train.py

from model import get_model
import torch
from transformers import BertTokenizer, Trainer, TrainingArguments
from datasets import load_dataset
import numpy as np
from sklearn.metrics import accuracy_score, precision_recall_fscore_support
from torch.utils.data import DataLoader
from sklearn.utils.class_weight import compute_class_weight

# Other imports and code remain the same...

# Compute class weights
class_weights = compute_class_weight(
    'balanced', classes=np.unique(train_dataset['labels']), y=train_dataset['labels'])
class_weights = torch.tensor(class_weights, dtype=torch.float)

# Update the model's classifier with class weights
model.classifier.weight.data = class_weights
# Load dataset dynamically or from a config
dataset_name = "NicolaiSivesind/human-vs-machine"
dataset = load_dataset(dataset_name)

tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')

def compute_metrics(pred):
    labels = pred.label_ids
    preds = np.argmax(pred.predictions, axis=1)
    precision, recall, f1, _ = precision_recall_fscore_support(labels, preds, average='binary')
    acc = accuracy_score(labels, preds)
    return {
        'accuracy': acc,
        'f1': f1,
        'precision': precision,
        'recall': recall
    }

def tokenize_function(examples):
    # Add any specific preprocessing steps if necessary
    return tokenizer(examples["text"], padding="max_length", truncation=True, max_length=512)

def get_tokenizer():
    try:
        return BertTokenizer.from_pretrained('./trained_model')
    except Exception:
        return BertTokenizer.from_pretrained('bert-base-uncased')

tokenized_dataset = dataset.map(tokenize_function, batched=True)
tokenized_dataset = tokenized_dataset.rename_column("original_label_name", "labels")
tokenized_dataset.set_format(type='torch', columns=['input_ids', 'attention_mask', 'labels'])

train_dataset = tokenized_dataset["train"]
eval_dataset = tokenized_dataset["validation"]
model = get_model()

# Make training arguments configurable
training_args = TrainingArguments(
    output_dir="./results",
    num_train_epochs=3,
    per_device_train_batch_size=8,
    per_device_eval_batch_size=8,
    warmup_steps=500,
    weight_decay=0.01,
    logging_dir='./logs',
    evaluation_strategy="steps",
    save_steps=500, # Save model every 500 steps
    logging_steps=100,
)

trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    compute_metrics=compute_metrics # Define this function to compute additional metrics
)

trainer.train()
model.save_pretrained("./trained_model")
tokenizer.save_pretrained("./trained_model")