PythonMLG/tests.py

import pytest
import torch

import config
from model import build_transformer
from dataset import TextDataset


@pytest.fixture(scope="class", autouse=False)
def get_data(request):
    dataset = TextDataset(config.DATA_PATH1, config.DATA_PATH2, config.INP_SEQ_LEN)
    dataloader = torch.utils.data.DataLoader(dataset=dataset, batch_size=1)
    model = build_transformer(dataset.rus_lang.n_words, config.INP_SEQ_LEN, dataset.rus_lang)
    model.load_checkpoint(config.CHECKPOINT_PATH)
    model = model.to(config.DEVICE)

    request.cls.model = model
    for batch in dataloader:
        request.cls.batch = batch
        break
    yield


@pytest.mark.usefixtures("get_data")
class TestBody:
    def test_encoder_seq(self):
        input_data = self.batch["encoder_input"].to(config.DEVICE)
        input_mask = self.batch["encoder_mask"].to(config.DEVICE)
        enc_out = self.model.encode(input_data, input_mask)
        assert enc_out.shape[1] == config.INP_SEQ_LEN

    def test_encoder_embed(self):
        input_data = self.batch["encoder_input"].to(config.DEVICE)
        input_mask = self.batch["encoder_mask"].to(config.DEVICE)
        enc_out = self.model.encode(input_data, input_mask)
        assert enc_out.shape[2] == 512

    def test_class_out_shape(self):
        input_data = self.batch["encoder_input"].to(config.DEVICE)
        input_mask = self.batch["encoder_mask"].to(config.DEVICE)
        enc_out = self.model.encode(input_data, input_mask)
        output = self.model.classify(enc_out)
        assert output.shape[-1] == 1

    def test_class_out_value(self):
        input_data = self.batch["encoder_input"].to(config.DEVICE)
        input_mask = self.batch["encoder_mask"].to(config.DEVICE)
        enc_out = self.model.encode(input_data, input_mask)
        output = self.model.classify(enc_out)
        probability = torch.sigmoid(output).item()
        assert 0 <= probability <= 1


if __name__ == "__main__":
    pytest.main()