label_1'], axis=1)
# read csv train
df_md_train = pd.read_json(base_dir + '/MD-Agreement_train.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_md_val = pd.read_json(base_dir + '/MD-Agreement_dev.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_md_train = get_dataset_labels(df_md_train)
df_md_val = get_dataset_labels(df_md_val)
df_brexit_train = pd.read_json(base_dir + '/HS-Brexit_train.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_brexit_val = pd.read_json(base_dir + '/HS-Brexit_dev.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_brexit_train = get_dataset_labels(df_brexit_train)
df_brexit_val = get_dataset_labels(df_brexit_val)
df_armis_train = pd.read_json(base_dir + '/ArMIS_train.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_armis_val = pd.read_json(base_dir + '/ArMIS_dev.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_armis_train = get_dataset_labels(df_armis_train)
df_armis_val = get_dataset_labels(df_armis_val)
df_conv_train = pd.read_json(base_dir + '/ConvAbuse_train.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_conv_val = pd.read_json(base_dir + '/ConvAbuse_dev.json', orient='index')[['text', 'hard_label', 'soft_label']]
df_conv_train = get_dataset_labels(df_conv_train)
df_conv_val = get_dataset_labels(df_conv_val)
df_brexit_train = clean_tweet(df_brexit_train)
df_brexit_val = clean_tweet(df_brexit_val)
df_md_train = clean_tweet(df_md_train)
df_md_val = clean_tweet(df_md_val)
df_conv_train = clean_tweet(df_conv_train)
df_conv_val = clean_tweet(df_conv_val)
frames = [df_md_train, df_md_val, df_armis_train, df_armis_val, df_conv_train, df_conv_val, df_brexit_train,
df_brexit_val]
""""frames = [df_md_train, df_armis_train, df_conv_train, df_brexit_train]"""
df_tot_train = pd.concat(frames)
df_tot_train = df_tot_train.reset_index()
df_tot_train = df_tot_train.drop(['soft_label_0', 'index', 'soft_label_1'], axis=1)
labels_train = df_brexit_train["disagreement"]
labels_dev = df_brexit_val["disagreement"]
frames = [labels_train, labels_dev]
df_lab = pd.concat(frames)
df_lab = df_lab.reset_index()
df_lab = df_lab.drop(['index'], axis=1)
labels_test = df_brexit_test["disagreement"]
gc.collect()
# Aggiusto indexes
df_tot_train.index = np.arange(1, len(df_tot_train) + 1)
df_test.index = np.arange(1, len(df_test) + 1)
lab_train = df_tot_train["disagreement"]
lab_test = df_test["disagreement"]
# Fase di predizione
# se ho già creato il dataframe con le distanze lo carico e basta altrimenti lo calcolo
print("INFONCE_Total Predictions.........................")
print()
print()
file_name = "contr_final.csv"
#emb_dataframe = pd.read_csv(file_name,sep=',')
emb_dataframe = create_dataframe_distances(df_tot_train,df_test,file_name,False,True,7)
# Predictions
labels_pred = predictions(df_test, emb_dataframe, lab_train, 7, 59)
prediction_metrics(lab_test,labels_pred)
# Mostro prediction per dataset
lunghezza1 = 3057
lunghezza2 = 145
lunghezza3 = 840
lunghezza4 = 168
pred_md = labels_pred[:lunghezza1]
pred_arm = labels_pred[lunghezza1:lunghezza1 + lunghezza2]
pred_conv = labels_pred[lunghezza1 + lunghezza2:lunghezza1 + lunghezza2 + lunghezza3]
pred_brexit = labels_pred[lunghezza1 + lunghezza2 + lunghezza3:]
prediction_metrics(df_brexit_test["disagreement"],pred_brexit)
prediction_metrics(df_armis_test["disagreement"],pred_arm)
prediction_metrics(df_conv_test["disagreement"],pred_conv)
prediction_metrics(df_md_test["disagreement"],pred_md)
print("INFONCE_Split Predictions.........................")
print()
print()
neibArmis = 22
neibMD = 105
neibConv = 19
neibBrexit = 50
preds = []
# se ho già creato il dataframe con le distanze lo carico e basta alttrimenti lo calcolo per ogni dataset
emb_dataframe = create_dataframe_distances(df_md_train,df_md_test,"contr_md.csv",False,True,7)
#emb_dataframe = pd.read_csv("contr_md.csv", sep=',')
# Predictions
labels_pred = predictions(df_md_test, emb_dataframe, df_md_train["disagreement"], 2, neibMD)
preds = preds + labels_pred
prediction_metrics(df_md_test["disagreement"],labels_pred)
emb_dataframe = create_dataframe_distances(df_armis_train,df_armis_test,"contr_armis.csv",False,True,7)
#emb_dataframe = pd.read_csv("contr_armis.csv", sep=',')
# Predictions
labels_pred = predictions(df_armis_test, emb_dataframe, df_armis_train["disagreement"], 2, neibArmis)
preds = preds + labels_pred
prediction_metrics(df_armis_test["disagreement"],labels_pred)
emb_dataframe = create_dataframe_distances(df_conv_train,df_conv_test,"contr_conv.csv",False,True,7)
#emb_dataframe = pd.read_csv("contr_conv.csv", sep=',')
# Predictions
labels_pred = predictions(df_conv_test, emb_dataframe, df_conv_train["disagreement"], 2, neibConv)
preds = preds + labels_pred
prediction_metrics(df_conv_test["disagreement"],labels_pred)
emb_dataframe = create_dataframe_distances(df_brexit_train,df_brexit_test,"contr_brexit.csv",False,True,7)
#emb_dataframe = pd.read_csv("contr_brexit.csv", sep=',')
# Predictions
labels_pred = predictions(df_brexit_test, emb_dataframe, df_brexit_train["disagreement"], 2, neibBrexit)
preds = preds + labels_pred
prediction_metrics(df_brexit_test["disagreement"],labels_pred)
print("prediction totale ")
acc_new = prediction_metrics(lab_test, preds)