test_text_classifiers.py

import random

import numpy as np
import pytest

from main import load_config
from tasks.data.data_loaders import TextDataLoader
from tasks.models.text_classifiers import DistilBERTModel, ModelFactory, TextEmbedder, MLModel, EmbeddingMLModel, \
    TfIdfEmbedder
from tasks.utils.evaluation import TextEvaluationRequest

@pytest.fixture()
def data_loader():
    # define text request
    text_request = TextEvaluationRequest()

    return TextDataLoader(text_request, light=True)

@pytest.fixture()
def train_dataset(data_loader):
    return data_loader.get_train_dataset()

@pytest.fixture()
def test_dataset(data_loader):
    return data_loader.get_test_dataset()


class TestDistilBERTModel:
    @pytest.fixture()
    def distilBERT_model(self):
        config = load_config("config_training_test.json")
        return ModelFactory.create_model(config)

    def test_trained_distilBERT(self, train_dataset, distilBERT_model, test_dataset):
        assert "DistilBERT" in distilBERT_model.description

        # train model
        distilBERT_model.train(train_dataset)

        # inference
        predictions = [distilBERT_model.predict(quote) for quote in test_dataset["quote"]]
        for prediction in predictions:
            assert prediction in range(8)

    def test_data_preprocessing(self, train_dataset, distilBERT_model):
        pre_processed_data = distilBERT_model.pre_process_data(train_dataset)
        assert pre_processed_data is not None
        assert pre_processed_data["train"].num_rows == 8
        assert pre_processed_data["test"].num_rows == 2

        for subset in ["train", "test"]:
            for feature_name in ['quote', 'label', 'input_ids', 'attention_mask']:
                assert feature_name in pre_processed_data[subset].features.keys()


class DummyEmbedder(TextEmbedder):
    def encode(self, text: str) -> np.ndarray:
        return np.random.rand(42)


class DummyMLModel(MLModel):
    def fit(self, X, y):
        pass

    def predict(self, X):
        return random.choice(range(8))


class TestEmbeddingMLModel:
    @pytest.fixture()
    def embeddingML(self):
        config = load_config("config_training_embedding_test.json")
        config["model"] = "EmbeddingMLModel"
        return ModelFactory.create_model(config)

    def test_EmbeddingML(self, train_dataset, embeddingML):
        assert "EmbeddingMLModel" in embeddingML.description

        # train model
        embeddingML.train(train_dataset)

        # inference
        assert embeddingML.predict("a quote") in range(8)

    def test_dummy_train_EmbeddingML(self, train_dataset):
        dummy_model = EmbeddingMLModel(embedder=DummyEmbedder(),
                                       ml_model=DummyMLModel())

        dummy_model.train(train_dataset)
        assert dummy_model.predict("dummy") in range(8)

class TestEmbedders:
    def test_tf_idf(self):
        embedder = TfIdfEmbedder()

        texts = [
            "hello world",
            "world hello",
            "yet another text",
            "this is a test",
            "this one as well"
        ]
        encoded_texts = embedder.encode(texts)
        assert encoded_texts.shape == (5, 11)