def build_representation_network(input_shape, filter_size1=3, filter_size2=6, conv_strides=1, avg_pool_strides=2):
input = Input(input_shape)
c1 = Conv2D(filters=filter_size1, kernel_size=3, strides=conv_strides, padding='same', activation='relu',
input_shape=input_shape)(input)
r1 = residual(filter_size1, filter_size1, c1)
r2 = residual(filter_size1, filter_size1, r1)
c2 = Conv2D(filters=filter_size2, kernel_size=3, strides=conv_strides, padding='same', activation='relu',
input_shape=(input_shape[0]/conv_strides, input_shape[0]/conv_strides, 3))(r2)
r3 = residual(filter_size2, filter_size2, c2)
r4 = residual(filter_size2, filter_size2, r3)
r5 = residual(filter_size2, filter_size2, r4)
a1 = AveragePooling2D(strides=avg_pool_strides)(r5)
r6 = residual(filter_size2, filter_size2, a1)
r7 = residual(filter_size2, filter_size2, r6)
r8 = residual(filter_size2, filter_size2, r7)
a2 = AveragePooling2D(strides=avg_pool_strides)(r8)
model = Model(inputs=input, outputs=a2)
return model
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 conv_block(feat_maps_out, prev):
prev = BatchNormalization(axis=-1)(prev) # Specifying the axis and mode allows for later merging
prev = Activation('relu')(prev)
prev = Conv2D(filters=feat_maps_out, kernel_size=3, padding='same')(prev)
prev = BatchNormalization(axis=-1)(prev) # Specifying the axis and mode allows for later merging
prev = Activation('relu')(prev)
prev = Conv2D(filters=feat_maps_out, kernel_size=3, padding='same')(prev)
return prev
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 build_model():
model = keras.Sequential()
model.add(
Conv2D(64, kernel_size=3, activation='relu', input_shape=(28, 28, 1)))
model.add(Conv2D(32, kernel_size=3, activation='relu'))
model.add(Flatten())
model.add(Dense(10, activation='softmax'))
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model
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 build_reward_network(shape, filter_size1=3, filter_size2=1):
input = Input(shape)
c1 = Conv2D(filters=filter_size1, kernel_size=(3, 3), strides=2, padding='same', activation='relu',
input_shape=shape)(input)
a1 = AveragePooling2D(strides=2)(c1)
c2 = Conv2D(filters=filter_size2, kernel_size=(3, 3), strides=1, padding='same', activation='relu',
input_shape=shape)(a1)
a2 = AveragePooling2D(strides=2)(c2)
f1 = Flatten()(a2)
model = Model(inputs=input, outputs=f1)
return model
def modelStandard(input_shape, parameter=None):
# define LSTM
model = Sequential()
model.add(
TimeDistributed(Conv2D(16, (2, 2), activation='relu'),
input_shape=input_shape))
model.add(Dropout(parameter['dropout']))
model.add(BatchNormalization())
model.add(TimeDistributed(MaxPooling2D(pool_size=(2, 2), strides=2)))
model.add(Dropout(parameter['dropout']))
model.add(TimeDistributed(Flatten()))
model.add(LSTM(parameter['cell1']))
# model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(RepeatVector(8))
model.add(LSTM(parameter['cell2'], return_sequences=True))
# model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(TimeDistributed(Dense(5, activation='softmax')))
# Replicates `model` on 8 GPUs.
# This assumes that your machine has 8 available GPUs.
#parallel_model = multi_gpu_model(model, gpus=2)
#parallel_model.compile(loss='categorical_crossentropy',
# optimizer='adam', metrics=['accuracy'])
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model
def modelC(row, col):
# define LSTM
model = Sequential()
model.add(
TimeDistributed(Conv2D(16, (2, 2), activation='relu'),
input_shape=(None, row, col, 1)))
model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(TimeDistributed(MaxPooling2D(pool_size=(2, 2))))
model.add(Dropout(0.25))
model.add(TimeDistributed(Flatten()))
model.add(LSTM(75))
# model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(RepeatVector(4))
model.add(LSTM(50, return_sequences=True))
# model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(TimeDistributed(Dense(4, activation='softmax')))
# Replicates `model` on 8 GPUs.
# This assumes that your machine has 8 available GPUs.
# parallel_model = multi_gpu_model(model, gpus=[2])
# parallel_model.compile(loss='categorical_crossentropy',
# optimizer='adam', metrics=['accuracy'])
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model
def build_dynamic_network(shape, filter_size=6, conv_strides=1):
input = Input(shape)
c1 = Conv2D(filters=filter_size, kernel_size=(3, 3), strides=conv_strides, padding='same', activation='relu',
input_shape=shape)(input)
r1 = residual(filter_size, filter_size, c1)
model = Model(inputs=input, outputs=r1)
return model
def __init__(self):
super(Z2Model, self).__init__()
# self.gcnn1 = tf.keras.layers.Conv2D(filters=20, kernel_size=(3, 3), activation='relu')
# self.gcnn2 = tf.keras.layers.Conv2D(filters=20, kernel_size=(3, 3), activation='relu')
# self.gcnn3 = tf.keras.layers.Conv2D(filters=20, kernel_size=(3, 3), activation='relu')
# self.gcnn4 = tf.keras.layers.Conv2D(filters=20, kernel_size=(3, 3), activation='relu')
# self.gcnn5 = tf.keras.layers.Conv2D(filters=20, kernel_size=(3, 3), activation='relu')
# self.gcnn6 = tf.keras.layers.Conv2D(filters=20, kernel_size=(3, 3), activation='relu')
# self.gcnn7 = tf.keras.layers.Conv2D(filters=20, kernel_size=(4, 4), activation='relu')
self.gcnn1 = ConvBatchLayer(
conv=Conv2D(filters=20, kernel_size=(3, 3), activation='relu'))
self.gcnn2 = ConvBatchLayer(
conv=Conv2D(filters=20, kernel_size=(3, 3), activation='relu'))
self.gcnn3 = ConvBatchLayer(
conv=Conv2D(filters=20, kernel_size=(3, 3), activation='relu'))
self.gcnn4 = ConvBatchLayer(
conv=Conv2D(filters=20, kernel_size=(3, 3), activation='relu'))
self.gcnn5 = ConvBatchLayer(
conv=Conv2D(filters=20, kernel_size=(3, 3), activation='relu'))
self.gcnn6 = ConvBatchLayer(
conv=Conv2D(filters=20, kernel_size=(3, 3), activation='relu'))
self.gcnn7 = ConvBatchLayer(conv=Conv2D(filters=9, kernel_size=(3, 3)))
self.flatten = Flatten()
self.dense = Dense(9)
def build_policy_network(shape, action_size, regularizer):
policy_input = Input(shape)
c1 = Conv2D(filters=1, kernel_size=1, padding='same', activation='linear',
kernel_regularizer=regularizer)(policy_input)
b1 = BatchNormalization(axis=-1)(c1)
l1 = LeakyReLU()(b1)
f1 = Flatten()(l1)
d1 = Dense(action_size, use_bias=False, activation='sigmoid', kernel_regularizer=regularizer)(f1)
policy_model = Model(inputs=policy_input, outputs=d1)
return policy_model
def build_value_network(shape, value_support_size):
value_input = Input(shape)
c1 = Conv2D(filters=1, kernel_size=1, padding='same', activation='linear')(value_input)
b1 = BatchNormalization(axis=-1)(c1)
l1 = LeakyReLU()(b1)
f1 = Flatten()(l1)
d2 = Dense(20, use_bias=False, activation='linear')(f1)
l2 = LeakyReLU()(d2)
d2 = Dense(value_support_size, use_bias=False, activation='tanh')(l2)
value_model = Model(inputs=value_input, outputs=d2)
return value_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 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 build_model():
model = Sequential()
model.add(Conv2D(filters=16, kernel_size=2, input_shape=(64, 64, 1), activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.2))
model.add(Conv2D(filters=32, kernel_size=2, activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.2))
model.add(Conv2D(filters=64, kernel_size=2, activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.2))
model.add(Conv2D(filters=128, kernel_size=2, activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.2))
model.add(GlobalAveragePooling2D())
model.add(Dense(9, activation='softmax'))
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 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 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 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 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 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
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 get_model(self):
model = Sequential()
model.add(Conv2D(32, kernel_size=(2, 2), activation='relu',
input_shape=(self.feature_dim_1, self.feature_dim_2, self.channel)))
model.add(Conv2D(64, kernel_size=(2, 2), activation='relu'))
model.add(Conv2D(128, kernel_size=(2, 2), activation='relu'))
model.add(MaxPool2D(pool_size=(1, 1)))
model.add(Dropout(0.5))
model.add(Conv2D(128, kernel_size=(2, 2), activation='relu'))
model.add(Conv2D(256, kernel_size=(2, 2), activation='relu'))
model.add(MaxPool2D(pool_size=(1, 1)))
model.add(Dropout(0.5))
model.add(Conv2D(128, kernel_size=(2, 2), activation='relu'))
model.add(Conv2D(256, kernel_size=(4, 4), activation='relu'))
model.add(MaxPool2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dropout(0.5))
model.add(Dense(256, kernel_regularizer=regularizers.l2(0.2), activation='relu'))
model.add(Dense(32, kernel_regularizer=regularizers.l2(0.2), activation='relu'))
model.add(Dense(self.num_classes, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='RMSProp', metrics=['accuracy'])
return model
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
plot_images(augmented_images)
val_image_gen = ImageDataGenerator(rescale=1. / 255)
val_data_gen = val_image_gen.flow_from_directory(batch_size=batch_size,
directory=val_dir,
target_size=(IMG_SHAPE,
IMG_SHAPE),
class_mode='sparse')
print("Configuring model...")
model = Sequential()
model.add(
Conv2D(16,
3,
padding='same',
activation='relu',
input_shape=(IMG_SHAPE, IMG_SHAPE, 3)))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(32, 3, padding='same', activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(64, 3, padding='same', activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dropout(0.2))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.2))
def skip_block(feat_maps_in, feat_maps_out, prev):
if feat_maps_in != feat_maps_out:
# This adds in a 1x1 convolution on shortcuts that map between an uneven amount of channels
prev = Conv2D(filters=feat_maps_out, kernel_size=1, padding='same')(prev)
return prev
print(f'X_train: {X_train.shape}, Y_train: {Y_train.shape}')
print(f'X_test: {X_test.shape}, Y_test: {Y_test.shape}')
X_train = X_train.reshape(*X_train.shape, 1).astype('float16') / 255
X_test = X_test.reshape(*X_test.shape, 1).astype('float16') / 255
Y_train = to_categorical(Y_train, 10, dtype='float16')
Y_test = to_categorical(Y_test, 10, dtype='float16')
print(f'X_train: {X_train.shape}, Y_train: {Y_train.shape}')
print(f'X_test: {X_test.shape}, Y_test: {Y_test.shape}')
model = Sequential()
model.add(
Conv2D(filters=32,
kernel_size=(3, 3),
activation='relu',
input_shape=(28, 28, 1)))
model.add(Conv2D(filters=64, kernel_size=(3, 3), activation='relu'))
model.add(MaxPool2D(pool_size=2))
model.add(Dropout(rate=0.25))
model.add(Flatten())
model.add(Dense(units=128, activation='relu'))
model.add(Dropout(rate=0.5))
model.add(Dense(units=10, activation='softmax'))
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
early_stop = EarlyStopping(monitor='val_loss', patience=10, verbose=0)
from tensorflow import keras
from tensorflow_core.python.keras import Sequential
from tensorflow_core.python.keras.layers.core import Dense, Dropout, Flatten
from tensorflow_core.python.keras.layers import Conv2D, MaxPooling2D
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)
x_test = x_test.reshape(x_test.shape[0], 28, 28, 1)
model = tf.keras.models.Sequential()
#-------------MOD------------------------
model.add(
Conv2D(16, kernel_size=(3, 3), activation='relu', input_shape=(28, 28, 1)))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dense(10, activation='softmax'))
#-------------------------------------
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
model.fit(x_train, y_train, epochs=5)
model.evaluate(x_test, y_test, verbose=2)
class_names = [
'T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat', 'Sandal', 'Shirt',
'Sneaker', 'Bag', 'Ankle boot'
]
# Create CNN Model
NUM_EPOCHS = 25
LR = 0.001
BATCH_SIZE = 32
model = tf.keras.Sequential()
model.add(
Conv2D(28, (3, 3),
strides=2,
padding='same',
activation='relu',
input_shape=(28, 28, 1)))
model.add(BatchNormalization())
model.add(MaxPooling2D((2, 2)))
model.add(Dropout(0.25))
model.add(Conv2D(64, (3, 3), padding='same', activation='relu'))
model.add(BatchNormalization())
model.add(MaxPooling2D((2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(512, activation='relu'))
model.add(Dense(256, activation='relu'))
model.add(Dense(64, activation='relu'))
