def load_model(
cls, model_path: str
) -> Union["ABCLabelingModel", "ABCClassificationModel"]:
model_config_path = os.path.join(model_path, 'model_config.json')
model_config = json.loads(open(model_config_path, 'r').read())
model = load_data_object(model_config)
model.embedding = load_data_object(model_config['embedding'])
model.text_processor = load_data_object(model_config['text_processor'])
model.label_processor = load_data_object(
model_config['label_processor'])
tf_model_str = json.dumps(model_config['tf_model'])
print(tf_model_str)
model.tf_model = tf.keras.models.model_from_json(
tf_model_str, custom_objects=kashgari.custom_objects)
if isinstance(model.tf_model.layers[-1], KConditionalRandomField):
model.layer_crf = model.tf_model.layers[-1]
model.tf_model.load_weights(
os.path.join(model_path, 'model_weights.h5'))
model.embedding.embed_model.load_weights(
os.path.join(model_path, 'embed_model_weights.h5'))
return model
def load_model(cls, model_path: str) -> 'Seq2Seq':
from kashgari.utils import load_data_object
model_config_path = os.path.join(model_path, 'model_config.json')
model_config = json.loads(open(model_config_path, 'r').read())
model = load_data_object(model_config)
# Load processors and embeddings
model.encoder_processor = load_data_object(model_config['encoder_processor'])
model.decoder_processor = load_data_object(model_config['decoder_processor'])
model.encoder_embedding = load_data_object(model_config['encoder_embedding'])
model.decoder_embedding = load_data_object(model_config['decoder_embedding'])
model._build_encoder_decoder()
# Load Model Weights
model.encoder_embedding.embed_model.load_weights(os.path.join(model_path, 'encoder_embed_weights.h5'))
model.decoder_embedding.embed_model.load_weights(os.path.join(model_path, 'decoder_embed_weights.h5'))
# ------ Fix Start -------
# load model issue on TF 2.3
# Unable to load weights saved in HDF5 format into a subclassed Model which has not created its variables yet.
# Call the Model first, then load the weights.
input_seq = model.encoder_processor.transform([['hello']],
seq_length=model.encoder_seq_length)
dec_input = tf.expand_dims([3], 0)
enc_hidden = tf.zeros((1, model.hidden_size))
dec_hidden = enc_hidden
enc_output, enc_hidden = model.encoder(input_seq, enc_hidden)
_ = model.decoder(dec_input, dec_hidden, enc_output)
# ------ Fix End -------
model.encoder.load_weights(os.path.join(model_path, 'encoder_weights.h5'))
model.decoder.load_weights(os.path.join(model_path, 'decoder_weights.h5'))
return model
def test_processor(self):
x_set, y_set = TestMacros.load_classification_corpus()
processor = ClassificationProcessor()
processor.build_vocab(x_set, y_set)
transformed_idx = processor.transform(y_set[20:40])
info_dict = processor.to_dict()
p2: ClassificationProcessor = load_data_object(info_dict)
assert (transformed_idx == p2.transform(y_set[20:40])).all()
assert y_set[20:40] == p2.inverse_transform(transformed_idx)
def load_model(
cls,
model_path: str,
custom_objects: Dict = None,
encoding: str = 'utf-8'
) -> Union["ABCLabelingModel", "ABCClassificationModel"]:
if custom_objects is None:
custom_objects = {}
if cls.__name__ not in custom_objects:
custom_objects[cls.__name__] = cls
model_config_path = os.path.join(model_path, 'model_config.json')
model_config = json.loads(
open(model_config_path, 'r', encoding=encoding).read())
model = load_data_object(model_config, custom_objects)
model.embedding = load_data_object(model_config['embedding'],
custom_objects)
model.text_processor = load_data_object(model_config['text_processor'],
custom_objects)
model.label_processor = load_data_object(
model_config['label_processor'], custom_objects)
tf_model_str = json.dumps(model_config['tf_model'])
model.tf_model = tf.keras.models.model_from_json(
tf_model_str, custom_objects=kashgari.custom_objects)
if isinstance(model.tf_model.layers[-1], KConditionalRandomField):
model.crf_layer = model.tf_model.layers[-1]
model.tf_model.load_weights(
os.path.join(model_path, 'model_weights.h5'))
model.embedding.embed_model.load_weights(
os.path.join(model_path, 'embed_model_weights.h5'))
return model
def test_text_processor(self):
x_set, y_set = TestMacros.load_labeling_corpus()
x_samples = random.sample(x_set, 5)
text_processor = SequenceProcessor(min_count=1)
text_processor.build_vocab(x_set, y_set)
text_idx = text_processor.transform(x_samples)
text_info_dict = text_processor.to_dict()
text_processor2: SequenceProcessor = load_data_object(text_info_dict)
text_idx2 = text_processor2.transform(x_samples)
sample_lengths = [len(i) for i in x_samples]
assert (text_idx2 == text_idx).all()
assert text_processor.inverse_transform(
text_idx, lengths=sample_lengths) == x_samples
assert text_processor2.inverse_transform(
text_idx2, lengths=sample_lengths) == x_samples
def test_multi_label_processor(self):
from kashgari.corpus import JigsawToxicCommentCorpus
file_path = TestMacros.jigsaw_mini_corpus_path
corpus = JigsawToxicCommentCorpus(file_path)
x_set, y_set = corpus.load_data()
corpus_gen = CorpusGenerator(x_set, y_set)
processor = ClassificationProcessor(multi_label=True)
processor.build_vocab_generator([corpus_gen])
transformed_idx = processor.transform(y_set[20:40])
info_dict = processor.to_dict()
p2: ClassificationProcessor = load_data_object(info_dict)
assert (transformed_idx == p2.transform(y_set[20:40])).all()
x1s = y_set[20:40]
x2s = p2.inverse_transform(transformed_idx)
for sample_x1, sample_x2 in zip(x1s, x2s):
assert sorted(sample_x1) == sorted(sample_x2)
def test_base_cases(self):
embedding = self.build_embedding()
x, y = SMP2018ECDTCorpus.load_data()
processor = SequenceProcessor()
processor.build_vocab(x, y)
embedding.setup_text_processor(processor)
samples = random.sample(x, sample_count)
res = embedding.embed(samples)
max_len = max([len(i) for i in samples]) + 2
if embedding.max_position is not None:
max_len = embedding.max_position
assert res.shape == (len(samples), max_len, embedding.embedding_size)
# Test Save And Load
embed_dict = embedding.to_dict()
embedding2 = load_data_object(embed_dict)
embedding2.setup_text_processor(processor)
assert embedding2.embed(samples).shape == (len(samples), max_len,
embedding.embedding_size)
def test_label_processor(self):
x_set, y_set = TestMacros.load_labeling_corpus()
text_processor = SequenceProcessor(build_vocab_from_labels=True,
min_count=1)
text_processor.build_vocab(x_set, y_set)
samples = random.sample(y_set, 20)
text_idx = text_processor.transform(samples)
text_info_dict = text_processor.to_dict()
text_processor2: SequenceProcessor = load_data_object(text_info_dict)
text_idx2 = text_processor2.transform(samples)
lengths = [len(i) for i in samples]
assert (text_idx2 == text_idx).all()
assert text_processor2.inverse_transform(text_idx,
lengths=lengths) == samples
assert text_processor2.inverse_transform(text_idx2,
lengths=lengths) == samples
text_idx3 = text_processor.transform(samples, seq_length=20)
assert [len(i) for i in text_idx3] == [20] * len(text_idx3)
