preprocessing_pipeline.py

from datasets import load_dataset, DatasetDict, load_from_disk
from transformers import AutoTokenizer

raw_dataset = load_dataset("imdb")

train_valid_split = raw_dataset["train"].train_test_split(
    test_size=0.1,
    seed=42
)

dataset = DatasetDict({
    "train": train_valid_split["train"],
    "validation": train_valid_split["test"],
    "test": raw_dataset["test"]
})

def filter_empty_or_short(example):
    text = example["text"]

    if text is None:
        return False

    text = text.strip()

    if text == "":
        return False

    if len(text) <= 10:
        return False

    return True

filtered_dataset = dataset.filter(filter_empty_or_short)

tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")

def tokenize_function(batch):
    return tokenizer(
        batch["text"],
        padding="max_length",
        truncation=True,
        max_length=128
    )

tokenized_dataset = filtered_dataset.map(
    tokenize_function,
    batched=True
)

tokenized_dataset.save_to_disk("./imdb_preprocessed_dataset")
tokenizer.save_pretrained("./bert_tokenizer")

loaded_dataset = load_from_disk("./imdb_preprocessed_dataset")

print(loaded_dataset)
print("train:", "train" in loaded_dataset)
print("validation:", "validation" in loaded_dataset)
print("test:", "test" in loaded_dataset)