prueba.py

# 2️⃣ Importar librerías
from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer
import evaluate
import numpy as np


# 3️⃣ Cargar dataset educativo
dataset = load_dataset("haider0941/Educational_Noneducational_Dataset")
print(dataset['train'][0])
print("----------3----------")

# Arreglo del dataset

def fix_dataset(example):
    # Extrae la primera clave (el texto)
    text = list(example.keys())[0]  # la clave que no sea 'educational'
    label = example['educational']  # etiqueta
    return {"text": text, "label": 0 if label == "educational" else 1}

dataset = dataset.map(fix_dataset)
print(dataset["train"][0])
# {'text': 'What is photosynthesis?', 'label': 0}

print("--------Dataset fix--------------")

# 4️⃣ Definir modelo y tokenizer
model_name = "distilbert-base-uncased"  # modelo ligero
tokenizer = AutoTokenizer.from_pretrained(model_name)
print("----------4----------")
# 5️⃣ Tokenizar dataset
def tokenize_function(example):
    return tokenizer(example["text"], padding="max_length", truncation=True, max_length=128)

tokenized_datasets = dataset.map(tokenize_function, batched=True)
print("----------5----------")

# 6️⃣ Convertir etiquetas a enteros → ELIMINADO, ya existe columna 'label'
# label2id = {"educational":0, "noneducational":1}
# tokenized_datasets = tokenized_datasets.map(lambda x: {"label": label2id[x["response_name"]]}, batched=False)
print("----------6----------")

# 7️⃣ Preparar modelo para clasificación
model = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=2)
print("----------7----------")

# 8️⃣ Definir métricas
accuracy = evaluate.load("accuracy")
f1 = evaluate.load("f1")

def compute_metrics(eval_pred):
    logits, labels = eval_pred
    predictions = logits.argmax(axis=-1)
    return {"accuracy": accuracy.compute(predictions=predictions, references=labels)["accuracy"],
            "f1": f1.compute(predictions=predictions, references=labels)["f1"]}
print("----------8----------")
# 9️⃣ Definir argumentos de entrenamiento
training_args = TrainingArguments(
    output_dir="./results",
    evaluation_strategy="epoch",
    learning_rate=5e-5,
    per_device_train_batch_size=16,
    per_device_eval_batch_size=16,
    num_train_epochs=3,
    weight_decay=0.01,
)
print("----------9----------")
#  🔟 Inicializar Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets["train"],
    eval_dataset=tokenized_datasets["test"],
    compute_metrics=compute_metrics,
)
print("----------10----------")
# 1️⃣1️⃣ Entrenar modelo
trainer.train()
print("----------11----------")
# 1️⃣2️⃣ Evaluar modelo
trainer.evaluate()
print("----------12----------")