def Create_Modle(max_seq_len, bert_ckpt_file):
with tf.io.gfile.GFile(bert_config_file,'r') as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = None
# Creating Model
bert = BertModelLayer.from_params(bert_params, name='bert')
# Keras Input Layer
input_ids = keras.layers.Input(shape=(max_seq_len, ), dtype='int32', name='input_ids')
bert_output = bert(input_ids)
# pirnting bert shape
print('Bert Shape: ', bert_output.shape)
cls_out = keras.layers.Lambda(lambda seq: seq[:,0,:])(bert_output)
cls_out = keras.layers.Dropout(0.5)(cls_out)
logits = keras.layers.Dense(units=768, activation='tanh')(cls_out)
logits = keras.layers.Dropout(0.5)(logits)
logits = keras.layers.Dense(units=len(classes), activation='softmax')(logits)
model = keras.Model(inputs=input_ids, outputs=logits)
model.build(input_shape = (None, max_seq_len))
load_stock_weights(bert, bert_ckpt_file) # loading weights
return model # returning model
def create_model(max_seq_len, bert_ckpt_file, classes):
with tf.io.gfile.GFile(config.bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = None
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = keras.layers.Input(shape=(max_seq_len, ),
dtype='int32',
name="input_ids")
bert_output = bert(input_ids)
print("bert shape", bert_output.shape)
cls_out = keras.layers.Lambda(lambda seq: seq[:, 0, :])(bert_output)
cls_out = keras.layers.Dropout(config.DROPOUT)(cls_out)
logits = keras.layers.Dense(units=768, activation="tanh")(cls_out)
logits = keras.layers.Dropout(config.DROPOUT)(logits)
logits = keras.layers.Dense(units=len(classes),
activation="softmax")(logits)
model = keras.Model(inputs=input_ids, outputs=logits)
model.build(input_shape=(None, max_seq_len))
load_stock_weights(bert, config.bert_ckpt_file)
print(model.summary())
model.compile(
optimizer=config.OPTIMIZER,
loss=config.LOSS,
metrics=[keras.metrics.SparseCategoricalAccuracy(name="acc")])
return model
def create_model(max_seq_len, classes, bert_ckpt_file):
with tf.io.gfile.GFile(config.BERT_CONFIG_FILE, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = None
bert = BertModelLayer.from_params(bert_params, name='bert')
input_ids = keras.layers.Input(shape=(max_seq_len, ),
dtype='int32',
name="input_ids")
bert_output = bert(input_ids)
print(f"Shape of BERT Embedding layer :{bert_output.shape}")
#input will be having a shape of (None,max_seq_len,hidden_layer(768))
#we can use lambda function to reshape it to (None,hidden_layer)
cls_out = keras.layers.Lambda(lambda seq: seq[:, 0, :])(bert_output)
cls_out = keras.layers.Dropout(0.5)(cls_out)
dense = keras.layers.Dense(units=768, activation="tanh")(cls_out)
dropout = keras.layers.Dropout(0.5)(dense)
output = keras.layers.Dense(units=len(classes),
activation="softmax")(dropout)
model = keras.Model(inputs=input_ids, outputs=output)
model.build(input_shape=(None, max_seq_len))
load_stock_weights(bert, bert_ckpt_file)
return model
def create_model(max_seq_len,adapter_size = 64): # Adapter size for adapter-bert
# Creating Base Layer from bert_config
with tf.io.gfile.GFile(BERT_CONFIG_FILE, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = adapter_size
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = keras.layers.Input(shape=(MAX_SEQ_LEN,), dtype='int32', name="input_ids")
output = bert(input_ids)
print("bert shape", output.shape)
cls_out = keras.layers.Lambda(lambda seq: seq[:, 0, :])(output)
cls_out = keras.layers.Dropout(0.5)(cls_out)
logits = keras.layers.Dense(units=768, activation="tanh")(cls_out)
logits = keras.layers.Dropout(0.5)(logits)
logits = keras.layers.Dense(units=2, activation="softmax")(logits)
model = keras.Model(inputs=input_ids, outputs=logits)
model.build(input_shape=(None, max_seq_len))
load_stock_weights(bert, BERT_CKPT_FILE)
if adapter_size is not None:
freeze_bert_layers(bert)
model.compile(optimizer=keras.optimizers.Adam(),
loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=[keras.metrics.SparseCategoricalAccuracy(name="acc")])
print(model.summary())
return model
def albert_params(albert_model: str):
"""Returns the ALBERT params for the specified TFHub model.
:param albert_model: either a model name or a checkpoint directory
containing an assets/albert_config.json
"""
if tf.io.gfile.isdir(albert_model):
config_file = os.path.join(
albert_model, "assets",
"albert_config.json") # google tfhub v2 weights
if not tf.io.gfile.exists(config_file):
config_file = os.path.join(
albert_model,
"albert_config.json") # google non-tfhub v2 weights
if tf.io.gfile.exists(config_file):
stock_config = loader.StockBertConfig.from_json_file(config_file)
else:
raise ValueError(
"No google-research ALBERT model found under:[{}] expecting albert_config.json or assets/albert_config.json"
.format(albert_model))
else:
if albert_model in albert_models_config: # google tfhub v1 weights
albert_config = albert_models_config[albert_model]
stock_config = loader.StockBertConfig.from_dict(
albert_config, return_instance=True, return_unused=False)
else:
raise ValueError(
"ALBERT model with name:[{}] not one of tfhub/google-research albert models, try one of:{}"
.format(albert_model, albert_models_tfhub))
params = loader.map_stock_config_to_params(stock_config)
return params
def create_model(max_seq_len, bert_ckpt_file):
with tf.io.gfile.GFile(bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = None
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = keras.layers.Input(shape=(max_seq_len, ),
dtype='int32',
name="input_ids")
bert_output = bert(input_ids)
print("bert shape", bert_output.shape)
cls_out = keras.layers.Lambda(lambda seq: seq[:, 0, :])(bert_output)
cls_out = keras.layers.Dropout(0.5)(cls_out)
logits = keras.layers.Dense(units=768, activation="tanh")(cls_out)
logits = keras.layers.Dropout(0.5)(logits)
logits = keras.layers.Dense(units=len(classes),
activation="softmax")(logits)
model = keras.Model(inputs=input_ids, outputs=logits)
model.build(input_shape=(None, max_seq_len))
load_stock_weights(bert, bert_ckpt_file)
return model
def model_factory(
name: str,
external_datasets: ExternalDatasets,
preprocessor: Preprocessor,
architecture: Architecture,
file_system: FileSystem,
) -> keras.Model:
"""The create_model method is a helper which accepts
max input sequence length and the number of intents
(classification bins/buckets). The logic returns a
BERT evaluator that matches the specified architecture.
:param name:
:type name:
:param external_datasets:
:type external_datasets:
:param preprocessor:
:type preprocessor:
:param architecture:
:type architecture:
:param file_system:
:type file_system:
:return:
:rtype:
"""
with tf.io.gfile.GFile(file_system.get_bert_config_path()) as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = None
bert = BertModelLayer.from_params(bert_params, name=name)
input_ids = keras.layers.Input(
shape=(preprocessor.max_sequence_length, ),
dtype='int32',
name="input_ids")
bert_output = bert(input_ids)
clf_out = keras.layers.Lambda(lambda seq: seq[:, 0, :])(bert_output)
clf_out = keras.layers.Dropout(
architecture.clf_out_dropout_rate)(clf_out)
logits = keras.layers.Dense(
units=BertModelParameters().bert_h_param,
activation=architecture.clf_out_activation)(clf_out)
logits = keras.layers.Dropout(architecture.logits_dropout_rate)(logits)
logits = keras.layers.Dense(
units=len(external_datasets.all_intents()),
activation=architecture.logits_activation)(logits)
model = keras.Model(inputs=input_ids, outputs=logits)
model.build(input_shape=(None, preprocessor.max_sequence_length))
load_stock_weights(bert, file_system.get_bert_model_path())
return model
def build(self,
max_seq_length,
bert_ckpt_file=bert_ckpt_file_location,
**kwargs):
optimizer = kwargs.get("optimizer", "adam")
metrics = kwargs.get("metrics", ['accuracy'])
adapter_size = kwargs.get("adapter_size", 64)
dropout_rate = kwargs.get('dropout_rate', 0.5)
# adapter_size = 64 # see - arXiv:1902.00751
# create the bert layer
with tf.io.gfile.GFile(
os.path.join(abs_path, bert_config_file_location),
"r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = adapter_size
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = tf.keras.layers.Input(shape=(max_seq_length, ),
dtype='int32',
name="input_ids")
output = bert(input_ids)
print("bert shape", output.shape)
cls_out = tf.keras.layers.Lambda(lambda seq: seq[:, 0, :])(output)
cls_out = tf.keras.layers.Dropout(0.5)(cls_out)
dense_out_1 = tf.keras.layers.Dense(units=768,
activation="tanh")(cls_out)
dense_out_1 = tf.keras.layers.Dropout(dropout_rate)(dense_out_1)
dense_out_2 = tf.keras.layers.Dense(units=200,
activation="softmax")(dense_out_1)
dense_out_2 = tf.keras.layers.Dropout(dropout_rate)(dense_out_2)
logits = tf.keras.layers.Dense(units=len(self.classes),
activation='softmax')(dense_out_2)
self.model = tf.keras.Model(inputs=input_ids, outputs=logits)
self.model.build(input_shape=(None, max_seq_length))
# load the pre-trained model weights
load_stock_weights(bert, os.path.join(abs_path, bert_ckpt_file))
# freeze weights if adapter-BERT is used
if adapter_size is not None:
freeze_bert_layers(bert)
self.model.compile(optimizer=optimizer,
loss=tf.keras.losses.SparseCategoricalCrossentropy(
from_logits=True),
metrics=metrics)
self.model.summary()
def create_model(self, type: str, adapter_size=None):
"""Creates a classification model. Input parameters:
type: "binary" to build a model for binary classification, "multi" for multiclass classification. """
self.type = type
# adapter_size = 64 # see - arXiv:1902.00751
if type == 'binary':
class_count = 2
elif type == 'multi':
class_count = 3
else:
raise TypeError("Choose a proper type of classification")
# create the bert layer
with tf.io.gfile.GFile(self._bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = adapter_size
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = keras.layers.Input(shape=(self.max_seq_len,), dtype='int32', name="input_ids")
# token_type_ids = keras.layers.Input(shape=(max_seq_len,), dtype='int32', name="token_type_ids")
# output = bert([input_ids, token_type_ids])
output = bert(input_ids)
print("bert shape", output.shape)
cls_out = keras.layers.Lambda(lambda seq: seq[:, 0, :])(output)
cls_out = keras.layers.Dropout(0.3)(cls_out)
logits = keras.layers.Dense(units=768, activation="relu")(cls_out)
# logits = keras.layers.Dropout(0.3)(logits)
# logits = keras.layers.Dense(units=256, activation="relu")(logits)
logits = keras.layers.Dropout(0.4)(logits)
logits = keras.layers.Dense(units=class_count, activation="softmax")(logits)
# model = keras.Model(inputs=[input_ids , token_type_ids], outputs=logits)
# model.build(input_shape=[(None, max_seq_len), (None, max_seq_len)])
model = keras.Model(inputs=input_ids, outputs=logits)
model.build(input_shape=(None, self.max_seq_len))
# load the pre-trained model weights
load_stock_weights(bert, self._bert_ckpt_file)
# freeze weights if adapter-BERT is used
if adapter_size is not None:
self.freeze_bert_layers(bert)
model.compile(optimizer=keras.optimizers.Adam(),
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
# loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
metrics=[tf.keras.metrics.SparseCategoricalAccuracy(name="acc")]
# metrics=[tf.keras.metrics.BinaryAccuracy(name="acc")]
)
model.summary()
self.model = model
def createMultiModelMaximum(max_seq_len, bert_ckpt_file, bert_config_file,
NUM_CLASS):
with GFile(bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = None
bert_layer = BertModelLayer.from_params(bert_params, name="bert")
bert_in = Input(shape=(max_seq_len, ),
dtype='int32',
name="input_ids_bert")
bert_inter = bert_layer(bert_in)
cls_out = Lambda(lambda seq: seq[:, 0, :])(bert_inter)
cls_out = Dropout(0.5)(cls_out)
bert_out = Dense(units=768, activation="tanh")(cls_out) # 768 before
load_stock_weights(bert_layer, bert_ckpt_file)
# image models:
inceptionv3 = InceptionV3(weights='imagenet', include_top=False)
resnet50 = ResNet50(weights='imagenet', include_top=False)
res_out = resnet50.output
res_out = GlobalAveragePooling2D()(res_out)
res_out = Dropout(0.5)(res_out)
res_out = Dense(2048)(res_out)
res_out = Dropout(0.5)(res_out)
res_out = Dense(768)(res_out)
inc_out = inceptionv3.output
inc_out = GlobalAveragePooling2D()(inc_out)
inc_out = Dropout(0.5)(inc_out)
inc_out = Dense(2048)(inc_out)
inc_out = Dropout(0.5)(inc_out)
inc_out = Dense(768)(inc_out)
# merge = Concatenate()([res_out, inc_out, bert_out])
merge = Maximum()([res_out, inc_out, bert_out])
# restliche Layer
x = Dense(2048)(merge)
x = Dropout(0.5)(x)
x = Dense(1024)(x)
x = Dropout(0.5)(x)
x = Dense(512)(x)
x = Dropout(0.5)(x)
output = Dense(NUM_CLASS, activation='softmax', name='output_layer')(x)
model = Model(inputs=[resnet50.input, inceptionv3.input, bert_in],
outputs=output)
plot_model(model,
to_file='multiple_inputs_text.png',
show_shapes=True,
dpi=600,
expand_nested=False)
return model, 17
def create_model(max_seq_len,
bert_config_file,
bert_ckpt_file,
adapter_size=64):
"""Creates a classification model."""
# adapter_size = 64 # see - arXiv:1902.00751
# max_seq_len
# create the bert layer
with tf.io.gfile.GFile(bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = adapter_size
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = keras.layers.Input(shape=(max_seq_len, ),
dtype='int32',
name="input_ids")
# token_type_ids = keras.layers.Input(shape=(max_seq_len,), dtype='int32', name="token_type_ids")
# output = bert([input_ids, token_type_ids])
output = bert(input_ids)
print("bert shape", output.shape)
cls_out = keras.layers.Lambda(lambda seq: seq[:, 0, :])(output)
cls_out = keras.layers.Dropout(0.5)(cls_out)
logits = keras.layers.Dense(units=768, activation="tanh")(cls_out)
logits = keras.layers.Dropout(0.5)(logits)
logits = keras.layers.Dense(units=6, activation="softmax")(logits)
# model = keras.Model(inputs=[input_ids, token_type_ids], outputs=logits)
# model.build(input_shape=[(None, max_seq_len), (None, max_seq_len)])
model = keras.Model(inputs=input_ids, outputs=logits)
model.build(input_shape=(None, max_seq_len))
# load the pre-trained model weights
load_stock_weights(bert, bert_ckpt_file)
# freeze weights if adapter-BERT is used
if adapter_size is not None:
freeze_bert_layers(bert)
model.compile(
optimizer=keras.optimizers.Adam(),
loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=[keras.metrics.SparseCategoricalAccuracy(name="acc")])
model.summary()
return model
def get_adapter_BERT_layer(model_dir, adapter_size):
"""Create a adapter-BERT layer
Arguments:
model_dir {str} -- Path to the pretrained model files
adapter_size {int} -- Size of adapter
Returns:
BERT -- BERT layer
"""
with open(model_dir + '/bert_config.json', 'r') as fd:
bc = StockBertConfig.from_json_string(fd.read())
params = map_stock_config_to_params(bc)
params.adapter_size = adapter_size
return get_bert_layer(params)
def create_bert_model(self, max_seq_len=18):
bc = None
with tf.io.gfile.GFile(self.bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert = BertModelLayer.from_params(map_stock_config_to_params(bc),
name="bert")
input_ids = keras.layers.Input(shape=(max_seq_len,), dtype='int32', name="input_ids")
token_type_ids = keras.layers.Input(shape=(max_seq_len,), dtype='int32', name="token_type_ids")
output = bert([input_ids, token_type_ids])
model = keras.Model(inputs=[input_ids, token_type_ids], outputs=output)
return model, bert, (input_ids, token_type_ids)
def albert_params(albert_model: str):
"""Returns the ALBERT params for the specified TFHub model.
:param albert_model: either a model name or a checkpoint directory
containing an assets/albert_config.json
"""
config_file = os.path.join(albert_model, "assets", "albert_config.json")
if tf.io.gfile.exists(config_file):
stock_config = loader.StockBertConfig.from_json_file(config_file)
elif albert_model in albert_models_config:
albert_config = albert_models_config[albert_model]
stock_config = loader.StockBertConfig(**albert_config)
else:
raise ValueError(
"ALBERT model with name:[{}] not one of tfhub/google-research albert models, try one of:{}"
.format(albert_model, albert_models_tfhub))
params = loader.map_stock_config_to_params(stock_config)
return params
def _load_bert(self, bert_config_file, bert_ckpt_file): try: with tf.io.gfile.GFile(bert_config_file, 'r') as gf: bert_config = StockBertConfig.from_json_string(gf.read()) bert_params = map_stock_config_to_params(bert_config) bert_params.adapter_size = None bert = BertModelLayer.from_params(bert_params, name='bert') except Exception as e: print(e) raise e input_ = keras.layers.Input(shape=(self.max_seq_len, ), dtype='int64', name="input_ids") x = bert(input_) # take the first embedding of BERT as the output embedding output_ = keras.layers.Lambda(lambda seq: seq[:,0,:])(x) model = keras.Model(inputs=input_, outputs=output_) model.build(input_shape=(None, self.max_seq_len)) load_stock_weights(bert, bert_ckpt_file) return model
def __init__(self):
self.max_len = 29
self.config_path = './Intent_cl/Bert_model/bert_config.json'
self.data = pd.read_csv('./Intent_cl/Intent_dataset/category_data.csv')
with open('./Intent_cl/Bert_model/vocab.json', 'r') as read_file:
self.vocab = json.loads(read_file.read())
with tf.io.gfile.GFile(self.config_path, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
self.bert_params = map_stock_config_to_params(bc)
self.bert_params.adapter_size = None
self.intent_model = keras.models.load_model(
'./Intent_cl/Bert_model/nomal_news_weather_etc_kobert_model_category.h5',
custom_objects={
"BertModelLayer":
BertModelLayer.from_params(self.bert_params, name="bert")
})
self.classes = self.data.intent.unique().tolist()
def create_model(max_seq_len, lr=1e-5):
"""
Creates a BERT classification model.
The model architecutre is raw input -> BERT input -> drop out layer to prevent overfitting -> dense layer that outputs predicted probability.
max_seq_len: the maximum sequence length
lr: learning rate of optimizer
"""
# create the bert layer
with tf.io.gfile.GFile(bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = keras.layers.Input(shape=(max_seq_len, ),
dtype='int32',
name="input_ids")
output = bert(input_ids)
print("bert shape", output.shape)
cls_out = keras.layers.Lambda(lambda seq: seq[:, 0, :])(output)
# Dropout layer
cls_out = keras.layers.Dropout(0.8)(cls_out)
# Dense layer with probibility output
logits = keras.layers.Dense(units=2, activation="softmax")(cls_out)
model = keras.Model(inputs=input_ids, outputs=logits)
model.build(input_shape=(None, max_seq_len))
# load the pre-trained model weights
load_stock_weights(bert, bert_ckpt_file)
model.compile(
optimizer=keras.optimizers.Adam(learning_rate=lr),
loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=[keras.metrics.SparseCategoricalAccuracy(name="acc")])
model.summary()
return model
def create_text_model(max_seq_len,
bert_ckpt_file,
bert_config_file,
NUM_CLASS,
overwriteLayerAndEmbeddingSize=False,
isPreTrained=False,
pathToBertModelWeights=None,
isTrainable=True):
with GFile(bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
if overwriteLayerAndEmbeddingSize:
bc.max_position_embeddings = max_seq_len
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = None
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = Input(shape=(max_seq_len, ), dtype='int32', name="input_ids")
bert_output = bert(input_ids)
print("bert shape", bert_output.shape)
cls_out = Lambda(lambda seq: seq[:, 0, :],
name='bert_output_layer_768')(bert_output)
cls_out = Dropout(0.5)(cls_out)
output = Dense(NUM_CLASS, activation="softmax")(cls_out) #
model_bert = Model(inputs=input_ids, outputs=output, name='BERT')
model_bert.build(input_shape=(None, max_seq_len))
if not isPreTrained:
load_stock_weights(bert, bert_ckpt_file)
return model_bert
else:
model_bert.load_weights(pathToBertModelWeights)
if not isTrainable:
for layer in model_bert.layers:
layer.trainable = False
return model_bert, 2
def create_model(max_seq_len, bert_ckpt_dir, bert_config_file):
with tf.io.gfile.GFile(bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = None
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = tf.keras.layers.Input(shape = (max_seq_len, ), dtype= tf.int32, name= 'input_ids')
bert_output = bert(input_ids)
bert_output = bert_output[:,0,:]
drop_out = tf.keras.layers.Dropout(0.5)(bert_output)
d_out = tf.keras.layers.Dense(768, activation='tanh')(drop_out)
logits = tf.keras.layers.Dropout(0.5)(d_out)
out = tf.keras.layers.Dense(2, activation='softmax')(logits)
model = tf.keras.models.Model(inputs = input_ids, outputs = out)
model.summary()
load_stock_weights(bert, bert_ckpt_file)
return model
def __init__(self, bert_config_file, bert_ckpt_file, max_seq_len, lr=1e-5):
"""
bert_config_file: path to bert configuration parameters
bert_ckpt_file: path to pretrained bert checkpoint
max_seq_len: maximum sequence lenght
lr: learning rate
"""
# create the bert layer
with tf.io.gfile.GFile(bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = keras.layers.Input(shape=(max_seq_len, ),
dtype='int32',
name="input_ids")
output = bert(input_ids)
cls_out = keras.layers.Lambda(lambda seq: seq[:, 0, :])(output)
# Dropout layer
cls_out = keras.layers.Dropout(0.8)(cls_out)
# Dense layer with probibility output
logits = keras.layers.Dense(units=2, activation="softmax")(cls_out)
model = keras.Model(inputs=input_ids, outputs=logits)
model.build(input_shape=(None, max_seq_len))
# load the pre-trained model weights
load_stock_weights(bert, bert_ckpt_file)
model.compile(
optimizer=keras.optimizers.Adam(learning_rate=lr),
loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=[keras.metrics.SparseCategoricalAccuracy(name="acc")])
self.model = model
def create_model(config, adapter_size=64):
"""Creates a classification model."""
# create the bert layer
with tf.io.gfile.GFile(config.bert_config_file, "r") as reader:
bc = StockBertConfig.from_json_string(reader.read())
bert_params = map_stock_config_to_params(bc)
bert_params.adapter_size = adapter_size
bert = BertModelLayer.from_params(bert_params, name="bert")
input_ids = keras.layers.Input(shape=(config.max_seq_len, ),
dtype='int32',
name="input_ids")
output = bert(input_ids)
matmul_qk = tf.matmul(output, output, transpose_b=True)
attention_weights = tf.nn.softmax(matmul_qk, axis=-1)
logits = tf.matmul(attention_weights, output)
logits = tf.reduce_sum(logits, axis=1) * config.attn_weight
cls_out = keras.layers.Lambda(lambda seq: seq[:, 0, :])(
output) * config.cls_weight
logits = cls_out + logits
logits = keras.layers.Dropout(0.5)(logits)
logits = keras.layers.LayerNormalization()(logits)
logits = keras.layers.Dense(units=len(config.classes))(logits)
model = keras.Model(inputs=input_ids, outputs=logits)
model.build(input_shape=(None, config.max_seq_len))
# load the pre-trained model weights
load_stock_weights(bert, config.bert_ckpt_file)
# freeze weights if adapter-BERT is used
# if adapter_size is not None:
# freeze_bert_layers(bert)
sigmoid_cross_entropy = tf.keras.losses.BinaryCrossentropy(
from_logits=True, label_smoothing=config.label_smoothing)
tfa_focal_loss = tfa.losses.SigmoidFocalCrossEntropy(
alpha=config.focal_alpha, gamma=config.focal_gamma, from_logits=True)
loss_func_list = {
"sigmoid_cross_entropy_loss": sigmoid_cross_entropy,
"focal_loss": tfa_focal_loss
}
model.compile(optimizer=keras.optimizers.Adam(),
loss=loss_func_list[config.loss_func],
metrics=[
MultiLabelAccuracy(batch_size=config.batch_size),
MultiLabelPrecision(batch_size=config.batch_size),
MultiLabelRecall(batch_size=config.batch_size),
MultiLabelF1(batch_size=config.batch_size),
HammingLoss(batch_size=config.batch_size)
])
model.summary()
return model
def albert_params(albert_model: str):
albert_config = albert_models_config[albert_model]
stock_config = loader.StockBertConfig(**albert_config)
params = loader.map_stock_config_to_params(stock_config)
return params
