ActiveLLM/setfit_modelclass.py at main · PEASEC/ActiveLLM · 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
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import transformers
import torch
from sklearn.metrics import accuracy_score, recall_score, precision_score, f1_score
from transformers import TrainingArguments, Trainer
from scipy.stats import spearmanr
from sklearn.metrics import matthews_corrcoef
import numpy as np
import logging
from datasets import Dataset
from datasets import load_dataset
import os
dir = os.path.dirname(os.path.abspath(__file__))
from transformers import set_seed
from sentence_transformers.losses import CosineSimilarityLoss
from setfit import SetFitModel, SetFitTrainer


def compute_metrics(y_pred, y_test):
    # make differentiation between binary and multi-class classification
    predictions, labels = y_pred, y_test
    if len(np.unique(labels)) == 2:
        return {
            "eval_Accuracy: " : accuracy_score(labels, predictions),
            "eval_F1: " : f1_score(labels, predictions, pos_label=1),
            "Precision_1: " : recall_score(labels, predictions, pos_label=1),
            "Recall_1: " : precision_score(labels, predictions, pos_label=1),
            "Precision_0: " : recall_score(labels, predictions, pos_label=0),
            "Recall_0: " : precision_score(labels, predictions, pos_label=0),
            # methew's correlation coefficient
            "eval_Corr: ": matthews_corrcoef(labels, predictions),
        }
    elif len(np.unique(labels)) >= 5:
        predictions = np.squeeze(predictions)
        return { #spearman correlation
            "eval_Accuracy: " : accuracy_score(labels, predictions),
            "eval_Corr: ": spearmanr(labels, predictions)[0],
        }
    else:
        return {
            "eval_Accuracy: " : accuracy_score(labels, predictions),
            "eval_F1: " : f1_score(labels, predictions, average="macro"),
            "Precision: " : recall_score(labels, predictions, average="macro"),
            "Recall: " : precision_score(labels, predictions, average="macro"),
        }

def set_seed(seed=109):
    # set seed
    torch.manual_seed(seed)
    torch.cuda.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)
    np.random.seed(seed)
    transformers.set_seed(seed)

# Class representing a model
class SetFitModelClass:
    def __init__(self, seed=109):
        # init a huggingface auto model based on the model name
        set_seed(seed)
        self.model = SetFitModel.from_pretrained("sentence-transformers/paraphrase-mpnet-base-v2").to("cuda")
        #self.tokenizer = AutoTokenizer.from_pretrained("sentence-transformers/paraphrase-mpnet-base-v2")

    def predict(self, texts, bs = 128):
        predictions = self.predictor(texts, bs).tolist()
        return np.argmax(predictions, axis=1)

    def predictor(self, texts, bs = 128):
        if isinstance(texts, str):
            texts = [texts]

        predictions = []
        with torch.no_grad():
            self.model.eval()
            if type(texts) is not tuple:
                break_len = len(texts)
            else:
                break_len = len(texts[0])

            for i in range(0, break_len, bs):
                if type(texts) is not tuple:
                    batch_x = self.tokenizer(texts[i:i+bs], padding=True, truncation=True, return_tensors="pt").to("cuda")
                else:
                    batch_x = self.tokenizer(texts[0][i:i+bs], texts[1][i:i+bs], padding=True, truncation=True, return_tensors="pt").to("cuda")
                predictions.extend(torch.nn.functional.softmax(self.model(**batch_x)[0], dim=1).cpu().detach().tolist())
        return np.array(predictions)

    def train(self, x_train, y_train, x_dev, y_dev, train_bs = 16, test_bs = 32):
        # transform x_train, y_train, x_dev, y_dev to lists of dictionaries where each dictionary represents a data instance
        # if x_train and x_dev are tuples, then the dictionaries have two keys: "text1" and "text2"
        # if x_train and x_dev are not tuples, then the dictionaries have one key: "text"
        if type(x_train) is not tuple:
            train_data = {"text": x_train, "label": y_train}
            test_data = {"text": x_dev, "label": y_dev}
        else:
            train_data = {"text1": x_train[0], "text2": x_train[1], "label": y_train}
            test_data = {"text1": x_dev[0], "text2": x_dev[1], "label": y_dev}

        # convert to datasets
        train_ds = Dataset.from_dict(train_data)
        test_ds = Dataset.from_dict(test_data)


        trainer = SetFitTrainer(
            model=self.model,
            train_dataset=train_ds,
            eval_dataset=test_ds,
            #loss_class=CosineSimilarityLoss,
            metric=compute_metrics,
            batch_size=train_bs,
            num_epochs=4,
            #num_iterations=20, # Number of text pairs to generate for contrastive learning
            #num_epochs=1 # Number of epochs to use for contrastive learning
            #column_mapping={"sentence": "text", "label": "label"}
        )

        trainer.train()
        # logging the results
        logging.info("Training results:")
        evaluations = trainer.evaluate()

        logging.info(evaluations)

        return evaluations

    # def evaluate(self, x_test, y_test):
    #     if type(x_test) is not tuple:
    #         test_encodings = self.tokenizer(x_test, padding=True, truncation=True, return_tensors="pt")
    #     else:
    #         test_encodings = self.tokenizer(*x_test, padding=True, truncation=True, return_tensors="pt")

    #     test_dataset = Dataset(test_encodings, y_test)
    #     trainer = Trainer(
    #         model=self.model,
    #         compute_metrics=compute_metrics,
    #     )
    #     logging.info("Test results:")
    #     logging.info(trainer.evaluate(test_dataset))


    def save(self, path):
        self.model.save_pretrained(path)

    def load(self, path, num_labels):
        self.model = AutoModelForSequenceClassification.from_pretrained(path, local_files_only=True, num_labels=num_labels)
        self.tokenizer = AutoTokenizer.from_pretrained(path, local_files_only=True)