def ModelShare():
tweet_a = Input(shape=(280, 256))
tweet_b = Input(shape=(280, 256))
# This layer can take as input a matrix
# and will return a vector of size 64
shared_lstm = LSTM(64, return_sequences=True, name='lstm')
# When we reuse the same layer instance
# multiple times, the weights of the layer
# are also being reused
# (it is effectively *the same* layer)
encoded_a = shared_lstm(tweet_a)
encoded_b = shared_lstm(tweet_b)
# We can then concatenate the two vectors:
merged_vector = concatenate([encoded_a, encoded_b], axis=-1)
# And add a logistic regression on top
predictions = Dense(1, activation='sigmoid')(merged_vector)
# We define a trainable model linking the
# tweet inputs to the predictions
model = Model(inputs=[tweet_a, tweet_b], outputs=predictions)
model.compile(optimizer='rmsprop',
loss='binary_crossentropy',
metrics=['accuracy'])
return model
def build(self):
inputs = Input(shape=self.input_shape, name='encoder_input')
x = Dense(self.intermediate_dim,
activation=self.activation_fct)(inputs)
z_mean = Dense(self.latent_dim, name='z_mean')(x)
z_log_var = Dense(self.latent_dim, name='z_log_var')(x)
# use reparameterization trick to push the sampling out as input
# note that "output_shape" isn't necessary with the TensorFlow backend
z = Lambda(sampling, output_shape=(self.latent_dim, ),
name='z')([z_mean, z_log_var])
# instantiate encoder model
encoder = Model(inputs, [z_mean, z_log_var, z], name='encoder')
# build decoder model
latent_inputs = Input(shape=(self.latent_dim, ), name='z_sampling')
x = Dense(self.intermediate_dim,
activation=self.activation_fct)(latent_inputs)
outputs = Dense(self.original_dim, activation='sigmoid')(x)
# instantiate decoder model
decoder = Model(latent_inputs, outputs, name='decoder')
# instantiate VAE model
outputs = decoder(encoder(inputs)[2])
vae = Model(inputs, outputs, name='vae_mlp')
# VAE Loss = mse_loss or xent_loss + kl_loss
reconstruction_loss = mse(inputs, outputs)
reconstruction_loss *= self.original_dim
kl_loss = 1 + z_log_var - K.square(z_mean) - K.exp(z_log_var)
kl_loss = K.sum(kl_loss, axis=-1)
kl_loss *= -0.5
vae_loss = K.mean(reconstruction_loss + kl_loss)
vae.add_loss(vae_loss)
vae.compile(optimizer=self.optimizer,
loss=self.loss,
metrics=['accuracy'])
x_train_split, x_valid_split = train_test_split(
self.x_train,
test_size=self.train_test_split,
random_state=self.seed)
vae.fit(x_train_split,
x_train_split,
batch_size=self.batch_size,
epochs=self.epochs,
verbose=self.verbosity,
shuffle=True,
validation_data=(x_valid_split, x_valid_split))
x_train_pred = vae.predict(self.x_train)
train_mse = np.mean(np.power(self.x_train - x_train_pred, 2), axis=1)
self.threshold = np.quantile(train_mse, 0.9)
self.vae = vae
def modelMasking(self, code_layer_type, input_dim, code_dim):
self.code_layer_type = code_layer_type
assert len(code_dim) > 0
if self.code_layer_type == 'lstm':
assert len(input_dim) == 2
input_data = Input(shape=(input_dim[0], input_dim[1]))
mask = Masking(mask_value=0.)(input_data)
if len(code_dim) == 1:
encoded = LSTM(code_dim[0])(mask)
decoded = RepeatVector(input_dim[0])(encoded)
elif len(code_dim) > 1:
encoded = mask
for i, units in enumerate(code_dim):
if i == len(code_dim) - 1:
encoded = LSTM(units)(encoded)
continue
encoded = LSTM(units, return_sequences=True)(encoded)
for i, units in enumerate(reversed(code_dim)):
if i == 1:
decoded = LSTM(units, return_sequences=True)(
RepeatVector(input_dim[0])(encoded))
elif i > 1:
decoded = LSTM(units, return_sequences=True)(decoded)
else:
raise ValueError("The codDim must be over 0.")
decoded = LSTM(input_dim[-1], return_sequences=True)(decoded)
self.model = Model(input_data, decoded)
elif self.code_layer_type == 'cov':
pass
elif self.code_layer_type == 'dense':
assert len(input_dim) == 1
input_data = Input(shape=(input_dim[0], ))
# encoded = input_data
# for i, units in enumerate(codeDim):
# encoded = Dense(units, activation='relu')(encoded)
# decoded = Dense(inputDim[-1], activation='sigmoid')(encoded)
# self.model = Model(input_data, decoded)
encoder = Dense(
code_dim[0],
activation="tanh",
activity_regularizer=regularizers.l1(10e-5))(input_data)
encoder = Dense(int(code_dim[0] / 2), activation="relu")(encoder)
decoder = Dense(int(code_dim[0] / 2), activation='tanh')(encoder)
decoder = Dense(input_dim[0], activation='relu')(decoder)
self.model = Model(input_data, decoder)
def __init__(self, latent_dim=49):
config = ConfigProto()
config.gpu_options.allow_growth = True
session = InteractiveSession(config=config)
# ENCODER
inp = Input((896, 896, 1))
e = Conv2D(32, (10, 10), activation='relu')(inp)
e = MaxPooling2D((10, 10))(e)
e = Conv2D(64, (6, 6), activation='relu')(e)
e = MaxPooling2D((10, 10))(e)
e = Conv2D(64, (3, 3), activation='relu')(e)
l = Flatten()(e)
l = Dense(49, activation='softmax')(l)
# DECODER
d = Reshape((7, 7, 1))(l)
d = Conv2DTranspose(64, (3, 3),
strides=8,
activation='relu',
padding='same')(d)
d = BatchNormalization()(d)
d = Conv2DTranspose(64, (3, 3),
strides=8,
activation='relu',
padding='same')(d)
d = BatchNormalization()(d)
d = Conv2DTranspose(64, (3, 3),
strides=2,
activation='relu',
padding='same')(d)
d = BatchNormalization()(d)
d = Conv2DTranspose(32, (3, 3), activation='relu', padding='same')(d)
decoded = Conv2D(1, (3, 3), activation='sigmoid', padding='same')(d)
self.CAD = tf.keras.Model(inp, decoded)
opt = tf.keras.optimizers.RMSprop(lr=0.0001, decay=1e-6)
self.CAD.compile(loss="binary_crossentropy",
optimizer=opt,
metrics=["accuracy"])
self.Flow = tf.keras.Sequential([
tf.keras.layers.LSTM(32, input_shape=(3, 2),
return_sequences=True),
tf.keras.layers.Dropout(0.4),
tf.keras.layers.Bidirectional(
tf.keras.layers.LSTM(32, return_sequences=True)),
tf.keras.layers.Dropout(0.4),
tf.keras.layers.TimeDistributed(
tf.keras.layers.Dense(10, activation='relu')),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(2, activation='relu')
])
opt = tf.keras.optimizers.RMSprop(lr=0.0001, decay=1e-6)
self.Flow.compile(loss="binary_crossentropy",
optimizer="adam",
metrics=["accuracy"])
print(self.Flow.summary())
print(self.CAD.summary())
def modelDemoStandardConvLSTMInception(input_shape, parameter=None):
# define LSTM
input = Input(shape=input_shape, name='main_input')
I_1 = TimeDistributed(Conv2D(16, (1, 1),
activation='relu',
padding='same',
name='C_1'),
name='I_11')(input)
I_1 = TimeDistributed(Conv2D(16, (5, 5),
activation='relu',
padding='same',
name='C_2'),
name='I_12')(I_1)
I_2 = TimeDistributed(MaxPooling2D((3, 3),
strides=(1, 1),
padding='same',
name='C_3'),
name='I_21')(input)
I_2 = TimeDistributed(Conv2D(16, (1, 1),
activation='relu',
padding='same',
name='C_4'),
name='I_22')(I_2)
concatenate_output = concatenate([I_1, I_2], axis=-1)
# x = TimeDistributed(Flatten())(x)
x = ConvLSTM2D(filters=32,
kernel_size=(3, 3),
padding='same',
return_sequences=False)(concatenate_output)
#x = MaxPooling2D((3, 3), strides=(1, 1), padding='same', name='M_1')(x)
x = (Flatten())(x)
x = RepeatVector(8)(x)
x = LSTM(50, return_sequences=True)(x)
output = TimeDistributed(Dense(8, activation='softmax'),
name='main_output')(x)
#with tensorflow.device('/cpu'):
model = Model(inputs=[input], outputs=[output])
# compile the model with gpu
#parallel_model = multi_gpu_model(model, gpus=2)
#parallel_model.compile(loss={'main_output': 'categorical_crossentropy'},
# loss_weights={'main_output': 1.}, optimizer='adam', metrics=['accuracy'])
#model = multi_gpu(model, gpus=[1, 2])
model.compile(loss={'main_output': 'categorical_crossentropy'},
loss_weights={'main_output': 1.},
optimizer='adam',
metrics=['accuracy'])
return model
def ModelVisualQuestionAnswering():
# First, let's define a vision model using a Sequential model.
# This model will encode an image into a vector.
vision_model = Sequential()
vision_model.add(
Conv2D(64, (3, 3),
activation='relu',
padding='same',
input_shape=(224, 224, 3)))
vision_model.add(Conv2D(64, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Conv2D(128, (3, 3), activation='relu', padding='same'))
vision_model.add(Conv2D(128, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Conv2D(256, (3, 3), activation='relu', padding='same'))
vision_model.add(Conv2D(256, (3, 3), activation='relu'))
vision_model.add(Conv2D(256, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Flatten())
# Now let's get a tensor with the output of our vision model:
image_input = Input(shape=(224, 224, 3))
encoded_image = vision_model(image_input)
# Next, let's define a language model to encode the question into a vector.
# Each question will be at most 100 words long,
# and we will index words as integers from 1 to 9999.
question_input = Input(shape=(100, ), dtype='int32')
embedded_question = Embedding(input_dim=10000,
output_dim=256,
input_length=100)(question_input)
encoded_question = LSTM(256)(embedded_question)
# Let's concatenate the question vector and the image vector:
merged = concatenate([encoded_question, encoded_image])
# And let's train a logistic regression over 1000 words on top:
output = Dense(1000, activation='softmax')(merged)
# This is our final model:
vqa_model = Model(inputs=[image_input, question_input], outputs=output)
return vqa_model
def model(self, code_layer_type, input_dim, code_dim):
self.code_layer_type = code_layer_type
assert len(code_dim) > 0
if self.code_layer_type == 'lstm':
assert len(input_dim) == 2
input_data = Input(shape=(input_dim[0], input_dim[1]))
if len(code_dim) == 1:
encoded = LSTM(code_dim[0])(input_data)
decoded = RepeatVector(input_dim[0])(encoded)
elif len(code_dim) > 1:
encoded = input_data
for i, units in enumerate(code_dim):
if i == len(code_dim) - 1:
encoded = LSTM(units)(encoded)
continue
encoded = LSTM(units, return_sequences=True)(encoded)
for i, units in enumerate(reversed(code_dim)):
if i == 1:
decoded = LSTM(units, return_sequences=True)(
RepeatVector(input_dim[0])(encoded))
elif i > 1:
decoded = LSTM(units, return_sequences=True)(decoded)
else:
raise ValueError("The codDim must be over 0.")
decoded = LSTM(input_dim[-1], return_sequences=True)(decoded)
self.model = Model(input_data, decoded)
elif self.code_layer_type == 'dense':
assert len(input_dim) == 1
input_data = Input(shape=(input_dim[0], ))
encoded = input_data
for i, units in enumerate(code_dim):
encoded = Dense(units, activation='relu')(encoded)
decoded = Dense(input_dim[-1], activation='sigmoid')(encoded)
self.model = Model(input_data, decoded)
elif self.code_layer_type == 'cov':
pass
def modelB(row, col, parameter=None):
# define LSTM
input = Input(shape=(None, row, col, 1), name='main_input')
''' x = TimeDistributed(Conv2D(16, (2, 2)))(input)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = Dropout(0.25)(x)
'''
# tower_1 = TimeDistributed(Conv2D(16, (1, 1), padding='same', activation='relu'))(input)
# tower_1 = TimeDistributed(Conv2D(16, (3, 3), padding='same', activation='relu'))(tower_1)
tower_2 = TimeDistributed(Conv2D(16, (1, 1), padding='same'))(input)
x = BatchNormalization()(tower_2)
x = Activation('relu')(x)
x = Dropout(0.25)(x)
tower_2 = TimeDistributed(Conv2D(16, (5, 5), padding='same'))(x)
x = BatchNormalization()(tower_2)
x = Activation('relu')(x)
tower_2 = Dropout(0.25)(x)
tower_3 = TimeDistributed(
MaxPooling2D((3, 3), strides=(1, 1), padding='same'))(input)
tower_3 = TimeDistributed(Conv2D(16, (1, 1), padding='same'))(tower_3)
x = BatchNormalization()(tower_3)
x = Activation('relu')(x)
tower_3 = Dropout(0.25)(x)
concatenate_output = concatenate([tower_2, tower_3], axis=-1)
x = TimeDistributed(MaxPooling2D(pool_size=(2, 2),
strides=2))(concatenate_output)
x = Dropout(0.25)(x)
x = TimeDistributed(Flatten())(x)
# convLstm = ConvLSTM2D(filters=40, kernel_size=(3, 3),padding='same', return_sequences=False)(x)
lstm_output = LSTM(75)(x)
lstm_output = BatchNormalization()(lstm_output)
# lstm_output = BatchNormalization()(convLstm)
# auxiliary_output = Dense(1, activation='sigmoid', name='aux_output')(lstm_output)
# auxiliary_input = Input(shape=(4,), name='aux_input')
# x = concatenate([lstm_output, auxiliary_input])
x = RepeatVector(4)(lstm_output)
x = LSTM(50, return_sequences=True)(x)
# model.add(Dropout(0.25))
x = BatchNormalization()(x)
output = TimeDistributed(Dense(4, activation='softmax'),
name='main_output')(x)
model = Model(inputs=[input], outputs=[output])
model.compile(loss={'main_output': 'categorical_crossentropy'},
loss_weights={'main_output': 1.},
optimizer='adam',
metrics=['accuracy'])
return model
def SingleOutputCNN(
input_shape,
output_shape,
cnns_per_maxpool=1,
maxpool_layers=1,
dense_layers=1,
dense_units=64,
dropout=0.25,
regularization=False,
global_maxpool=False,
name='',
) -> Model:
function_name = cast(types.FrameType,
inspect.currentframe()).f_code.co_name
model_name = f"{function_name}-{name}" if name else function_name
# model_name = seq([ function_name, name ]).filter(lambda x: x).make_string("-") # remove dependency on pyfunctional - not in Kaggle repo without internet
inputs = Input(shape=input_shape)
x = inputs
for cnn1 in range(0, maxpool_layers):
for cnn2 in range(1, cnns_per_maxpool + 1):
x = Conv2D(32 * cnn2,
kernel_size=(3, 3),
padding='same',
activation='relu')(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = BatchNormalization()(x)
x = Dropout(dropout)(x)
if global_maxpool:
x = GlobalMaxPooling2D()(x)
x = Flatten()(x)
for nn1 in range(0, dense_layers):
if regularization:
x = Dense(dense_units,
activation='relu',
kernel_regularizer=regularizers.l2(0.01),
activity_regularizer=regularizers.l1(0.01))(x)
else:
x = Dense(dense_units, activation='relu')(x)
x = BatchNormalization()(x)
x = Dropout(dropout)(x)
x = Dense(output_shape, activation='softmax')(x)
model = Model(inputs, x, name=model_name)
# plot_model(model, to_file=os.path.join(os.path.dirname(__file__), f"{name}.png"))
return model
def ModelResidual():
# input tensor for a 3-channel 256x256 image
x = Input(shape=(256, 256, 3))
# 3x3 conv with 3 output channels (same as input channels)
y = Conv2D(3, (3, 3), padding='same')(x)
# this returns x + y.
z = add([x, y])
model = Model(inputs=[x], outputs=z)
model.compile(optimizer='rmsprop',
loss='binary_crossentropy',
metrics=['accuracy'])
return model
def ModelSharedVision():
# First, define the vision modules
digit_input = Input(shape=(27, 27, 1))
x = Conv2D(64, (3, 3))(digit_input)
x = Conv2D(64, (3, 3))(x)
x = MaxPooling2D((2, 2))(x)
out = Flatten()(x)
vision_model = Model(digit_input, out)
# Then define the tell-digits-apart model
digit_a = Input(shape=(27, 27, 1))
digit_b = Input(shape=(27, 27, 1))
# The vision model will be shared, weights and all
out_a = vision_model(digit_a)
out_b = vision_model(digit_b)
concatenated = concatenate([out_a, out_b])
out = Dense(1, activation='sigmoid')(concatenated)
classification_model = Model([digit_a, digit_b], out)
return classification_model
def modelA(row, col):
# define LSTM
input = Input(shape=(None, row, col, 1), name='main_input')
x = TimeDistributed(Conv2D(16, (2, 2), activation='relu'))(input)
x = Dropout(0.25)(x)
x = BatchNormalization()(x)
x = TimeDistributed(MaxPooling2D(pool_size=(2, 2), strides=2))(x)
x = Dropout(0.25)(x)
x = TimeDistributed(Flatten())(x)
lstm_output = LSTM(75)(x)
lstm_output = BatchNormalization()(lstm_output)
auxiliary_output = Dense(1, activation='sigmoid',
name='aux_output')(lstm_output)
auxiliary_input = Input(shape=(4, ), name='aux_input')
x = concatenate([lstm_output, auxiliary_input])
x = RepeatVector(8)(x)
x = LSTM(50, return_sequences=True)(x)
# model.add(Dropout(0.25))
x = BatchNormalization()(x)
output = TimeDistributed(Dense(5, activation='softmax'),
name='main_output')(x)
model = Model(inputs=[input, auxiliary_input],
outputs=[output, auxiliary_output])
model.compile(loss={
'main_output': 'categorical_crossentropy',
'aux_output': 'binary_crossentropy'
},
loss_weights={
'main_output': 1.,
'aux_output': 0.2
},
optimizer='adam',
metrics=['accuracy'])
return model
def get_autoencoder_model128(img_width=128, img_height=128): # Built for 128x128
autoencoder = Sequential()
autoencoder.add(Input(shape=(img_width, img_height, 1)))
autoencoder.add(Conv2D(64, (5, 5), activation='relu', padding='same'))
autoencoder.add(MaxPooling2D((2, 2), padding='same'))
autoencoder.add(Conv2D(128, (3, 3), activation='relu', padding='same'))
autoencoder.add(MaxPooling2D((2, 2), padding='same'))
# Decoder
autoencoder.add(Conv2D(128, (3, 3), activation='relu', padding='same'))
autoencoder.add(UpSampling2D((2, 2)))
autoencoder.add(Conv2D(64, (3, 3), activation='relu', padding='same'))
autoencoder.add(UpSampling2D((2, 2)))
autoencoder.add(Conv2D(1, (5, 5), activation='sigmoid', padding='same'))
autoencoder.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
autoencoder.summary()
return autoencoder
def create_model(step: Tensorflow2ModelStep) -> tf.keras.Model:
"""
Create a TensorFlow v2 sequence to sequence (seq2seq) encoder-decoder model.
:param step: The base Neuraxle step for TensorFlow v2 (Tensorflow2ModelStep)
:return: TensorFlow v2 Keras model
"""
# shape: (batch_size, seq_length, input_dim)
encoder_inputs = Input(shape=(None, step.hyperparams['input_dim']),
batch_size=None,
dtype=tf.dtypes.float32,
name='encoder_inputs')
last_encoder_outputs, last_encoders_states = _create_encoder(
step, encoder_inputs)
decoder_outputs = _create_decoder(step, last_encoder_outputs,
last_encoders_states)
return Model(encoder_inputs, decoder_outputs)
def modelDemoStandard(row, col):
# define LSTM
input = Input(shape=(None, row, col, 1), name='main_input')
x = TimeDistributed(Conv2D(16, (2, 2), activation='relu'))(input)
x = TimeDistributed(Flatten())(x)
lstm_output = LSTM(75)(x)
x = RepeatVector(4)(lstm_output)
x = LSTM(50, return_sequences=True)(x)
output = TimeDistributed(Dense(4, activation='softmax'),
name='main_output')(x)
model = Model(inputs=[input], outputs=[output])
model.compile(loss={'main_output': 'categorical_crossentropy'},
loss_weights={'main_output': 1.},
optimizer='adam',
metrics=['accuracy'])
return model
def ModelInception():
input_img = Input(shape=(256, 256, 3))
tower_1 = Conv2D(64, (1, 1), padding='same', activation='relu')(input_img)
tower_1 = Conv2D(64, (3, 3), padding='same', activation='relu')(tower_1)
tower_2 = Conv2D(64, (1, 1), padding='same', activation='relu')(input_img)
tower_2 = Conv2D(64, (5, 5), padding='same', activation='relu')(tower_2)
tower_3 = MaxPooling2D((3, 3), strides=(1, 1), padding='same')(input_img)
tower_3 = Conv2D(64, (1, 1), padding='same', activation='relu')(tower_3)
output = concatenate([tower_1, tower_2, tower_3], axis=1)
model = Model(inputs=[input_img], outputs=output)
model.compile(optimizer='rmsprop',
loss='binary_crossentropy',
metrics=['accuracy'])
return model
def get_model(X, N_class, total_words=86627, EMBEDDING_DIM=100, maxlen=53):
embeddings_index = {}
f = open('glove.6B/glove.6B.100d.txt')
for line in f:
values = line.split()
word = values[0]
coefs = np.asarray(values[1:], dtype='float32')
embeddings_index[word] = coefs
f.close()
embedding_matrix = np.zeros((total_words, EMBEDDING_DIM))
for word, i in X.items():
embedding_vector = embeddings_index.get(word)
if embedding_vector is not None:
embedding_matrix[i] = embedding_vector
inp = Input(shape=(maxlen, ), dtype='int32')
embedding = Embedding(total_words,
EMBEDDING_DIM,
embeddings_initializer=Constant(embedding_matrix),
input_length=maxlen,
trainable=False)(inp)
x = LSTM(300, dropout=0.25, recurrent_dropout=0.25,
return_sequences=True)(embedding)
x = Dropout(0.25)(x)
merged = Attention_COSTUM(maxlen)(x)
merged = Dense(256, activation='relu')(merged)
merged = Dropout(0.25)(merged)
merged = BatchNormalization()(merged)
outp = Dense(N_class, activation='softmax')(merged)
AttentionLSTM = Model(inputs=inp, outputs=outp)
AttentionLSTM.compile(loss='sparse_categorical_crossentropy',
optimizer='adam',
metrics=['acc'])
return AttentionLSTM
def modelDemoStandardConvLSTM(row, col):
# define LSTM
input = Input(shape=(None, row, col, 1), name='main_input')
# x = TimeDistributed(Flatten())(x)
x = ConvLSTM2D(filters=75,
kernel_size=(3, 3),
padding='same',
return_sequences=False)(input)
x = (Flatten())(x)
x = RepeatVector(4)(x)
x = LSTM(50, return_sequences=True)(x)
output = TimeDistributed(Dense(4, activation='softmax'),
name='main_output')(x)
model = Model(inputs=[input], outputs=[output])
model.compile(loss={'main_output': 'categorical_crossentropy'},
loss_weights={'main_output': 1.},
optimizer='adam',
metrics=['accuracy'])
return model
def train_model(x_data, y_data, rain_data, k):
k_fold = KFold(n_splits=k, shuffle=True, random_state=0)
#stratified_k_fold = StratifiedKFold(n_splits=k, shuffle=True, random_state=0)
model_number = 0
#for train_idx, val_idx in tqdm(stratified_k_fold.split(x_data, rain_data)):
for train_idx, val_idx in tqdm(k_fold.split(x_data)):
x_train, y_train = x_data[train_idx], y_data[train_idx]
x_val, y_val = x_data[val_idx], y_data[val_idx]
input_layer = Input(x_train.shape[1:])
output_layer = train2_unet2_model(input_layer, 32)
model = Model(input_layer, output_layer)
callbacks_list = [
# 스케쥴러?
#tf.keras.callbacks.ReduceLROnPlateau(
# monitor='val_loss',
# patience=3,
# factor=0.8
#),
tf.keras.callbacks.ModelCheckpoint(
filepath='./models/model' + str(model_number) + '.h5',
monitor='score',
save_best_only=True,
#save_weights_only=True,
verbose=1
)
]
model.compile(loss='mae', optimizer=RAdamOptimizer(learning_rate=1e-3)
, metrics=[score, maeOverFscore_keras, fscore_keras])
# stratified_k_fold 사용시 batch_size는 최소 128에서 256이 되어야한다.
model.fit(x_train, y_train, epochs=50, batch_size=128, shuffle=True, validation_data=(x_val, y_val),
callbacks=callbacks_list)
model_number += 1
def build(self):
input_img = Input(shape=(28, 28, 1))
cnn = Conv2D(32, (3, 3), activation='relu', padding='same')(input_img)
cnn = MaxPooling2D((2, 2), padding='same')(cnn)
cnn = Conv2D(32, (3, 3), activation='relu', padding='same')(cnn)
cnn = MaxPooling2D((2, 2), padding='same')(cnn)
cnn = Conv2D(32, (3, 3), activation='relu', padding='same')(cnn)
encoded = MaxPooling2D((2, 2), padding='same')(cnn)
cnn = Conv2D(32, (3, 3), activation='relu', padding='same')(encoded)
cnn = UpSampling2D((2, 2))(cnn)
cnn = Conv2D(32, (3, 3), activation='relu', padding='same')(cnn)
cnn = UpSampling2D((2, 2))(cnn)
cnn = Conv2D(32, (3, 3), activation='relu')(cnn)
cnn = UpSampling2D((2, 2))(cnn)
decoded = Conv2D(1, (3, 3), activation='sigmoid', padding='same')(cnn)
cnn_autoencoder = Model(input_img, decoded)
cnn_autoencoder.compile(optimizer='adam', loss='binary_crossentropy')
x_train = self.x_train.reshape(-1, 28, 28, 1)
x_train_split, x_valid_split = train_test_split(x_train, test_size=self.train_test_split,
random_state=self.seed)
cnn_autoencoder.fit(x_train_split, x_train_split,
epochs=self.epochs,
batch_size=self.batch_size,
validation_data=(x_valid_split, x_valid_split),
verbose=self.verbosity)
x_train_pred = cnn_autoencoder.predict(x_train)
mse = np.mean(np.power(x_train - x_train_pred, 2), axis=1)
# Semi-supervised due to given threshold
self.threshold = np.quantile(mse, 0.9)
self.cnn_autoencoder = cnn_autoencoder
def _build_model(self, hl1_dims, hl2_dims, hl3_dims, input_layer_size,
output_layer_size, optimizer, loss):
input_v = Input(shape=(1, input_layer_size))
branch_v = Flatten()(input_v)
branch_v = Dense(hl1_dims, activation='relu')(branch_v)
branch_v = BatchNormalization()(branch_v)
branch_v = Dense(hl2_dims, activation='relu')(branch_v)
branch_v = BatchNormalization()(branch_v)
out_v = Dense(hl3_dims, activation='relu')(branch_v)
out_v = BatchNormalization()(out_v)
out_v = Dense(output_layer_size, activation='linear')(out_v)
model = Model(inputs=input_v, outputs=out_v)
#model = Model(inputs=m_v.inputs, outputs=out_v)
# print(model.summary())
model.compile(
optimizer=optimizer, loss=loss
) # Use Huber Loss Function for DQN based on TensorBoard analysis
return model
def modelStandardB(row, col):
# define LSTM
input_img = Input(shape=(None, row, col, 1), name='input')
x = TimeDistributed(Conv2D(16, (2, 2), activation='relu'))(input_img)
x = Dropout(0.25)(x)
x = BatchNormalization()(x)
x = TimeDistributed(MaxPooling2D(pool_size=(2, 2), strides=2))(x)
x = Dropout(0.25)(x)
x = TimeDistributed(Flatten())(x)
x = LSTM(75)(x)
# model.add(Dropout(0.25))
x = BatchNormalization()(x)
x = RepeatVector(4)(x)
x = LSTM(50, return_sequences=True)(x)
# model.add(Dropout(0.25))
x = BatchNormalization()(x)
output = TimeDistributed(Dense(4, activation='softmax'))(x)
model = Model(input_img, output)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model
# 가중치의 체크 포인트 이름 저장
checkpoint_path = ".saved_weight/cp.ckpt"
checkpoint_dir = os.path.dirname(checkpoint_path)
# 체크포인트 콜백 만들기
cp_callback = tf.keras.callbacks.ModelCheckpoint(checkpoint_path, save_weights_only=True, verbose=1)
# 데이터 파이프라인 만들기
train_dataset = tf.data.Dataset.from_generator(trainGenerator, (tf.float32, tf.float32),
(tf.TensorShape([40, 40, my_col]), tf.TensorShape([40, 40, 1])))
# batch크기가 512로 너무 큰거 같다는 생각이 듬 논문을 참고하면 32에서 64가 이미지 분석에는 적당하다고 봄
train_dataset = train_dataset.batch(32).prefetch(1)
input_layer = Input((40, 40, my_col))
# 처음 시작 뉴론 32개 말고 다른것으로 변경해 볼 것
#output_layer = resnet_model(input_layer)
output_layer = unet_model(input_layer, 32)
model = Model(input_layer, output_layer)
# 저장된 가중치 불러오기
model.load_weights(checkpoint_path)
# ################## compile 메서드로 모형 완성 ###########################
adam = tf.keras.optimizers.Adam(learning_rate=0.0005, beta_1=0.9, beta_2=0.999, epsilon=1e-07,amsgrad=True, name='Adam')
model.compile(loss="mae", optimizer=adam, metrics=[maeOverFscore_keras, fscore_keras, "accuracy"])
# ################## fit 메서드로 트레이닝 #################################
def ModelVideoQuestionAnswering():
# First, let's define a vision model using a Sequential model.
# This model will encode an image into a vector.
vision_model = Sequential()
vision_model.add(
Conv2D(64, (3, 3),
activation='relu',
padding='same',
input_shape=(224, 224, 3)))
vision_model.add(Conv2D(64, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Conv2D(128, (3, 3), activation='relu', padding='same'))
vision_model.add(Conv2D(128, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Conv2D(256, (3, 3), activation='relu', padding='same'))
vision_model.add(Conv2D(256, (3, 3), activation='relu'))
vision_model.add(Conv2D(256, (3, 3), activation='relu'))
vision_model.add(MaxPooling2D((2, 2)))
vision_model.add(Flatten())
# Now let's get a tensor with the output of our vision model:
image_input = Input(shape=(224, 224, 3))
encoded_image = vision_model(image_input)
# Next, let's define a language model to encode the question into a vector.
# Each question will be at most 100 words long,
# and we will index words as integers from 1 to 9999.
question_input = Input(shape=(100, ), dtype='int32')
embedded_question = Embedding(input_dim=10000,
output_dim=256,
input_length=100)(question_input)
encoded_question = LSTM(256)(embedded_question)
# Let's concatenate the question vector and the image vector:
merged = concatenate([encoded_question, encoded_image])
# And let's train a logistic regression over 1000 words on top:
output = Dense(1000, activation='softmax')(merged)
# This is our final model:
# vqa_model = Model(inputs=[image_input, question_input], outputs=output)
video_input = Input(shape=(100, 224, 224, 3))
# This is our video encoded via the previously trained vision_model (weights are reused)
encoded_frame_sequence = TimeDistributed(vision_model)(
video_input) # the output will be a sequence of vectors
encoded_video = LSTM(256)(
encoded_frame_sequence) # the output will be a vector
# This is a model-level representation of the question encoder, reusing the same weights as before:
question_encoder = Model(inputs=question_input, outputs=encoded_question)
# Let's use it to encode the question:
video_question_input = Input(shape=(100, ), dtype='int32')
encoded_video_question = question_encoder(video_question_input)
# And this is our video question answering model:
merged = concatenate([encoded_video, encoded_video_question])
output = Dense(1000, activation='softmax')(merged)
video_qa_model = Model(inputs=[video_input, video_question_input],
outputs=output)
return video_qa_model
def init_model(self, mode="resnet50"):
"""
初始化模型
:param mode: 模型类型
:return: 模型名称和基础模型
"""
if mode == "resnet50v2":
from tensorflow.keras.applications.resnet_v2 import ResNet50V2
model_name = 'rotnet_v3_resnet50v2_{epoch:02d}_{val_acc:.4f}.hdf5'
base_model = ResNet50V2(weights='imagenet',
include_top=False,
input_shape=self.input_shape)
elif mode == "resnet50":
from tensorflow.keras.applications.resnet import ResNet50
model_name = 'rotnet_v3_resnet50_{epoch:02d}_{val_acc:.4f}.hdf5'
base_model = ResNet50(weights='imagenet',
include_top=False,
input_shape=self.input_shape)
elif mode == "mobilenetv2":
from tensorflow.keras.applications.mobilenet_v2 import MobileNetV2
model_name = 'rotnet_v3_mobilenetv2_{epoch:02d}_{val_acc:.4f}.hdf5'
base_model = MobileNetV2(weights='imagenet',
include_top=False,
input_shape=self.input_shape)
elif mode == "densenet121":
from tensorflow.python.keras.applications.densenet import DenseNet121
model_name = 'rotnet_v3_densenet121_{epoch:02d}_{val_acc:.4f}.hdf5'
base_model = DenseNet121(weights='imagenet',
include_top=False,
input_shape=self.input_shape)
else:
raise Exception("[Exception] mode {} 不支持!!".format(mode))
# freeze
for layer in base_model.layers:
layer.trainable = True
x = base_model.output
# if mode == "mobilenetv2":
# x = Dense(128, activation="relu")(x)
x = Flatten()(x)
if self.is_hw_ratio: # 是否使用宽高比
x1 = base_model.output
x1 = Flatten()(x1)
input_ratio = Input(shape=(1, ), name='ratio')
x2 = Dense(1, activation='relu')(input_ratio)
x = concatenate([x1, x2])
final_output = Dense(self.nb_classes,
activation='softmax',
name='fc360')(x)
model = Model(inputs=base_model.input, outputs=final_output)
# model.summary()
# 优化器
if self.nb_classes == 360:
metrics = ["acc", angle_error]
else:
metrics = ["acc"]
model.compile(loss='categorical_crossentropy',
optimizer=SGD(lr=0.0004, momentum=0.9),
metrics=metrics)
if self.model_path:
model.load_weights(self.model_path)
print('[Info] 加载模型的路径: {}'.format(self.model_path))
return model_name, model
from tensorflow.keras.optimizers import SGD
from tensorflow.keras.callbacks import TensorBoard
from tensorflow.keras.models import load_model
import datetime
from sklearn.model_selection import train_test_split
import numpy as np
input_data = np.load('dog_breed_og_flp_trans_data_test.npy')
print(input_data.shape)
input_labels = np.load('dog_breed_og_flp_trans_labels_test.npy')
# X_train, X_val, y_train, y_val = train_test_split(data_in, labels_in, test_size=0.2, random_state=42)
# print("done splitting")
new_input = Input(shape=(224,224,3)) # should be 224,224
# graph changes over epochs
# logdir = "logs/scalars/" + datetime.time().strftime("%Y%m%d-%H%M%S")
# tensorboard_callback = TensorBoard(log_dir=logdir)
model = Sequential()
model.add(ResNet50(weights='imagenet', include_top=False, input_tensor=new_input, pooling='avg'))
model.trainable = False
model.add(Flatten())
model.add(Dense(120, activation='softmax'))
print(model.summary())
model.compile(loss='categorical_crossentropy',
optimizer=SGD(lr=0.001, momentum=0.0, decay=0.0, nesterov=False),
def construct_keras_api_model(embedding_weights):
# input_no_time_no_repeat = Input(shape=max_len, dtype='int32')
# embedded_no_time_no_repeat = Embedding(
# creative_id_window,embedding_size,weights=[embedding_weights],trainable=False
# )(input_no_time_no_repeat)
# ==================================================================================
Input_fix_creative_id = Input(shape=(math.ceil(time_id_max / period_days) *
period_length),
dtype='int32',
name='input_fix_creative_id')
Embedded_fix_creative_id = Embedding(
creative_id_window,
embedding_size,
weights=[embedding_weights],
trainable=False)(Input_fix_creative_id)
# ==================================================================================
# input_no_time_with_repeat = Input(shape=max_len, dtype='int32')
# embedded_no_time_with_repeat = Embedding(creative_id_window,embedding_size,weights=[embedding_weights],trainable=False)(input_no_time_with_repeat)
# ----------------------------------------------------------------------
GM_x = keras.layers.GlobalMaxPooling1D()(Embedded_fix_creative_id)
GM_x = Dropout(0.5)(GM_x)
GM_x = Dense(embedding_size // 2, kernel_regularizer=l2(0.001))(GM_x)
GM_x = BatchNormalization()(GM_x)
GM_x = Activation('relu')(GM_x)
GM_x = Dropout(0.5)(GM_x)
GM_x = Dense(embedding_size // 4, kernel_regularizer=l2(0.001))(GM_x)
GM_x = BatchNormalization()(GM_x)
GM_x = Activation('relu')(GM_x)
GM_x = Dense(1, 'sigmoid')(GM_x)
# ----------------------------------------------------------------------
GA_x = GlobalAveragePooling1D()(Embedded_fix_creative_id)
GA_x = Dropout(0.5)(GA_x)
GA_x = Dense(embedding_size // 2, kernel_regularizer=l2(0.001))(GA_x)
GA_x = BatchNormalization()(GA_x)
GA_x = Activation('relu')(GA_x)
GA_x = Dropout(0.5)(GA_x)
GA_x = Dense(embedding_size // 4, kernel_regularizer=l2(0.001))(GA_x)
GA_x = BatchNormalization()(GA_x)
GA_x = Activation('relu')(GA_x)
GA_x = Dense(1, 'sigmoid')(GA_x)
# ==================================================================================
Conv_creative_id = Conv1D(embedding_size, 15, 5,
activation='relu')(Embedded_fix_creative_id)
# ----------------------------------------------------------------------
Conv_GM_x = MaxPooling1D(7)(Conv_creative_id)
Conv_GM_x = Conv1D(embedding_size, 2, 1, activation='relu')(Conv_GM_x)
Conv_GM_x = GlobalMaxPooling1D()(Conv_GM_x)
Conv_GM_x = Dropout(0.5)(Conv_GM_x)
Conv_GM_x = Dense(embedding_size // 2,
kernel_regularizer=l2(0.001))(Conv_GM_x)
Conv_GM_x = BatchNormalization()(Conv_GM_x)
Conv_GM_x = Activation('relu')(Conv_GM_x)
Conv_GM_x = Dropout(0.5)(Conv_GM_x)
Conv_GM_x = Dense(embedding_size // 4,
kernel_regularizer=l2(0.001))(Conv_GM_x)
Conv_GM_x = BatchNormalization()(Conv_GM_x)
Conv_GM_x = Activation('relu')(Conv_GM_x)
Conv_GM_x = Dense(1, 'sigmoid')(Conv_GM_x)
# ----------------------------------------------------------------------
Conv_GA_x = AveragePooling1D(7)(Conv_creative_id)
Conv_GA_x = Conv1D(embedding_size, 2, 1, activation='relu')(Conv_GA_x)
Conv_GA_x = GlobalAveragePooling1D()(Conv_GA_x)
Conv_GA_x = Dropout(0.5)(Conv_GA_x)
Conv_GA_x = Dense(embedding_size // 2,
kernel_regularizer=l2(0.001))(Conv_GA_x)
Conv_GA_x = BatchNormalization()(Conv_GA_x)
Conv_GA_x = Activation('relu')(Conv_GA_x)
Conv_GA_x = Dropout(0.5)(Conv_GA_x)
Conv_GA_x = Dense(embedding_size // 4,
kernel_regularizer=l2(0.001))(Conv_GA_x)
Conv_GA_x = BatchNormalization()(Conv_GA_x)
Conv_GA_x = Activation('relu')(Conv_GA_x)
Conv_GA_x = Dense(1, 'sigmoid')(Conv_GA_x)
# ----------------------------------------------------------------------
LSTM_x = Conv1D(embedding_size, 14, 7, activation='relu')(Conv_creative_id)
LSTM_x = LSTM(embedding_size, return_sequences=True)(LSTM_x)
LSTM_x = LSTM(embedding_size, return_sequences=True)(LSTM_x)
LSTM_x = LSTM(embedding_size)(LSTM_x)
LSTM_x = Dropout(0.5)(LSTM_x)
LSTM_x = Dense(embedding_size // 2, kernel_regularizer=l2(0.001))(LSTM_x)
LSTM_x = BatchNormalization()(LSTM_x)
LSTM_x = Activation('relu')(LSTM_x)
LSTM_x = Dropout(0.5)(LSTM_x)
LSTM_x = Dense(embedding_size // 4, kernel_regularizer=l2(0.001))(LSTM_x)
LSTM_x = BatchNormalization()(LSTM_x)
LSTM_x = Activation('relu')(LSTM_x)
LSTM_x = Dense(1, 'sigmoid')(LSTM_x)
# ----------------------------------------------------------------------
concatenated = concatenate([
GM_x,
GA_x,
Conv_GM_x,
Conv_GA_x,
LSTM_x,
],
axis=-1)
output_tensor = Dense(1, 'sigmoid')(concatenated)
keras_api_model = Model(
[
# input_no_time_no_repeat,
Input_fix_creative_id,
# input_no_time_with_repeat,
],
output_tensor)
keras_api_model.summary()
plot_model(keras_api_model, to_file='model/keras_api_word2vec_model.png')
print('-' * 5 + ' ' * 3 + "编译模型" + ' ' * 3 + '-' * 5)
keras_api_model.compile(optimizer=optimizers.RMSprop(lr=RMSProp_lr),
loss=losses.binary_crossentropy,
metrics=[metrics.binary_accuracy])
return keras_api_model
def build_model(self, model_name, query_dim, terms_dim, output_dim,
word_embedding):
self.model_name = model_name
query_input = Input(shape=(query_dim, ), name='query_input')
terms_input = Input(shape=(terms_dim, ), name='terms_input')
if model_name == 'lstm':
embedding_feature_block = Sequential(layers=[
Embedding(word_embedding.vocabulary_size,
word_embedding.dimensions,
weights=[word_embedding.embedding_matrix],
trainable=True,
mask_zero=False),
BatchNormalization(),
LSTM(64, return_sequences=True)
])
elif model_name == 'bilstm':
embedding_feature_block = Sequential(layers=[
Embedding(word_embedding.vocabulary_size,
word_embedding.dimensions,
weights=[word_embedding.embedding_matrix],
trainable=True,
mask_zero=False),
BatchNormalization(),
Bidirectional(LSTM(64, return_sequences=True))
])
else: # default cnn
embedding_feature_block = Sequential(layers=[
Embedding(word_embedding.vocabulary_size,
word_embedding.dimensions,
weights=[word_embedding.embedding_matrix],
trainable=True,
mask_zero=False),
BatchNormalization(),
Conv1D(filters=64, kernel_size=3, strides=1),
MaxPooling1D(pool_size=3)
])
# Features
query_feature = embedding_feature_block(query_input)
terms_feature = embedding_feature_block(terms_input)
# Query-Terms alignment
attention = Dot(axes=-1)([query_feature, terms_feature])
softmax_attention = Lambda(lambda x: softmax(x, axis=1),
output_shape=unchanged_shape)(attention)
terms_aligned = Dot(axes=1)([softmax_attention, terms_feature])
# Aligned features
if model_name == 'lstm':
flatten_layer = LSTM(128, return_sequences=False)(terms_aligned)
elif model_name == 'bilstm':
flatten_layer = Bidirectional(LSTM(
128, return_sequences=False))(terms_aligned)
else: # default cnn
merged_cnn = Conv1D(filters=128, kernel_size=3,
strides=1)(terms_aligned)
merged_cnn = MaxPooling1D(pool_size=3)(merged_cnn)
flatten_layer = Flatten()(merged_cnn)
# Output
dense = BatchNormalization()(flatten_layer)
dense = Dense(64, activation='sigmoid')(dense)
out = Dense(output_dim, activation='linear')(dense)
self.model = Model(inputs=[query_input, terms_input], outputs=out)
self.model.compile(optimizer='adam', loss=losses.mean_squared_error)
self.model.summary()
df = StandardScaler().fit_transform(X=df)
kfold = KFold(n_splits=5, shuffle=True)
rmseList = []
rrseList = []
def rmse(y_true, y_pred):
return K.sqrt(K.mean(K.square(y_pred - y_true)))
opt = SGD(lr=0.001, momentum=0.6, nesterov=False)
print("begin training")
for i, (train, test) in enumerate(kfold.split(X)):
model_in = Input(shape=(len(unique_user, )))
x = Dense(10, activation='relu')(model_in)
model_out1 = Dense(10, activation='relu')(model_in)
model_out2 = Dense(10, activation='relu')(model_in)
model_out = Dense(len(unique_movies), activation='sigmoid')(model_out2)
model = Model(inputs=model_in, outputs=model_out)
model.compile(loss='mae', optimizer=opt, metrics=['mae', 'acc'])
history = model.fit(X[train],
Y[train],
validation_data=(X[test], Y[test]),
epochs=400,
batch_size=10,
verbose=0)
scores = model.evaluate(X[test], Y[test])
rmseList.append(scores[1])
