def get_model(self):
num_words = len(self.dataset.alphabet)
num_labels = len(self.dataset.labels)
model = Sequential()
# MIR Embedding turns positive integers into dense vectors, for 1st layer of model. Why? Paper uses one-hot
model.add(
Embedding(
num_words, # MIR input dimension
self.config.embed_size, # MIR dense embedding
mask_zero=True))
model.add(Dropout(self.config.input_dropout))
for _ in range(self.config.recurrent_stack_depth):
model.add(
Bidirectional(
LSTM(self.config.num_lstm_units, return_sequences=True)))
# MIR return_sequences means return full sequence, not just last output
model.add(Dropout(self.config.output_dropout))
# MIR Does the paper have both input- and output-dropout?
model.add(TimeDistributed(Dense(num_labels, activation='softmax')))
# TODO Add Viterbi decoder here, see Kuru et al.
optimizer = Adam(lr=self.config.learning_rate, clipnorm=1.0)
# MIR Add precision, recall, f1 metrics for all labels
extra_metrics = {
"precision_null": keras_metrics.precision(label=0),
"precision_loc": keras_metrics.precision(label=1),
"precision_org": keras_metrics.precision(label=2),
"precision_per": keras_metrics.precision(label=3),
"recall_null": keras_metrics.recall(label=0),
"recall_loc": keras_metrics.recall(label=1),
"recall_org": keras_metrics.recall(label=2),
"recall_per": keras_metrics.recall(label=3),
"f1_null": keras_metrics.f1_score(label=0),
"f1_loc": keras_metrics.f1_score(label=1),
"f1_org": keras_metrics.f1_score(label=2),
"f1_per": keras_metrics.f1_score(label=3),
}
model.compile(optimizer=optimizer,
loss='categorical_crossentropy',
metrics=[
'categorical_accuracy',
self.non_null_label_accuracy,
].extend(extra_metrics))
# MIR non_null_label_accuracy is a func
return model
def construct_model(self):
self.model = Sequential()
self.model.add(
Dense(units=100, activation='relu', input_dim=self.n_inputs))
self.model.add(Dense(units=64, activation='relu'))
self.model.add(Dense(units=self.n_classes, activation='softmax'))
# Calculate precision for the second label.
pa = keras_metrics.precision(label=1)
pna = keras_metrics.precision(label=0)
# Calculate recall for the first label.
ra = keras_metrics.recall(label=1)
rna = keras_metrics.recall(label=0)
self.model.compile(loss=self.loss, \
optimizer=self.optimizer, \
metrics=['accuracy',pa,pna,ra,rna])
def build_model_architecture(x_shape):
model = Sequential()
model.add(
Dense(128,
input_dim=x_shape,
activation='relu',
kernel_regularizer=l2(0.01)))
model.add(Dropout(0.2))
model.add(Dense(64, activation='relu', kernel_regularizer=l2(0.01)))
model.add(Dropout(0.2))
# for binary classifiers
model.add(Dense(1, activation='sigmoid'))
model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=[
'accuracy',
keras_metrics.precision(),
keras_metrics.recall()
])
model.summary()
# early stop
early_stop = EarlyStopping(monitor='val_loss',
min_delta=0.01,
mode='auto',
verbose=1,
patience=5)
return early_stop, model
def evaluate_model(self):
if os.path.isfile(self.weights_file):
print("Saved model found")
self.model = load_model(self.weights_file, custom_objects={'binary_recall': keras_metrics.recall(), 'binary_precision': keras_metrics.precision(
), 'categorical_recall': keras_metrics.recall(), 'categorical_precision': keras_metrics.precision(), 'recall': keras_metrics.recall(), 'precision': keras_metrics.precision()})
print("Saved model loaded successfully")
self.model.compile(optimizer=self.optimizer,
loss=self.args.get('loss'),
metrics=["accuracy", keras_metrics.recall(), keras_metrics.precision()])
test_datagen = ImageDataGenerator(
rescale=1./255,
shear_range=0.5,
zoom_range=0.5,
horizontal_flip=True,
vertical_flip=True,
rotation_range=90,
)
test_set = test_datagen.flow_from_directory(
self.test_path,
target_size=self.target_size,
batch_size=1,
class_mode=self.args.get('class_mode'),
)
self.history_test = self.model.evaluate_generator(
test_set, steps=self.steps_per_epoch)
print(self.model.metrics_names, self.history_test)
else:
print("Model not found")
def compile_model_for_fine_tuning(self, lr=0.0001):
# we need to recompile the model for these modifications to take effect
self.model.compile(
optimizer=optimizers.SGD(lr=lr, momentum=0.9),
loss='sparse_categorical_crossentropy',
metrics=['accuracy', f1_score, keras_metrics.precision(), keras_metrics.recall()]
)
def compile_model(self, lr=0.001):
optimizer = optimizers.Adam(learning_rate=lr, beta_1=0.9, beta_2=0.999, amsgrad=False)
self.model.compile(
loss='sparse_categorical_crossentropy',
optimizer=optimizer,
metrics=['accuracy', f1_score, keras_metrics.precision(), keras_metrics.recall()]
)
def cnn(filters,
pooling_size=2,
epochs=15,
table_folder="/",
kernel_size=3,
input_dim=34,
batch_size=32,
nb_filters=34,
time_from=32,
time_to=8,
downsample_ratio=None,
oversample=None):
timesteps = time_from - time_to
X_train, X_test, y_train, y_test, churn_number, total_number, feature_names = import_and_preprocess_table(
timesteps, time_from, time_to, filters, table_folder, downsample_ratio,
oversample)
print("Creating layers...")
model = Sequential()
model.add(
Conv1D(nb_filters,
kernel_size=kernel_size,
input_shape=(timesteps, input_dim),
activation='relu'))
model.add(MaxPooling1D(pooling_size))
# model.add(Conv1D(nb_filters, kernel_size=kernel_size, activation='relu'))
# model.add(Conv1D(nb_filters*2, kernel_size=3, activation='relu'))
# model.add(GlobalAveragePooling1D())
# model.add(Dropout(0.5))
model.add(Flatten())
# model.add(Dense(128, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
print("Compiling model...")
model.compile(loss='mean_squared_error',
optimizer='rmsprop',
metrics=[
'accuracy',
keras_metrics.precision(),
keras_metrics.recall(),
keras_metrics.f1_score()
])
print("Fitting model...")
print(model.summary())
callback = [EarlyStopping(monitor='loss', patience=5)]
history = model.fit(X_train,
y_train,
validation_data=(X_test, y_test),
batch_size=batch_size,
epochs=epochs) #, callbacks=callback)
score = model.evaluate(X_test, y_test, batch_size=batch_size)
y_pred = model.predict(X_test)
log_to_csv("cnn", score, history, filters,
table_folder, input_dim, batch_size, time_from, time_to,
model.to_json(), nb_filters, kernel_size)
return [score, history, churn_number, total_number, y_pred]
def test_save_load(self):
custom_objects = {
"true_positive": keras_metrics.true_positive(sparse=self.sparse),
"true_negative": keras_metrics.true_negative(sparse=self.sparse),
"false_positive": keras_metrics.false_positive(sparse=self.sparse),
"false_negative": keras_metrics.false_negative(sparse=self.sparse),
"precision": keras_metrics.precision(sparse=self.sparse),
"recall": keras_metrics.recall(sparse=self.sparse),
"f1_score": keras_metrics.f1_score(sparse=self.sparse),
"sin": keras.backend.sin,
"abs": keras.backend.abs,
}
x, y = self.samples(100)
self.model.fit(x, y, epochs=10)
with tempfile.NamedTemporaryFile() as file:
self.model.save(file.name, overwrite=True)
model = keras.models.load_model(file.name,
custom_objects=custom_objects)
expected = self.model.evaluate(x, y)[1:]
received = model.evaluate(x, y)[1:]
self.assertEqual(expected, received)
def seq_boi(tokenizer): #loads the function and creates a BI_rnn network
embedding_matrix = get_embedding_vectors(tokenizer)
model = Sequential()
model.add(Embedding(len(tokenizer.word_index)+1,
EMBEDDING_SIZE,
weights=[embedding_matrix],
trainable=False,
input_length =SEQUENCE_LENGTH))
from keras.layers import Dense, Activation, Flatten
model.add(Dropout(0.8))
model.add(Conv1D(filters,
kernel_size,
padding='same',
activation='softmax',
strides=1))
model.add(MaxPooling1D(pool_size=pool_size))
model.add(Bidirectional(LSTM(1024))) #creates given network
model.add(Dense(2, activation='relu'))
model.add(Dense(2, activation='softmax'))
plot_model(model, to_file='model_plot.png', show_shapes=True, show_layer_names=True)
model.compile(optimizer="adam", loss="categorical_crossentropy",
metrics=["accuracy", keras_metrics.precision(), keras_metrics.recall()])
model.summary()
return model
def get_model(alpha=1, depth_multiplier=1, pooling='avg', lr=0.00001):
base_mobilenetv2_model = MobileNetV2(alpha=alpha,
depth_multiplier=depth_multiplier,
input_shape=(224, 224, 1),
include_top=False,
weights=None,
classes=1,
pooling=pooling)
top_model = Sequential()
top_model.add(Dense(512))
top_model.add(Dropout(0.5))
top_model.add(Dense(1, activation='sigmoid'))
# Create model.
model = Model(inputs=base_mobilenetv2_model.input,
outputs=top_model(base_mobilenetv2_model.output))
optimizer = Adam(lr=lr)
model.compile(
optimizer=optimizer,
loss='binary_crossentropy',
metrics=[keras.metrics.binary_accuracy,
keras_metrics.precision()])
return model
def test_metrics(self):
tp = keras_metrics.true_positive()
fp = keras_metrics.false_positive()
fn = keras_metrics.false_negative()
precision = keras_metrics.precision()
recall = keras_metrics.recall()
model = keras.models.Sequential()
model.add(keras.layers.Dense(1, activation="sigmoid", input_dim=2))
model.add(keras.layers.Dense(1, activation="softmax"))
model.compile(optimizer="sgd",
loss="binary_crossentropy",
metrics=[tp, fp, fn, precision, recall])
samples = 1000
x = numpy.random.random((samples, 2))
y = numpy.random.randint(2, size=(samples, 1))
model.fit(x, y, epochs=1, batch_size=10)
metrics = model.evaluate(x, y, batch_size=10)[1:]
tp_val = metrics[0]
fp_val = metrics[1]
fn_val = metrics[2]
precision = metrics[3]
recall = metrics[4]
expected_precision = tp_val / (tp_val + fp_val)
expected_recall = tp_val / (tp_val + fn_val)
self.assertAlmostEqual(expected_precision, precision, delta=0.05)
self.assertAlmostEqual(expected_recall, recall, delta=0.05)
def create_model(input_length):
print ('Creating model...')
model = Sequential()
model.add(Dense(64, input_dim=9, init='uniform'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(64, init='uniform'))
model.add(Activation('tanh'))
model.add(BatchNormalization())
model.add(Dropout(0.5))
model.add(Dense(1, init='uniform'))
model.add(Activation('softmax'))
sgd = SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
adam = Adam(lr=0.05)
print ('Compiling...')
model.compile(loss='categorical_crossentropy',
optimizer=Adam(lr=0.12, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False),
metrics=['accuracy',keras_metrics.precision(), keras_metrics.recall(),keras_metrics.false_positive(), keras_metrics.false_negative()])
#keras_metrics.false_positive(), keras_metrics.false_negative()
return model
def load(self, file_name=None):
"""
:param file_name: [model_file_name, weights_file_name]
:return:
"""
with self.graph.as_default():
with self.session.as_default():
try:
model_name = file_name[0]
weights_name = file_name[1]
if model_name is not None:
# load the model
filepath = os.path.join(self.model_folder, model_name)
self.model = tf.keras.models.load_model(
filepath,
custom_objects={
"f1_score": f1_score,
"binary_precision": keras_metrics.precision(),
"binary_recall": keras_metrics.recall(),
})
if weights_name is not None:
# load the weights
weights_path = os.path.join(self.model_folder,
weights_name)
self.model.load_weights(weights_path)
print("Neural Network loaded: ")
print('\t' + "Neural Network model: " + model_name)
print('\t' + "Neural Network weights: " + weights_name)
return True
except Exception as e:
print(e)
return False
def baseline_model(self):
model = Sequential()
model.add(
Dense(30,
input_dim=75,
kernel_initializer='normal',
activation='elu'))
# model.add(ActivityRegularization(l1=0.05, l2=0.05))
model.add(Dropout(0.3, noise_shape=None, seed=None))
model.add(Dense(20, kernel_initializer='normal', activation='elu'))
model.add(Dense(1, kernel_initializer='normal', activation='sigmoid'))
model.compile(loss='binary_crossentropy',
optimizer='adamax',
metrics=[
'accuracy',
keras_metrics.precision(),
keras_metrics.recall()
])
if self.summary == True:
print(model.summary())
if self.plot == True:
plot_model(model,
to_file='model_plot.png',
show_shapes=True,
show_layer_names=False)
return model
def get_model(tokenizer, lstm_units):
"""
Constructs the model,
Embedding vectors => LSTM => 2 output Fully-Connected neurons with softmax activation
"""
# get the GloVe embedding vectors
embedding_matrix = get_embedding_vectors(tokenizer)
model = Sequential()
model.add(
Embedding(
len(tokenizer.word_index) + 1,
EMBEDDING_SIZE,
weights=[embedding_matrix],
trainable=False,
input_length=SEQUENCE_LENGTH,
))
model.add(LSTM(lstm_units, recurrent_dropout=0.2))
model.add(Dropout(0.3))
model.add(Dense(2, activation="sigmoid"))
# compile as rmsprop optimizer
# aswell as with recall metric
model.compile(
optimizer="adam",
loss="categorical_crossentropy",
metrics=[
"accuracy",
keras_metrics.precision(),
keras_metrics.recall()
],
)
model.summary()
return model
def __model__(self):
"""Build & compile model keras.
:return: (Keras.Sequential) model deep
"""
# TODO refactor this shit code
mobilenet = MobileNet(weights='imagenet',
include_top=False,
input_shape=(224, 224, 3))
init = mobilenet.output
pool1 = GlobalAveragePooling2D()(init)
l1 = Dense(1024)(pool1)
act1 = Activation(activation="relu")(l1)
drop1 = Dropout(0.2)(act1)
l2 = Dense(self.number_classes)(drop1)
output = Activation(activation="softmax")(l2)
model = Model(inputs=mobilenet.input, outputs=output)
for layer in model.layers[:-6]:
layer.trainable = False
metrics = [
'accuracy',
keras_metrics.precision(),
keras_metrics.recall()
]
model.compile(optimizer='Adam', loss=self.loss, metrics=metrics)
return model
def __model__(self):
"""Build & compile the keras model.
:return: (Keras.Sequential) model deep
"""
model = Sequential()
model.add(
Conv2D(64, (5, 5), activation='relu', input_shape=(48, 48, 1)))
model.add(MaxPooling2D(pool_size=(5, 5), strides=(2, 2)))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(AveragePooling2D(pool_size=(3, 3), strides=(2, 2)))
model.add(Conv2D(128, (3, 3), activation='relu'))
model.add(Conv2D(128, (3, 3), activation='relu'))
model.add(AveragePooling2D(pool_size=(3, 3), strides=(2, 2)))
model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(len(self.classes), activation='softmax'))
metrics = [
'accuracy',
keras_metrics.precision(),
keras_metrics.recall()
]
model.compile(loss='categorical_crossentropy',
optimizer="Adam",
metrics=metrics)
return model
def build_bert(nclass):
#config_path = str(config_path)
#checkpoint_path = str(checkpoint_path)
# 读取bert预训练模型
# keras_bert是在Keras下对Bert最好的封装是
# 真正调用bert的就这么一行代码
bert_model = load_trained_model_from_checkpoint(config_path, checkpoint_path, seq_len=None)
for l in bert_model.layers:
l.trainable = True
x1_in = Input(shape=(None,))
x2_in = Input(shape=(None,))
x = bert_model([x1_in, x2_in])
# # 取出[CLS]对应的向量用来做分类
x = Lambda(lambda x: x[:, 0])(x)
p = Dense(nclass, activation='softmax')(x)
model = Model([x1_in, x2_in], p)
model.compile(loss='binary_crossentropy',
optimizer=Adam(1e-5),
metrics=['accuracy', keras_metrics.precision(), keras_metrics.recall()])
print(model.summary())
return model
def train_cnn(x1, x2, y, num_epoch, batch_size):
# Get and compile model
model = siamese_net()
model.summary()
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy', keras_metrics.precision(), keras_metrics.recall(), keras_metrics.f1_score()])
# Shuffle data
x, y = unison_shuffled_copies(np.array([x1, x2]), y)
x_train, x_test, y_train, y_test = split_data(x, y, dual_input=True)
# Train model
start_time = time()
history = model.fit(x=x_train, y=y_train, epochs=num_epoch, verbose=1, validation_data=(x_test, y_test), shuffle=True, batch_size=batch_size, callbacks=[callbacks.EarlyStopping(monitor='val_loss', patience=2)])
end_time = time()
time_taken = end_time - start_time
epochs_trained = len(history.history['loss'])
print("Took {} to train over {} epochs with batch size {}".format(end_time - start_time, epochs_trained, batch_size))
# Plot train vs test metric graphs
plot_graph(history.history['loss'], history.history['val_loss'], 'loss', 'Loss', "loss_epoch{}_batch{}.png".format(epochs_trained, batch_size), 200)
plot_graph(history.history['acc'], history.history['val_acc'], 'accuracy', 'Accuracy', "acc_epoch{}_batch{}.png".format(epochs_trained, batch_size), 201)
plot_graph(history.history['recall'], history.history['val_recall'], 'recall', 'Recall', "rec_epoch{}_batch{}.png".format(epochs_trained, batch_size), 202)
plot_graph(history.history['precision'], history.history['val_precision'], 'precision', 'Precision', "pre_epoch{}_batch{}.png".format(epochs_trained, batch_size), 203)
plot_graph(history.history['f1_score'], history.history['val_f1_score'], 'f1', 'F1 Score', "f1_epoch{}_batch{}.png".format(epochs_trained, batch_size), 204)
# Save metrics
save_metrics(history, time_taken, num_epoch, batch_size, "metrics_epoch{}_batch{}.txt".format(epochs_trained, batch_size))
# Save model
save_model(model, 'model_epoch{}_batch{}.h5'.format(epochs_trained, batch_size))
def init_model(self):
image_shape = (IMAGE_HEIGHT, IMAGE_WIDTH, 3)
self.datagen = ImageDataGenerator(rotation_range=20,
width_shift_range=0.15,
height_shift_range=0.15,
zoom_range=0.1,
horizontal_flip=True,
vertical_flip=True,
fill_mode='nearest')
input_a, input_b = Input(shape=image_shape), Input(shape=image_shape)
if cnt.USE_VGG:
shared_model = get_shared_model_vgg(image_shape)
else:
shared_model = get_shared_model(image_shape)
if cnt.PRE_TRAIN_CLASSIFIER:
self.init_classifier()
self.classifier.load_weights(cnt.CLASSIFIER_MODEL_PATH)
shared_model.set_weights(self.classifier.layers[1].get_weights())
for layer in shared_model.layers:
layer.trainable = False
nlayer1 = shared_model(input_a)
nlayer2 = shared_model(input_b)
n_layer = Flatten()
nlayer1 = n_layer(nlayer1)
nlayer2 = n_layer(nlayer2)
n_layer = Dense(cnt.EMBEDDING_SIZE, activation='relu')
nlayer1 = n_layer(nlayer1)
nlayer2 = n_layer(nlayer2)
n_layer = BatchNormalization()
nlayer1 = n_layer(nlayer1)
nlayer2 = n_layer(nlayer2)
n_layer = Lambda(lambda x: K.l2_normalize(x, axis=1))
nlayer1 = n_layer(nlayer1)
nlayer2 = n_layer(nlayer2)
n_layer = Lambda(lambda x: K.sqrt(
K.sum(K.square(x[0] - x[1]), axis=1, keepdims=True)))(
[nlayer1, nlayer2])
out = Dense(1, activation="sigmoid")(n_layer)
self.model = Model(inputs=[input_a, input_b], outputs=[out])
adam = optimizers.Adam(lr=0.001)
self.model.compile(
optimizer=adam,
loss="mean_squared_error",
metrics=['accuracy',
km.precision(label=0),
km.recall(label=0)])
def MLP(train_x, train_y, test_x, test_y, epoch, batch_size):
# One hidden layer with 60 neurons
model = models.Sequential()
layer = layers.Dense(200,
activation="relu",
input_shape=(train_x.shape[1], ),
kernel_regularizer=regularizers.l2(1e-10),
activity_regularizer=regularizers.l1(1e-10))
model.add(layer)
model.add(layers.Dropout(0.1, noise_shape=None, seed=None))
#model.add(layers.Dropout(0.2, noise_shape=None, seed=None))
model.add(
layers.Dense(100,
activation="relu",
kernel_regularizer=regularizers.l2(1e-10),
activity_regularizer=regularizers.l1(1e-10)))
model.add(layers.Dropout(0.1, noise_shape=None, seed=None))
model.add(
layers.Dense(60,
activation="relu",
kernel_regularizer=regularizers.l2(1e-10),
activity_regularizer=regularizers.l1(1e-10)))
model.add(
layers.Dense(30,
activation="relu",
kernel_regularizer=regularizers.l2(1e-10),
activity_regularizer=regularizers.l1(1e-10)))
#model.add(layers.Dropout(0.1, noise_shape=None, seed=None))
model.add(
layers.Dense(10,
activation="relu",
kernel_regularizer=regularizers.l2(1e-10),
activity_regularizer=regularizers.l1(1e-10)))
model.add(layers.Dense(1, activation="sigmoid"))
model.summary()
adam = Adam(lr=0.0001,
beta_1=0.9,
beta_2=0.999,
epsilon=None,
decay=0.0,
amsgrad=False)
model.compile(loss='binary_crossentropy',
metrics=[
'accuracy',
keras_metrics.precision(),
keras_metrics.recall()
],
optimizer=adam)
# Get the loss and accuracy by using cross validation
results_mlp = model.fit(train_x,
train_y,
epochs=epoch,
batch_size=batch_size,
validation_data=(test_x, test_y))
#predict_label = model.predict_classes(train_x)
weights_mlp = layer.get_weights()
return results_mlp, weights_mlp
def train(model, X_train, X_test, y_train, y_test, save_path):
model.compile(Adam(lr=0.001), loss="sparse_categorical_crossentropy", \
metrics=['accuracy', keras_metrics.precision(), keras_metrics.recall()])
model.fit(X_train, y_train, \
validation_data=(X_test, y_test), \
batch_size=32, \
epochs=20
)
model.save(save_path)
def getModel(self):
num_classes = 10
optimizer = RMSprop(lr=0.001, rho=0.9, epsilon=1e-08, decay=0.0)
model = Sequential()
model.add(
Conv2D(128, (3, 3),
input_shape=(28, 28, 1),
activation='relu',
padding='same'))
model.add(AveragePooling2D(pool_size=(2, 2), strides=2))
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(
Conv2D(256, (3, 3),
input_shape=(28, 28, 1),
activation='relu',
padding='same'))
model.add(AveragePooling2D(pool_size=(2, 2), strides=2))
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(
Conv2D(512, (3, 3),
input_shape=(28, 28, 1),
activation='relu',
padding='same'))
model.add(AveragePooling2D(pool_size=(2, 2), strides=2))
model.add(Dropout(0.3))
model.add(BatchNormalization())
model.add(GlobalMaxPooling2D())
model.add(Dense(1024, activation='relu'))
model.add(BatchNormalization())
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax', name='predict'))
model.compile(loss='categorical_crossentropy',
optimizer=optimizer,
metrics=[
'accuracy',
keras_metrics.precision(),
keras_metrics.recall()
])
from keras.models import model_from_json
model_json = model.to_json()
with open("modelBrabo.json", "w") as json_file:
json_file.write(model_json)
return model
def load_model(self, verbose=0):
self.model = load_model(
self.save_name(),
custom_objects={'binary_precision': km.precision()})
if verbose >= 1:
print(self.model.summary())
history_name = self.save_name().split('.')[0] + '.json'
with open(history_name, 'r') as history_file:
self.history = json.load(history_file)
def lstm5(filters,
epochs=15,
table_folder="/",
save_file=None,
input_dim=34,
batch_size=32,
time_from=32,
time_to=8,
downsample_ratio=None,
oversample=None):
timesteps = time_from - time_to
X_train, X_test, y_train, y_test, churn_number, total_number, feature_names = import_and_preprocess_table(
timesteps, time_from, time_to, filters, table_folder, downsample_ratio,
oversample)
print("Creating layers...")
model = Sequential()
model.add(
LSTM(34, input_length=timesteps, input_dim=34, return_sequences=True))
model.add(Dropout(0.2))
model.add(LSTM(input_dim, return_sequences=True))
model.add(Dropout(0.2))
# model.add(LSTM(input_dim, return_sequences=True))
# model.add(Dropout(0.2))
# model.add(LSTM(input_dim, return_sequences=True))
# model.add(Dropout(0.2))
model.add(LSTM(input_dim))
model.add(Dropout(0.2))
model.add(Dense(1, activation='sigmoid'))
print("Compiling model...")
model.compile(loss='mean_squared_error',
optimizer='rmsprop',
metrics=[
'accuracy',
keras_metrics.precision(),
keras_metrics.recall(),
keras_metrics.f1_score()
])
print("Fitting model...")
print(model.summary())
callback = [EarlyStopping(monitor='val_loss', patience=5)]
history = model.fit(X_train,
y_train,
validation_data=(X_test, y_test),
batch_size=batch_size,
epochs=epochs) #, callbacks=callback)
score = model.evaluate(X_test, y_test, batch_size=batch_size)
y_pred = model.predict(X_test)
log_to_csv("lstm", score, history, filters, table_folder, input_dim,
batch_size, time_from, time_to, model.to_json())
return [score, history, churn_number, total_number, y_pred]
def main(model_name, model_file_path, training_data_path, epochs=20):
samples, labels = get_samples_labels(training_data_path)
vocabulary, tokenizer = get_vocabulary_tokenizer(samples)
num_classes = len(INDEX_NAME_CLASS_MAP)
num_tokens = len(vocabulary)
tokenized_samples = [tokenizer.tokenize(sample.text) for sample in samples]
max_document_length = tokenizer.get_max_length(tokenized_samples)
transformer = TextTransformer(class_map=NAME_INDEX_CLASS_MAP,
max_sequence_length=max_document_length,
vocabulary=vocabulary,
tokenizer=tokenizer)
X = [transformer.transform(sample.text) for sample in samples]
sample_batch_provider = SampleBatchProvider(
X=X,
y=labels,
num_labels=num_classes,
max_document_length=max_document_length,
max_token_length=num_tokens)
X, y = sample_batch_provider.get_batch(X, labels)
experiment = NgramCNNModel()
model = experiment.get_model(max_document_length=max_document_length,
num_classes=num_classes,
vocabulary_size=len(vocabulary))
model.compile(optimizer=Adam(),
loss="binary_crossentropy",
metrics=[
"accuracy",
keras_metrics.precision(),
keras_metrics.recall()
])
plot_model(model, show_shapes=True, to_file=model_name + '.png')
callbacks = [
GradientDebugger(),
TensorBoard(log_dir='/tmp/shooting_ngram_cnn'),
ModelCheckpoint(model_file_path,
monitor='val_acc',
verbose=1,
save_best_only=True,
mode='max'),
EarlyStopping(monitor='val_acc', patience=5, mode='max'),
ReduceLROnPlateau(factor=0.1, verbose=1),
]
model.fit(x=[X, X, X],
y=y,
batch_size=None,
epochs=epochs,
verbose=1,
callbacks=callbacks,
validation_split=0.2,
shuffle=True,
steps_per_epoch=20,
validation_steps=100)
def load_model(self, filepath):
print('[Model] Loading model from file %s' % filepath)
self.model = tf.keras.models.load_model(filepath,
custom_objects={
"f1_score":
f1_score,
"binary_precision":
keras_metrics.precision(),
"binary_recall":
keras_metrics.recall(),
})
def create_model_single():
opt = keras.optimizers.SGD(lr=0.01)
model = create_base()
model.compile(optimizer=opt,
loss="categorical_crossentropy",
metrics=[
"accuracy",
keras_metrics.precision(),
keras_metrics.recall()
])
return model
def create_model_multi():
opt = keras.optimizers.SGD(lr=0.01)
model = create_base()
parallel_model = multi_gpu_model(model, gpus=2)
parallel_model.compile(optimizer=opt,
loss="categorical_crossentropy",
metrics=[
"accuracy",
keras_metrics.precision(),
keras_metrics.recall()
])
return parallel_model
def model_impl():
x_train, y_train, x_test, y_test = get_train_data()
model = Sequential()
#model.add(LSTM(300, activation='relu', input_shape=(15, 300)))
model.add(LSTM(300, activation='tanh', input_shape=(15, 300)))
model.add(Dropout(0.2))
model.add(Dense(1, activation='sigmoid'))
model.summary()
model.compile(loss='binary_crossentropy', optimizer='sgd',
metrics=['accuracy', keras_metrics.precision(), keras_metrics.recall()])
model.fit(x_train, y_train, verbose=1, epochs=20, validation_data=(x_test, y_test))
model.save('lstm_model.h5')