Function test_resize_same_results

spacy/tests/pipeline/test_textcat.py:503–545

(name, textcat_config)

)
# fmt: on
def test_resize_same_results(name, textcat_config):
    # Ensure that the resized textcat classifiers still produce the same results for old labels
    fix_random_seed(0)
    nlp = English()
    pipe_config = {"model": textcat_config}
    textcat = nlp.add_pipe(name, config=pipe_config)

    train_examples = []
    for text, annotations in TRAIN_DATA_SINGLE_LABEL:
        train_examples.append(Example.from_dict(nlp.make_doc(text), annotations))
    optimizer = nlp.initialize(get_examples=lambda: train_examples)
    assert textcat.model.maybe_get_dim("nO") in [2, None]

    for i in range(5):
        losses = {}
        nlp.update(train_examples, sgd=optimizer, losses=losses)

    # test the trained model before resizing
    test_text = "I am happy."
    doc = nlp(test_text)
    assert len(doc.cats) == 2
    pos_pred = doc.cats["POSITIVE"]
    neg_pred = doc.cats["NEGATIVE"]

    # test the trained model again after resizing
    textcat.add_label("NEUTRAL")
    doc = nlp(test_text)
    assert len(doc.cats) == 3
    assert doc.cats["POSITIVE"] == pos_pred
    assert doc.cats["NEGATIVE"] == neg_pred
    assert doc.cats["NEUTRAL"] <= 1

    for i in range(5):
        losses = {}
        nlp.update(train_examples, sgd=optimizer, losses=losses)

    # test the trained model again after training further with new label
    doc = nlp(test_text)
    assert len(doc.cats) == 3
    assert doc.cats["POSITIVE"] != pos_pred
    assert doc.cats["NEGATIVE"] != neg_pred
    for cat in doc.cats:
        assert doc.cats[cat] <= 1


def test_error_with_multi_labels():