SEPC/task.py at main · SELGroup/SEPC · 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
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
import os
from enum import Enum, unique

import numpy as np
import torch
from datasets import load_dataset
from sklearn.metrics import accuracy_score, f1_score
from torch.utils.data import TensorDataset
from scipy.stats import pearsonr, spearmanr


@unique
class Task(Enum):
    # classification
    tweeteval_emoji = "tweeteval_emoji"
    tweeteval_emotion = "tweeteval_emotion"
    tweeteval_hate = "tweeteval_hate"
    tweeteval_irony = "tweeteval_irony"
    tweeteval_offensive = "tweeteval_offensive"
    tweeteval_sentiment = "tweeteval_sentiment"
    tweeteval_stance = "tweeteval_stance"
    isear_v3 = "isear_v3"
    meld = "meld"
    goemotions = "goemotions"

    # regression
    glue_sts_b = "glue_sts_b"
    claire_v2 = "claire_v2"
    emobank = "emobank"

    def num_classes(self, tasks_config):
        return tasks_config[self]['num_classes']


class TaskConfig:
    def __init__(self, dataset_script, columns, batch_size, eval_batch_size, metrics, task_type="cls", bins_num=None):
        self.dataset_script = dataset_script
        self.columns = columns
        self.batch_size = batch_size
        self.eval_batch_size = eval_batch_size
        self.metrics = metrics
        self.task_type = task_type
        if task_type == 'res':
            self.bins_num = bins_num


def define_dataset_config(fine_tune_task=None, batch_size=64, pair_flag=False):
    task_columns = ["label", "sentence1", "sentence2"] if pair_flag else ["label", "sentence1"]
    datasets_config = {
        # classification
        Task.isear_v3: TaskConfig(dataset_script="./datasets/isear_v3.py", columns=task_columns,
                                  batch_size=batch_size, eval_batch_size=batch_size * 2, metrics=[accuracy_score, f1_score]),
        Task.goemotions: TaskConfig(dataset_script="./datasets/goemotions.py", columns=task_columns,
                                    batch_size=batch_size, eval_batch_size=batch_size * 2, metrics=[accuracy_score, f1_score]),
        Task.meld: TaskConfig(dataset_script="./datasets/meld.py", columns=task_columns,
                              batch_size=batch_size, eval_batch_size=batch_size * 2, metrics=[accuracy_score, f1_score]),
        Task.tweeteval_emoji: TaskConfig(dataset_script="./datasets/tweeteval_emoji.py", columns=task_columns,
                                         batch_size=batch_size, eval_batch_size=batch_size * 2, metrics=[accuracy_score, f1_score]),
        Task.tweeteval_emotion: TaskConfig(dataset_script="./datasets/tweeteval_emotion.py", columns=task_columns,
                                           batch_size=batch_size, eval_batch_size=batch_size * 2,
                                           metrics=[accuracy_score, f1_score]),
        Task.tweeteval_hate: TaskConfig(dataset_script="./datasets/tweeteval_hate.py", columns=task_columns,
                                        batch_size=batch_size, eval_batch_size=batch_size * 2,
                                        metrics=[accuracy_score, f1_score]),
        Task.tweeteval_irony: TaskConfig(dataset_script="./datasets/tweeteval_irony.py", columns=task_columns,
                                         batch_size=batch_size, eval_batch_size=batch_size * 2,
                                         metrics=[accuracy_score, f1_score]),
        Task.tweeteval_offensive: TaskConfig(dataset_script="./datasets/tweeteval_offensive.py", columns=task_columns,
                                             batch_size=batch_size, eval_batch_size=batch_size * 2,
                                             metrics=[accuracy_score, f1_score]),
        Task.tweeteval_sentiment: TaskConfig(dataset_script="./datasets/tweeteval_sentiment.py", columns=task_columns,
                                             batch_size=batch_size, eval_batch_size=batch_size * 2,
                                             metrics=[accuracy_score, f1_score]),
        Task.tweeteval_stance: TaskConfig(dataset_script="./datasets/tweeteval_stance.py", columns=task_columns,
                                          batch_size=batch_size, eval_batch_size=batch_size * 2,
                                          metrics=[accuracy_score, f1_score]),
        # regression
        Task.glue_sts_b: TaskConfig(dataset_script="./datasets/glue_sts_b.py", columns=task_columns,
                                    batch_size=batch_size, eval_batch_size=batch_size * 2,
                                    metrics=[pearsonr, spearmanr], task_type="res", bins_num = 5),
        Task.claire_v2: TaskConfig(dataset_script="./datasets/claire_v2.py", columns=task_columns,
                                   batch_size=batch_size, eval_batch_size=batch_size * 2,
                                   metrics=[pearsonr, spearmanr], task_type="res", bins_num = 4),
        Task.emobank: TaskConfig(dataset_script="./datasets/emobank.py", columns=task_columns,
                                 batch_size=batch_size, eval_batch_size=batch_size * 2,
                                 metrics=[pearsonr, spearmanr], task_type="res"),

    }
    if fine_tune_task is not None:
        if type(fine_tune_task) == Task and fine_tune_task in datasets_config.keys():
            # single task
            datasets_config = dict((k, v) for k, v in datasets_config.items() if k == fine_tune_task)
        else:
            datasets_config = dict((k, v) for k, v in datasets_config.items() if k in fine_tune_task)

    print("## datasets_config: ", datasets_config)
    print("## sentences_pair_flag: ", pair_flag)
    print("## fine_tune_task: ", fine_tune_task)
    return datasets_config


def define_tasks_config(datasets_config, dataset_percentage=1.0, cache_dir="./datasets/cache/", class_weights_flag=False, noise_rate=0.0):
    tasks_config = {}
    for id, (task, task_config) in enumerate(datasets_config.items()):
        print("task: ", task)
        print("task_config: ", task_config)

        if not os.path.isdir(cache_dir): os.makedirs(cache_dir)
        dataset_dic = load_dataset(path=task_config.dataset_script, cache_dir=cache_dir, split=None)
        train_dataset, val_dataset, test_dataset = dataset_dic["train"], dataset_dic["validation"], dataset_dic["test"]
        len_dataset = len(train_dataset)
        print("## dataset_dic.shape: ", dataset_dic.shape)
        if task_config.task_type == 'cls':
            class_names = train_dataset.features['label'].names
            num_classes = train_dataset.features['label'].num_classes

        if noise_rate > 0:
            # add noise to train_dataset
            train_label = np.array(train_dataset['label']).squeeze()
            num_noise = int(len_dataset * noise_rate)
            noise_idx = np.random.choice(np.arange(len_dataset), num_noise, False)
            noise_label = np.random.choice(np.arange(num_classes), num_noise, True)
            train_label[noise_idx] = noise_label
            train_dataset = train_dataset.remove_columns('label').add_column('label', train_label)
        # convert res label to cls label
        if task_config.task_type == 'res':
            train_label = np.array(train_dataset['label']).squeeze()
            _, bins = np.histogram(train_label, bins=task_config.bins_num)
            bins[-1] = bins[-1] + 1e-7 # 容纳右边界

        train_dataset = train_dataset.select(
            list(np.random.choice(np.arange(len_dataset), int(len_dataset * dataset_percentage if dataset_percentage <= 1 else dataset_percentage), False)))
        train_loader = torch.utils.data.DataLoader(train_dataset, num_workers=0, batch_size=task_config.batch_size, shuffle=len_dataset > 0)
        dev_loader = torch.utils.data.DataLoader(val_dataset, num_workers=0, batch_size=task_config.eval_batch_size, shuffle=False)
        test_loader = torch.utils.data.DataLoader(test_dataset, num_workers=0, batch_size=task_config.eval_batch_size, shuffle=False)

        if class_weights_flag:
            import sys, importlib
            sys.path.append("./datasets")
            dataset_module = importlib.import_module(task.name)
            formatted_name = "".join([w.capitalize() for w in task.name.split('_')]) + "_Dataset"
            dataset_class = getattr(dataset_module, formatted_name)
            class_weights = list(dataset_class.LABEL2WEIGHT.values())
        else:
            class_weights = None

        if task_config.task_type == "res":
            tasks_config[task] = dict(
                task_id=id,
                class_names=list(range(train_dataset.features['label'].length)),
                num_classes=train_dataset.features['label'].length,
                columns=task_config.columns,
                train_loader=train_loader,
                dev_loader=dev_loader,
                test_loader=test_loader,
                test_dataset=test_dataset,
                train_dataset=train_dataset,
                class_weights=class_weights,
                task_type=task_config.task_type,
                batch_size=task_config.batch_size,
                bins = bins,
            )
        else:
            tasks_config[task] = dict(
                task_id=id,
                class_names=class_names,
                num_classes=num_classes,
                columns=task_config.columns,
                train_loader=train_loader,
                dev_loader=dev_loader,
                test_loader=test_loader,
                test_dataset=test_dataset,
                train_dataset=train_dataset,
                class_weights=class_weights,
                task_type=task_config.task_type,
                batch_size=task_config.batch_size,
            )
    return tasks_config