is-online-sexism/transformer.py at main · rudygarrido/is-online-sexism · GitHub

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
import torch
from transformers import AutoTokenizer, DataCollatorWithPadding, AutoModelForSequenceClassification, \
    TrainingArguments, Trainer, AutoConfig
import pandas as pd
import evaluate
import numpy as np
from sklearn.model_selection import train_test_split

id2label = {0: "none", 1: "1.1 threats of harm", 2: "1.2 incitement and encouragement of harm", 3: "2.1 descriptive "
                                                                                                   "attacks",
            4: "2.2 aggressive and emotive attacks", 5: "2.3 dehumanising attacks & overt sexual objectification",
            6: "3.1 casual use of gendered slurs, profanities, and insults", 7: "3.2 immutable gender differences and "
                                                                                "gender stereotypes",
            8: "3.3 backhanded gendered compliments", 9: "3.4 condescending explanations or unwelcome advice",
            10: "4.1 supporting mistreatment of individual women", 11: "4.2 supporting systemic discrimination "
                                                                       "against women as a group"}
label2id = {"none": 0, "1.1 threats of harm": 1, "1.2 incitement and encouragement of harm": 2, "2.1 descriptive "
                                                                                                "attacks": 3,
            "2.2 aggressive and emotive attacks": 4, "2.3 dehumanising attacks & overt sexual objectification": 5,
            "3.1 casual use of gendered slurs, profanities, and insults": 6, "3.2 immutable gender differences and "
                                                                             "gender stereotypes": 7, "3.3 backhanded "
                                                                                                      "gendered "
                                                                                                      "compliments":
                8, "3.4 condescending explanations or unwelcome advice": 9, "4.1 supporting mistreatment of "
                                                                            "individual women": 10, "4.2 supporting "
                                                                                                    "systemic "
                                                                                                    "discrimination "
                                                                                                    "against women as "
                                                                                                    "a group": 11}

METRICS_AVERAGE_TYPE = "micro"


class SexismDatasset(torch.utils.data.Dataset):
    def __init__(self, encodings, labels):
        self.encodings = encodings
        self.labels = labels

    def __getitem__(self, idx):
        item = {key: torch.tensor(val[idx]) for key, val in self.encodings.items()}
        item['labels'] = torch.tensor(self.labels[idx])
        return item

    def __len__(self):
        return len(self.labels)


def compute_metrics(eval_pred):
    predictions, labels = eval_pred
    predictions = np.argmax(predictions, axis=-1)

    results = {}
    results.update(f1_metric.compute(predictions=predictions, references=labels, average=METRICS_AVERAGE_TYPE))
    results.update(recall_metric.compute(predictions=predictions, references=labels, average=METRICS_AVERAGE_TYPE))
    results.update(precision_metric.compute(predictions=predictions, references=labels, average=METRICS_AVERAGE_TYPE))
    results.update(accuracy_metric.compute(predictions=predictions, references=labels))

    return results


f1_metric = evaluate.load("f1")
recall_metric = evaluate.load("recall")
precision_metric = evaluate.load("precision")
accuracy_metric = evaluate.load("accuracy")
config = AutoConfig.from_pretrained("distilbert-base-uncased")
tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased", config=config)
dataset = pd.read_csv('train_all_tasks.csv')
dataset.drop(labels=["rewire_id", "label_category", "label_sexist"], axis=1, inplace=True)
dataset.rename(columns={"label_vector": "label"}, inplace=True)
dataset['label'] = dataset['label'].map(
    {"none": 0, "1.1 threats of harm": 1, "1.2 incitement and encouragement of harm": 2, "2.1 descriptive "
                                                                                         "attacks": 3,
     "2.2 aggressive and emotive attacks": 4, "2.3 dehumanising attacks & overt sexual objectification": 5,
     "3.1 casual use of gendered slurs, profanities, and insults": 6, "3.2 immutable gender differences and "
                                                                      "gender stereotypes": 7, "3.3 backhanded "
                                                                                               "gendered "
                                                                                               "compliments":
         8, "3.4 condescending explanations or unwelcome advice": 9, "4.1 supporting mistreatment of "
                                                                     "individual women": 10, "4.2 supporting "
                                                                                             "systemic "
                                                                                             "discrimination "
                                                                                             "against women as "
                                                                                             "a group": 11})
train, test = train_test_split(dataset, test_size=0.2, shuffle=True, stratify=dataset['label'])
train_texts, train_labels = train["text"].tolist(), train["label"].tolist()
test_texts, test_labels = test["text"].tolist(), test["label"].tolist()


train_encodings = tokenizer(train_texts, truncation=True, padding=True)
test_encodings = tokenizer(test_texts, truncation=True, padding=True)

train_dataset = SexismDatasset(train_encodings, train_labels)
test_dataset = SexismDatasset(test_encodings, test_labels)


data_collator = DataCollatorWithPadding(tokenizer=tokenizer)

def model_init():
    return AutoModelForSequenceClassification.from_pretrained(
    "distilbert-base-uncased", num_labels=12, id2label=id2label, label2id=label2id
)

training_args = TrainingArguments(
    output_dir="is_online_sexism_model",
    learning_rate=2e-5,
    per_device_train_batch_size=16,
    per_device_eval_batch_size=16,
    num_train_epochs=2,
    weight_decay=0.01,
    evaluation_strategy="epoch",
    save_strategy="epoch",
    load_best_model_at_end=True,
    push_to_hub=False,
)

trainer = Trainer(
    model_init=model_init,
    args=training_args,
    train_dataset=train_dataset,
    eval_dataset=test_dataset,
    tokenizer=tokenizer,
    data_collator=data_collator,
    compute_metrics=compute_metrics,
)

best_trial = trainer.hyperparameter_search(
    direction="maximize",
    backend="ray",
    n_trials=2
)

print(best_trial)