Python allennlp.models.load_archive() Examples

def test_train_model(self):
        params = lambda: Params(
            {
                "model": {"type": "constant"},
                "dataset_reader": {"type": "sequence_tagging"},
                "train_data_path": SEQUENCE_TAGGING_DATA_PATH,
                "validation_data_path": SEQUENCE_TAGGING_DATA_PATH,
                "data_loader": {"batch_size": 2},
                "trainer": {"type": "no_op"},
            }
        )

        serialization_dir = self.TEST_DIR / "serialization_directory"
        train_model(params(), serialization_dir=serialization_dir)
        archive = load_archive(str(serialization_dir / "model.tar.gz"))
        model = archive.model
        assert model.forward(torch.tensor([1, 2, 3]))["class"] == torch.tensor(98)
        assert model.vocab.get_vocab_size() == 9 

def test_extend_embedder_vocab(self):
        model_archive = str(
            self.FIXTURES_ROOT / "basic_classifier" / "serialization" / "model.tar.gz"
        )
        trained_model = load_archive(model_archive).model

        original_weight = trained_model._text_field_embedder.token_embedder_tokens.weight
        assert tuple(original_weight.shape) == (213, 10)

        counter = {"tokens": {"unawarded": 1}}
        trained_model.vocab._extend(counter)
        trained_model.extend_embedder_vocab()

        extended_weight = trained_model._text_field_embedder.token_embedder_tokens.weight
        assert tuple(extended_weight.shape) == (214, 10)

        assert torch.all(original_weight == extended_weight[:213, :]) 

def test_predictor():
    question_json = {"id": "1700", "question_tokens": ["@start@", "For", "what", "does", "a", "stove", "generally", "generate", "heat", "?", "@end@"], "choice_tokens_list": [["@start@", "warming", "the", "air", "in", "the", "area", "@end@"], ["@start@", "heating", "nutrients", "to", "appropriate", "temperatures", "@end@"], ["@start@", "entertaining", "various", "visitors", "and", "guests", "@end@"], ["@start@", "to", "create", "electrical", "charges", "@end@"]], "facts_tokens_list": [["@start@", "UML", "can", "generate", "code", "@end@"], ["@start@", "generate", "is", "a", "synonym", "of", "beget", "@end@"], ["@start@", "Heat", "is", "generated", "by", "a", "stove", "@end@"], ["@start@", "A", "sonnet", "is", "generally", "very", "structured", "@end@"], ["@start@", "A", "fundamentalist", "is", "generally", "right", "-", "wing", "@end@"], ["@start@", "menstruation", "is", "generally", "crampy", "@end@"], ["@start@", "an", "erection", "is", "generally", "pleasurable", "@end@"], ["@start@", "gunfire", "is", "generally", "lethal", "@end@"], ["@start@", "ejaculating", "is", "generally", "pleasurable", "@end@"], ["@start@", "Huddersfield", "is", "generally", "urban", "@end@"], ["@start@", "warming", "is", "a", "synonym", "of", "calefacient", "@end@"], ["@start@", "heat", "is", "related", "to", "warming", "air", "@end@"], ["@start@", "a", "stove", "is", "for", "warming", "food", "@end@"], ["@start@", "an", "air", "conditioning", "is", "for", "warming", "@end@"], ["@start@", "The", "earth", "is", "warming", "@end@"], ["@start@", "a", "heat", "source", "is", "for", "warming", "up", "@end@"], ["@start@", "A", "foyer", "is", "an", "enterance", "area", "@end@"], ["@start@", "Being", "nosey", "is", "not", "appropriate", "@end@"], ["@start@", "seize", "is", "a", "synonym", "of", "appropriate", "@end@"], ["@start@", "a", "fitting", "room", "is", "used", "for", "something", "appropriate", "@end@"], ["@start@", "appropriate", "is", "a", "synonym", "of", "allow", "@end@"], ["@start@", "appropriate", "is", "similar", "to", "befitting", "@end@"], ["@start@", "appropriate", "is", "similar", "to", "grade", "-", "appropriate", "@end@"], ["@start@", "grade", "-", "appropriate", "is", "similar", "to", "appropriate", "@end@"], ["@start@", "A", "parlor", "is", "used", "for", "entertaining", "guests", "@end@"], ["@start@", "a", "back", "courtyard", "is", "for", "entertaining", "guests", "@end@"], ["@start@", "guest", "is", "a", "type", "of", "visitor", "@end@"], ["@start@", "a", "family", "room", "is", "for", "entertaining", "guests", "@end@"], ["@start@", "cooking", "a", "meal", "is", "for", "entertaining", "guests", "@end@"], ["@start@", "buying", "a", "house", "is", "for", "entertaining", "guests", "@end@"], ["@start@", "having", "a", "party", "is", "for", "entertaining", "guests", "@end@"], ["@start@", "a", "dining", "area", "is", "used", "for", "entertaining", "guests", "@end@"], ["@start@", "visitor", "is", "related", "to", "guest", "@end@"], ["@start@", "guest", "is", "related", "to", "visitor", "@end@"], ["@start@", "Electrical", "charges", "are", "additive", "@end@"], ["@start@", "Lightning", "is", "an", "electrical", "charge", "@end@"], ["@start@", "electrons", "have", "electrical", "charge", "@end@"], ["@start@", "A", "judge", "is", "in", "charge", "in", "a", "courtroom", "@end@"], ["@start@", "charge", "is", "a", "synonym", "of", "accusation", "@end@"], ["@start@", "A", "consultant", "can", "charge", "a", "fee", "to", "a", "client", "@end@"], ["@start@", "charge", "is", "a", "synonym", "of", "commission", "@end@"], ["@start@", "charge", "is", "a", "synonym", "of", "cathexis", "@end@"], ["@start@", "charge", "is", "not", "cash", "@end@"], ["@start@", "arraign", "entails", "charge", "@end@"], ["@start@", "a", "stove", "generates", "heat", "for", "cooking", "usually", "@end@"], ["@start@", "preferences", "are", "generally", "learned", "characteristics", "@end@"], ["@start@", "a", "windmill", "does", "not", "create", "pollution", "@end@"], ["@start@", "temperature", "is", "a", "measure", "of", "heat", "energy", "@end@"], ["@start@", "a", "hot", "something", "is", "a", "source", "of", "heat", "@end@"], ["@start@", "the", "moon", "does", "not", "contain", "water", "@end@"], ["@start@", "sunlight", "produces", "heat", "@end@"], ["@start@", "an", "oven", "is", "a", "source", "of", "heat", "@end@"], ["@start@", "a", "hot", "substance", "is", "a", "source", "of", "heat", "@end@"], ["@start@", "a", "car", "engine", "is", "a", "source", "of", "heat", "@end@"], ["@start@", "as", "the", "amount", "of", "rainfall", "increases", "in", "an", "area", ",", "the", "amount", "of", "available", "water", "in", "that", "area", "will", "increase", "@end@"], ["@start@", "sound", "can", "travel", "through", "air", "@end@"], ["@start@", "the", "greenhouse", "effect", "is", "when", "carbon", "in", "the", "air", "heats", "a", "planet", "'s", "atmosphere", "@end@"], ["@start@", "a", "community", "is", "made", "of", "many", "types", "of", "organisms", "in", "an", "area", "@end@"], ["@start@", "air", "is", "a", "vehicle", "for", "sound", "@end@"], ["@start@", "rainfall", "is", "the", "amount", "of", "rain", "an", "area", "receives", "@end@"], ["@start@", "an", "animal", "requires", "air", "for", "survival", "@end@"], ["@start@", "humidity", "is", "the", "amount", "of", "water", "vapor", "in", "the", "air", "@end@"], ["@start@", "if", "some", "nutrients", "are", "in", "the", "soil", "then", "those", "nutrients", "are", "in", "the", "food", "chain", "@end@"], ["@start@", "as", "heat", "is", "transferred", "from", "something", "to", "something", "else", ",", "the", "temperature", "of", "that", "something", "will", "decrease", "@end@"], ["@start@", "uneven", "heating", "causes", "convection", "@end@"], ["@start@", "as", "temperature", "during", "the", "day", "increases", ",", "the", "temperature", "in", "an", "environment", "will", "increase", "@end@"], ["@start@", "uneven", "heating", "of", "the", "Earth", "'s", "surface", "cause", "wind", "@end@"], ["@start@", "an", "animal", "needs", "to", "eat", "food", "for", "nutrients", "@end@"], ["@start@", "soil", "contains", "nutrients", "for", "plants", "@end@"], ["@start@", "if", "two", "objects", "have", "the", "same", "charge", "then", "those", "two", "materials", "will", "repel", "each", "other", "@end@"], ["@start@", "water", "is", "an", "electrical", "conductor", "@end@"], ["@start@", "a", "battery", "is", "a", "source", "of", "electrical", "energy", "@end@"], ["@start@", "metal", "is", "an", "electrical", "energy", "conductor", "@end@"], ["@start@", "when", "an", "electrical", "circuit", "is", "working", "properly", ",", "electrical", "current", "runs", "through", "the", "wires", "in", "that", "circuit", "@end@"], ["@start@", "brick", "is", "an", "electrical", "insulator", "@end@"], ["@start@", "wood", "is", "an", "electrical", "energy", "insulator", "@end@"], ["@start@", "a", "toaster", "converts", "electrical", "energy", "into", "heat", "energy", "for", "toasting", "@end@"]], "gold_label": 1, "gold_facts": {"fact1": "a stove generates heat for cooking usually", "fact2": "cooking involves heating nutrients to higher temperatures"}, "label_probs": [0.002615198493003845, 0.9686304330825806, 0.008927381597459316, 0.01982697658240795], "label_ranks": [3, 0, 2, 1], "predicted_label": 1, }

    inputs = question_to_predictor_input(question_json)
    inputs = predictor_input_to_pred_input_with_full_question_text(inputs)
    print(json.dumps(inputs, indent=4))

    archive = load_archive('_trained_models/model_CN5_1202.tar.gz')
    predictor = Predictor.from_archive(archive, 'predictor-qa-mc-with-know-visualize')

    result = predictor.predict_json(inputs)

    print(result) 

def test_train_model_distributed(self):
        if torch.cuda.device_count() >= 2:
            devices = [0, 1]
        else:
            devices = [-1, -1]

        params = lambda: Params(
            {
                "model": {
                    "type": "simple_tagger",
                    "text_field_embedder": {
                        "token_embedders": {"tokens": {"type": "embedding", "embedding_dim": 5}}
                    },
                    "encoder": {"type": "lstm", "input_size": 5, "hidden_size": 7, "num_layers": 2},
                },
                "dataset_reader": {"type": "sequence_tagging"},
                "train_data_path": SEQUENCE_TAGGING_DATA_PATH,
                "validation_data_path": SEQUENCE_TAGGING_DATA_PATH,
                "data_loader": {"batch_size": 2},
                "trainer": {"num_epochs": 2, "optimizer": "adam"},
                "distributed": {"cuda_devices": devices},
            }
        )

        out_dir = os.path.join(self.TEST_DIR, "test_distributed_train")
        train_model(params(), serialization_dir=out_dir)

        # Check that some logs specific to distributed
        # training are where we expect.
        serialized_files = os.listdir(out_dir)
        assert "out_worker0.log" in serialized_files
        assert "out_worker1.log" in serialized_files
        assert "model.tar.gz" in serialized_files

        # Check we can load the serialized model
        assert load_archive(out_dir).model 

def test_inspect_model_parameters(self):
        model_archive = str(
            self.FIXTURES_ROOT / "basic_classifier" / "serialization" / "model.tar.gz"
        )
        parameters_inspection = str(
            self.FIXTURES_ROOT / "basic_classifier" / "parameters_inspection.json"
        )
        model = load_archive(model_archive).model
        with open(parameters_inspection) as file:
            parameters_inspection_dict = json.load(file)
        assert parameters_inspection_dict == util.inspect_parameters(model) 

def fine_tune_model_from_file_paths(model_archive_path     ,
                                    config_file     ,
                                    serialization_dir     ,
                                    overrides      = u"",
                                    extend_vocab       = False,
                                    file_friendly_logging       = False)         :
    u"""
    A wrapper around :func:`fine_tune_model` which loads the model archive from a file.

    Parameters
    ----------
    model_archive_path : ``str``
        Path to a saved model archive that is the result of running the ``train`` command.
    config_file : ``str``
        A configuration file specifying how to continue training.  The format is identical to the
        configuration file for the ``train`` command, but any contents in the ``model`` section is
        ignored (as we are using the provided model archive instead).
    serialization_dir : ``str``
        The directory in which to save results and logs. We just pass this along to
        :func:`fine_tune_model`.
    overrides : ``str``
        A JSON string that we will use to override values in the input parameter file.
    file_friendly_logging : ``bool``, optional (default=False)
        If ``True``, we make our output more friendly to saved model files.  We just pass this
        along to :func:`fine_tune_model`.
    """
    # We don't need to pass in `cuda_device` here, because the trainer will call `model.cuda()` if
    # necessary.
    archive = load_archive(model_archive_path)
    params = Params.from_file(config_file, overrides)
    return fine_tune_model(model=archive.model,
                           params=params,
                           serialization_dir=serialization_dir,
                           extend_vocab=extend_vocab,
                           file_friendly_logging=file_friendly_logging) 

def test_fine_tune_does_not_expand_vocab_by_default(self):
        params = Params.from_file(self.config_file)
        # snli2 has a new token in it
        params[u"train_data_path"] = unicode(self.FIXTURES_ROOT / u'data' / u'snli2.jsonl')

        model = load_archive(self.model_archive).model

        # By default, no vocab expansion.
        fine_tune_model(model, params, self.serialization_dir) 

def test_fine_tune_runtime_errors_with_vocab_expansion(self):
        params = Params.from_file(self.config_file)
        params[u"train_data_path"] = unicode(self.FIXTURES_ROOT / u'data' / u'snli2.jsonl')

        model = load_archive(self.model_archive).model

        # If we do vocab expansion, we get a runtime error because of the embedding.
        with pytest.raises(RuntimeError):
            fine_tune_model(model, params, self.serialization_dir, extend_vocab=True) 

def test_fine_tune_nograd_regex(self):
        original_model = load_archive(self.model_archive).model
        name_parameters_original = dict(original_model.named_parameters())
        regex_lists = [[],
                       [u".*attend_feedforward.*", u".*token_embedder.*"],
                       [u".*compare_feedforward.*"]]
        for regex_list in regex_lists:
            params = Params.from_file(self.config_file)
            params[u"trainer"][u"no_grad"] = regex_list
            shutil.rmtree(self.serialization_dir, ignore_errors=True)
            tuned_model = fine_tune_model(model=original_model,
                                          params=params,
                                          serialization_dir=self.serialization_dir)
            # If regex is matched, parameter name should have requires_grad False
            # If regex is matched, parameter name should have same requires_grad
            # as the originally loaded model
            for name, parameter in tuned_model.named_parameters():
                if any(re.search(regex, name) for regex in regex_list):
                    assert not parameter.requires_grad
                else:
                    assert parameter.requires_grad\
                    == name_parameters_original[name].requires_grad
        # If all parameters have requires_grad=False, then error.
        with pytest.raises(Exception) as _:
            params = Params.from_file(self.config_file)
            params[u"trainer"][u"no_grad"] = [u"*"]
            shutil.rmtree(self.serialization_dir, ignore_errors=True)
            tuned_model = fine_tune_model(model=original_model,
                                          params=params,
                                          serialization_dir=self.serialization_dir) 

def __init__(self, model_path, top_k=3, cuda_device=-1):
        archive = load_archive(model_path,
                               cuda_device=cuda_device)

        config = archive.config
        prepare_environment(config)
        model = archive.model
        model.eval()

        self.model = model

        self._tokenizer = CharacterTokenizer()
        self._token_indexers = {'tokens': SingleIdTokenIndexer()}
        self._id_to_label = model.vocab.get_index_to_token_vocabulary(namespace='labels')
        self._top_k = top_k 

def eval_model(db: FeverDocDB, args) -> Model:
    archive = load_archive(args.archive_file, cuda_device=args.cuda_device)

    config = archive.config
    ds_params = config["dataset_reader"]

    model = archive.model
    model.eval()

    reader = FEVERReader(db,
                                 sentence_level=ds_params.pop("sentence_level",False),
                                 wiki_tokenizer=Tokenizer.from_params(ds_params.pop('wiki_tokenizer', {})),
                                 claim_tokenizer=Tokenizer.from_params(ds_params.pop('claim_tokenizer', {})),
                                 token_indexers=TokenIndexer.dict_from_params(ds_params.pop('token_indexers', {})))

    logger.info("Reading training data from %s", args.in_file)
    data = reader.read(args.in_file).instances

    actual = []
    predicted = []

    if args.log is not None:
        f = open(args.log,"w+")

    for item in tqdm(data):
        if item.fields["premise"] is None or item.fields["premise"].sequence_length() == 0:
            cls = "NOT ENOUGH INFO"
        else:
            prediction = model.forward_on_instance(item, args.cuda_device)
            cls = model.vocab._index_to_token["labels"][np.argmax(prediction["label_probs"])]

        if "label" in item.fields:
            actual.append(item.fields["label"].label)
        predicted.append(cls)

        if args.log is not None:
            if "label" in item.fields:
                f.write(json.dumps({"actual":item.fields["label"].label,"predicted":cls})+"\n")
            else:
                f.write(json.dumps({"predicted":cls})+"\n")

    if args.log is not None:
        f.close()


    if len(actual) > 0:
        print(accuracy_score(actual, predicted))
        print(classification_report(actual, predicted))
        print(confusion_matrix(actual, predicted))

    return model 

def eval_model(db: FeverDocDB, args) -> Model:
    archive = load_archive(args.archive_file, cuda_device=args.cuda_device, overrides=args.overrides)

    config = archive.config
    ds_params = config["dataset_reader"]

    model = archive.model
    model.eval()

    reader = FEVERReader(db,
                                 sentence_level=ds_params.pop("sentence_level",False),
                                 wiki_tokenizer=Tokenizer.from_params(ds_params.pop('wiki_tokenizer', {})),
                                 claim_tokenizer=Tokenizer.from_params(ds_params.pop('claim_tokenizer', {})),
                                 token_indexers=TokenIndexer.dict_from_params(ds_params.pop('token_indexers', {})))


    while True:

        claim = input("enter claim (or q to quit) >>")
        if claim.lower() == "q":
            break

        ranker = retriever.get_class('tfidf')(tfidf_path=args.model)

        p_lines = []
        pages,_ = ranker.closest_docs(claim,5)

        for page in pages:
            lines = db.get_doc_lines(page)
            lines = [line.split("\t")[1] if len(line.split("\t")[1]) > 1 else "" for line in lines.split("\n")]

            p_lines.extend(zip(lines, [page] * len(lines), range(len(lines))))

        scores = tf_idf_sim(claim, [pl[0] for pl in p_lines])
        scores = list(zip(scores, [pl[1] for pl in p_lines], [pl[2] for pl in p_lines], [pl[0] for pl in p_lines]))
        scores = list(filter(lambda score: len(score[3].strip()), scores))
        sentences_l = list(sorted(scores, reverse=True, key=lambda elem: elem[0]))

        sentences = [s[3] for s in sentences_l[:5]]
        evidence = " ".join(sentences)


        print("Best pages: {0}".format(repr(pages)))

        print("Evidence:")
        for idx,sentence in enumerate(sentences_l[:5]):
            print("{0}\t{1}\t\t{2}\t{3}".format(idx+1, sentence[0], sentence[1],sentence[3]) )

        item = reader.text_to_instance(evidence, claim)

        prediction = model.forward_on_instance(item, args.cuda_device)
        cls = model.vocab._index_to_token["labels"][np.argmax(prediction["label_probs"])]
        print("PREDICTED: {0}".format(cls))
        print()