Simple Python example

Just like the quickstart example foy .yaml, in this simple experiment we will:

1. Use the toys Amazon Dataset and add 'item_' and 'user_' prefixes to each item and user ids
2. Train the GPT2Rec using the distilgpt2 checkpoint on the SequentialSideInfoTask
3. Evaluate results using hit@10 and hit@5 metrics

The model trained will be saved into models/simple_experiment path and the metrics results into reports/metrics/simple_experiment path

Run the experiment

from src.data.datasets.amazon_dataset import AmazonDataset
from src.data.tasks.tasks import SequentialSideInfoTask
from src.evaluate.evaluator import RecEvaluator
from src.evaluate.metrics.ranking_metrics import Hit
from src.model.models.gpt import GPT2Rec
from src.model.trainer import RecTrainer

if __name__ == "__main__":

    # data phase
    ds = AmazonDataset("toys", add_prefix_items_users=True)

    ds_splits = ds.get_hf_datasets()  # this returns a dict of hf datasets

    train_split = ds_splits["train"]
    val_split = ds_splits["validation"]
    test_split = ds_splits["test"]

    # model phase
    model = GPT2Rec("distilgpt2",
                    training_tasks_str=["SequentialSideInfoTask"],
                    all_unique_labels=list(ds.all_items))

    trainer = RecTrainer(model,
                         n_epochs=10,
                         batch_size=8,
                         train_sampling_fn=ds.sample_train_sequence,
                         output_dir="models/simple_experiment")

    trainer.train(train_split)

    # eval phase
    evaluator = RecEvaluator(model, eval_batch_size=4)

    evaluator.evaluate_suite(test_split,
                             tasks_to_evaluate={SequentialSideInfoTask(): [Hit(k=10), Hit(k=5)]},
                             output_dir="reports/metrics/simple_experiment")

As you can see, it's very easy to perform a complete experiment also via the Python API!

Reducing dataset size for testing purposes

If you really want to get a glimpse of the final results of the process without having to wait a lot, since AmazonDataset is quite big, you could cut the datasets size of each split for testing purposes!

This is the new data phase in which each dataset split size has been reduced

Data phase with reduced dataset size for testing

from datasets import Dataset
from src.data.datasets.amazon_dataset import AmazonDataset

# data phase
ds = AmazonDataset("toys", add_prefix_items_users=True)

ds_splits = ds.get_hf_datasets()

train_split = Dataset.from_dict(ds_splits["train"][:100])
val_split = Dataset.from_dict(ds_splits["validation"][:100])
test_split = Dataset.from_dict(ds_splits["test"][:100])

As you can see, the full integration with state-of-the-art libraries makes LaikaLLM feel like home!