experiments/model_finetuning_hate.py

"""Experiment.

```
python model_finetuning_hate.py -m "roberta-base" -d "hate_temporal"
```
"""
import argparse
import json
import logging
import math
import os
import re
from os.path import join as pj
from shutil import copyfile, rmtree
from glob import glob

import numpy as np
import evaluate
from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer
from huggingface_hub import Repository

logging.basicConfig(format="%(asctime)s %(levelname)-8s %(message)s", level=logging.INFO, datefmt="%Y-%m-%d %H:%M:%S")
os.environ["WANDB_DISABLED"] = "true"

EVAL_STEP = 500
RANDOM_SEED = 42
N_TRIALS = 10
URL_RE = re.compile(r"https?:\/\/[\w\.\/\?\=\d&#%_:/-]+")
HANDLE_RE = re.compile(r"@\w+")


def preprocess_bernice(text):
    text = HANDLE_RE.sub("@USER", text)
    text = URL_RE.sub("HTTPURL", text)
    return text


def preprocess_timelm(text):
    text = HANDLE_RE.sub("@user", text)
    text = URL_RE.sub("http", text)
    return text


def preprocess(model_name, text):
    if model_name == "jhu-clsp/bernice":
        return preprocess_bernice(text)
    if "twitter-roberta-base" in model_name:
        return preprocess_timelm(text)
    return text


def sigmoid(x):
    return 1 / (1 + math.exp(-x))


def main(
        dataset: str = "tweettemposhift/tweet_temporal_shift",
        dataset_type: str = "hate_temporal",
        model: str = "roberta-base",
        skip_train: bool = False,
        skip_test: bool = False,
        skip_upload: bool = False):

    model_alias = f"hate-{dataset_type}-{os.path.basename(model)}"
    output_dir = f"ckpt/{model_alias}"
    best_model_path = pj(output_dir, "best_model")

    tokenizer = AutoTokenizer.from_pretrained(model)
    dataset = load_dataset(dataset, dataset_type)
    tokenized_datasets = dataset.map(
        lambda x: tokenizer(
            [preprocess(model, t) for t in x["text"]],
            padding="max_length",
            truncation=True,
            max_length=64),
        batched=True
    )
    tokenized_datasets = tokenized_datasets.rename_column("gold_label_binary", "label")
    metric_accuracy = evaluate.load("accuracy")
    metric_f1 = evaluate.load("f1")

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

    def compute_metric_all(eval_pred):
        logits, labels = eval_pred
        predictions = np.argmax(logits, axis=-1)
        return {
            "f1": metric_f1.compute(predictions=predictions, references=labels)["f1"],
            "accuracy": metric_accuracy.compute(predictions=predictions, references=labels)["accuracy"]
        }

    if not skip_train:
        logging.info("training model")
        trainer = Trainer(
            model=AutoModelForSequenceClassification.from_pretrained(model, num_labels=2),
            args=TrainingArguments(
                output_dir=output_dir,
                evaluation_strategy="steps",
                eval_steps=EVAL_STEP,
                seed=RANDOM_SEED
            ),
            train_dataset=tokenized_datasets["train"],
            eval_dataset=tokenized_datasets["validation"],
            compute_metrics=compute_metric_search,
            model_init=lambda x: AutoModelForSequenceClassification.from_pretrained(
                model, return_dict=True, num_labels=2,
            )
        )

        best_run = trainer.hyperparameter_search(
            hp_space=lambda trial: {
                "learning_rate": trial.suggest_float("learning_rate", 1e-6, 1e-4, log=True),
                "per_device_train_batch_size": trial.suggest_categorical(
                    "per_device_train_batch_size", [8, 16, 32]
                ),
            },
            direction="maximize",
            backend="optuna",
            n_trials=N_TRIALS
        )
        for n, v in best_run.hyperparameters.items():
            setattr(trainer.args, n, v)
        trainer.train()
        trainer.save_model(best_model_path)

    if not skip_test:
        logging.info("testing model")
        test_split = ["test"]
        if dataset_type.endswith("temporal"):
            test_split += ["test_1", "test_2", "test_3", "test_4"]
        summary_file = pj(best_model_path, "summary.json")
        if os.path.exists(summary_file):
            with open(summary_file) as f:
                metric = json.load(f)
        else:
            metric = {}
        for single_test in test_split:
            trainer = Trainer(
                model=AutoModelForSequenceClassification.from_pretrained(best_model_path, num_labels=2),
                args=TrainingArguments(
                    output_dir=output_dir,
                    evaluation_strategy="no",
                    seed=RANDOM_SEED
                ),
                train_dataset=tokenized_datasets["train"],
                eval_dataset=tokenized_datasets[single_test],
                compute_metrics=compute_metric_all
            )
            metric.update({f"{single_test}/{k}": v for k, v in trainer.evaluate().items()})
        logging.info(json.dumps(metric, indent=4))
        with open(summary_file, "w") as f:
            json.dump(metric, f)

    if not skip_upload:
        logging.info("uploading to huggingface")
        model_organization = "tweettemposhift"
        model_instance = AutoModelForSequenceClassification.from_pretrained(best_model_path, num_labels=2)
        model_instance.push_to_hub(f"{model_organization}/{model_alias}", use_auth_token=True)
        tokenizer.push_to_hub(f"{model_organization}/{model_alias}", use_auth_token=True)
        repo = Repository(model_alias, f"{model_organization}/{model_alias}")
        for i in glob(f"{best_model_path}/*"):
            if not os.path.exists(f"{model_alias}/{os.path.basename(i)}"):
                copyfile(i, f"{model_alias}/{os.path.basename(i)}")
        repo.push_to_hub()
        rmtree(model_alias)


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Fine-tuning language model.")
    parser.add_argument("-m", "--model", help="transformer LM", default="roberta-base", type=str)
    parser.add_argument("-d", "--dataset-type", help='dataset type', default="hate_temporal", type=str)
    parser.add_argument("--skip-train", action="store_true")
    parser.add_argument("--skip-test", action="store_true")
    parser.add_argument("--skip-upload", action="store_true")
    opt = parser.parse_args()
    main(
        dataset_type=opt.dataset_type,
        model=opt.model,
        skip_train=opt.skip_train,
        skip_test=opt.skip_test,
        skip_upload=opt.skip_upload,
    )