nlpx

Usage Sample ''''''''''''

.. code:: python

    from nlpx.text_token import Tokenizer
    from nlpx.model.classifier import TextCNNClassifier
    from nlpx.model.wrapper import ClassModelWrapper
    from nlpx.dataset import TokenDataset, PaddingTokenCollator

    if __name__ == '__main__':
        classes = ['class1', 'class2', 'class3'...]
        texts = [[str],]
        labels = [0, 0, 1, 2, 1...]
        tokenizer = Tokenizer.from_texts(texts, min_freq=5)
        sent = 'I love you'
        tokens = tokenizer.encode(sent, max_length=6)
        # [101, 66, 88, 99, 102, 0]
        sent = tokenizer.decode(sent)
        # ['<BOS>', 'I', 'love', 'you', '<EOS>', '<PAD>']

        tokens = tokenizer.batch_encode(texts, padding=False)
        X_train, X_test, y_train, y_test = train_test_split(tokens, labels, test_size=0.2)
        train_set = TokenDataset(X_train, y_train)
        test_set = TokenDataset(X_test, y_test)

        model = TextCNNClassifier(embed_dim=128, vocab_size=tokenizer.vocab_size, num_classes=len(classes))
        model_wrapper = ClassModelWrapper(model, classes=classes)
        model_wrapper.train(train_set, test_set, show_progress=True, collate_fn=PaddingTokenCollator(tokenizer.pad))

        result = model_wrapper.evaluate(test_set, collate_fn=PaddingTokenCollator(tokenizer.pad))
        # 0.953125

        result = model_wrapper.predict(torch.tensor(test_tokens, dtype=torch.long))
        # [0, 1]

        result = model_wrapper.predict_proba(torch.tensor(test_tokens, dtype=torch.long))
        # ([0, 1], array([0.99439645, 0.99190724], dtype=float32))

        result = model_wrapper.predict_classes(torch.tensor(test_tokens, dtype=torch.long))
        # ['class1', 'class2']

        result = model_wrapper.predict_classes_proba(torch.tensor(test_tokens, dtype=torch.long))
        # (['class1', 'class2'], array([0.99439645, 0.99190724], dtype=float32))