BERT_train.py

# pip install transformers
# pip install datasets

from datasets import Dataset
from datasets import load_metric
from xml.etree.ElementTree import ElementTree
import pandas as pd
import numpy as np
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer

# Initialization
tokenizer = AutoTokenizer.from_pretrained("bert-base-cased")
model = AutoModelForSequenceClassification.from_pretrained("bert-base-cased", num_labels=3)
metric = load_metric("accuracy")
training_args = TrainingArguments(output_dir="../test_trainer", evaluation_strategy="epoch")


def extract_data(file):
    fileTree = ElementTree()
    fileTree.parse(file)
    root = fileTree.getroot()

    questions = []

    for thread in root:
        for child in thread:
            subject = child.find("RelQSubject").text
            body = child.find("RelQBody").text
            content = "{} {}".format(subject, body)
            tag = child.attrib["RELQ_FACT_LABEL"]

            questions.append((tag, content))

    questions_dataframe = pd.DataFrame(questions)
    questions_dataframe.columns = ["label", "text"]

    questions_dataframe["label"] = questions_dataframe["label"].replace({'Factual': 0, 'Opinion': 1, 'Socializing': 2})

    return questions_dataframe


def tokenize_function(examples):
    return tokenizer(examples["text"], padding="max_length", truncation=True)


def compute_metrics(eval_pred):
    logits, labels = eval_pred
    predictions = np.argmax(logits, axis=-1)
    return metric.compute(predictions=predictions, references=labels)


train_df = extract_data("questions_train.xml")
dev_df = extract_data("questions_dev.xml")

train_dataset = Dataset.from_pandas(train_df)
dev_dataset = Dataset.from_pandas(dev_df)

tokenized_train_dataset = train_dataset.map(tokenize_function, batched=True)
tokenized_dev_dataset = dev_dataset.map(tokenize_function, batched=True)

small_train_dataset = tokenized_train_dataset.shuffle(seed=42).select(range(5))
small_dev_dataset = tokenized_dev_dataset.shuffle(seed=42).select(range(1))

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=small_train_dataset,
    eval_dataset=small_dev_dataset,
    compute_metrics=compute_metrics,
)

trainer.train()
trainer.save_model("saved_model")