def __disc_img_first_part(self):
model = Sequential()
#model.add(Reshape((self.width*self.height*self.channels,), input_shape=(self.height, self.width, self.channels)))
model.add(InputLayer(input_shape=(self.width, self.height, self.channels)))
model.add(Conv2D(filters=32, kernel_size=3, padding='same'))
model.add(BatchNormalization(momentum=0.9))
model.add(Activation('relu'))
model.add(Conv2D(filters=64, kernel_size=3, padding='same'))
model.add(BatchNormalization(momentum=0.9))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Conv2D(filters=128, kernel_size=3, padding='same'))
model.add(BatchNormalization(momentum=0.9))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Flatten())
return model
def create_cnn_model(img_size):
input_num_units = (img_size, img_size, 3)
pool_size = (2, 2)
hidden_num_units = 500
output_num_units = 3
model = Sequential([
InputLayer(input_shape=input_num_units),
Conv2D(25, (5, 5), activation='relu'),
MaxPooling2D(pool_size=pool_size),
Conv2D(25, (5, 5), activation='relu'),
MaxPooling2D(pool_size=pool_size),
Conv2D(25, (4, 4), activation='relu'),
Flatten(),
Dense(units=hidden_num_units, activation='relu'),
Dense(units=output_num_units,
input_dim=hidden_num_units,
activation='softmax'),
])
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
return model
def net(self):
model = Sequential()
model.add(InputLayer(input_shape=(None, None, 1)))
model.add(Conv2D(64, (3, 3), activation='relu', padding='same'))
model.add(
Conv2D(64, (3, 3), activation='relu', padding='same', strides=2))
model.add(Conv2D(128, (3, 3), activation='relu', padding='same'))
model.add(
Conv2D(128, (3, 3), activation='relu', padding='same', strides=2))
model.add(Conv2D(256, (3, 3), activation='relu', padding='same'))
model.add(
Conv2D(256, (3, 3), activation='relu', padding='same', strides=2))
model.add(Conv2D(512, (3, 3), activation='relu', padding='same'))
model.add(Conv2D(256, (3, 3), activation='relu', padding='same'))
model.add(Conv2D(128, (3, 3), activation='relu', padding='same'))
model.add(UpSampling2D((2, 2)))
model.add(Conv2D(64, (3, 3), activation='relu', padding='same'))
model.add(UpSampling2D((2, 2)))
model.add(Conv2D(32, (3, 3), activation='relu', padding='same'))
model.add(Conv2D(2, (3, 3), activation='tanh', padding='same'))
model.add(UpSampling2D((2, 2)))
model.compile(optimizer='rmsprop', loss='mse')
return model
def merge_with_taken(model, x_without_first, rc0 = 10):
""" Create a new model taking rc0-shaped inputs which are then merged with x_without_first from @see split_x_rc0 """
# input with size just d
input_tensor = InputLayer(input_shape = (rc0,))
# side of the original image
n = int(model.inputs[0].shape[1])
# creating a new model
model_upscale = Sequential()
model_upscale.add(input_tensor)
# adding all x but first red column
model_upscale.add(Lambda(lambda y : tf.pad(tf.reshape(y, shape = (-1, rc0, 1, 1)),
paddings = ([0, 0], [0, n - rc0], [0, n - 1], [0, 2])) + x_without_first))
# adding the rest of VGG15
model_upscale.add(model)
# sanity check: when fed x_taken, will reconstruct the input correctly
#assert np.all(K.function([model_upscale.input], [model_upscale.layers[0].output])([x_taken])[0] == x_orig), "Input is handled in a wrong way"
return model_upscale
def stack1(data, target_shape, l2=1e-3, shared=True):
model = Sequential()
model.add(
InputLayer(name='numeric_columns__',
input_shape=[len(data.numeric_columns)]))
model.add(Reshape([-1, 6]))
model.add(Permute((2, 1)))
if shared:
model.add(
Conv1D(1,
1,
kernel_regularizer=regularizers.l2(l2),
kernel_initializer='zero',
activation='sigmoid'))
else:
model.add(
LocallyConnected1D(1,
1,
kernel_regularizer=regularizers.l2(l2),
kernel_initializer='zero',
activation='sigmoid'))
model.add(Reshape([6]))
return model
def get_model():
return Sequential([
# encoder starts here
InputLayer(input_shape=(200, 200, 3)),
Conv2D(16, (4, 4), activation='relu', padding='same'),
MaxPooling2D((2, 2), padding='same'),
Conv2D(8, (3, 3), activation='relu', padding='same'),
MaxPooling2D((2, 2), padding='same'),
Conv2D(8, (3, 3), activation='relu', padding='same'),
MaxPooling2D((2, 2), padding='same'),
Flatten(),
Dense(1000, activation='relu'),
# bottleneck
Dense(5000, activation='relu'),
Reshape((25, 25, 8)),
Conv2D(8, (4, 4), activation='relu', padding='same'),
UpSampling2D((2, 2)),
Conv2D(8, (3, 3), activation='relu', padding='same'),
UpSampling2D((2, 2)),
Conv2D(16, (3, 3), activation='relu', padding='same'),
UpSampling2D((2, 2)),
Conv2D(3, (3, 3), activation='relu', padding='same')
])
def train(self, training_data):
self.n = len(max(training_data, key=len))
self.N = len(training_data)
input_size = np.zeros((self.N, self.n), dtype='int32')
output_size = np.zeros((self.N, self.n), dtype='int32')
iter = 0
for tagged_sentence in training_data:
train_sentences_num, train_tags_num = [], []
for word, tag in tagged_sentence:
try:
train_sentences_num.append(w2i[word])
except KeyError:
train_sentences_num.append(w2i[b'UNW123'])
train_tags_num.append(t2i[tag])
train_sentences_num_paded = sequence.pad_sequences(
[train_sentences_num], maxlen=self.n)
train_tags_num_padded = sequence.pad_sequences([train_tags_num],
maxlen=self.n)
input_size[iter, :] = train_sentences_num_paded
output_size[iter, :] = train_tags_num_padded
iter = iter + 1
output_size = keras.utils.to_categorical(output_size,
num_classes=len(t2i))
self.model.add(InputLayer(input_shape=(self.n, )))
self.model.add(Embedding(len(w2i), 100, mask_zero=True))
self.model.add(Bidirectional(LSTM(50, return_sequences=True)))
self.model.add(Dense(len(t2i), activation='softmax'))
self.model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
print(self.model.summary())
self.model.fit(input_size, output_size, batch_size=32, epochs=4)
pass
def add_noise_layer(self, layer_name):
# previous_layer_name="maeno layer no namae"
# before_noise_layer = self.model.layers[layer_name]
print("before modeling: self.model.get_layer(block1_conv2).output = ",
self.model.get_layer("block1_conv2").output)
# model_layers = self.model.layers[1:]
model_copy = Sequential()
model_copy.add(InputLayer(input_shape=self.model.input_shape[1:]))
model_layers = []
for layer in self.model.layers[1:]:
try:
model_layers.extend(layer.layers)
except:
model_layers.append(layer)
print("model_layers")
# for i, layer in enumerate(self.model.layers[1:]):
for i, layer in enumerate(model_layers):
layer_config = layer.get_config()
# if type(layer_config) == list:
# print(len(layer_config))
# print(layer.layers)
# print("type(layer) =", type(layer))
# print("layer_config =", layer_config)
# model_copy.add(type(layer)(**layer_config))
# model_copy.add(type(layer)(**{'filters':layer_config["filters"], 'kernel_size':layer_config["kernel_size"]}))
# model_copy.add(type(layer)(filters=layer_config["filters"], kernel_size=layer_config["kernel_size"]))
model_copy.add(
type(layer)(**layer_config, weights=layer.get_weights()))
if layer.name == layer_name:
model_copy.add(
noise_layer(noiselevel=self.noiselevel, name="noise"))
if int(self.nb_gpus) > 1:
self.model_multiple_gpu = multi_gpu_model(model_copy,
gpus=self.nb_gpus)
else:
self.model_multiple_gpu = model_copy
def construct_hrs_model(dataset,
model_indicator,
blocks_definition,
load_weights=True):
# get structure from model_indicator
structure = [int(ss[:-1]) for ss in model_indicator.split('[')[1:]]
nb_block = len(structure)
# assert nb blocks
assert len(structure) == len(
blocks_definition
), 'arg structure and block_definition need to have the same length'
# assert weights exist
weights_dir = './Model/%s_Models/%s' % (dataset, model_indicator)
assert os.path.exists(weights_dir), '%s does not exist' % weights_dir
# input
img_rows, img_cols, img_channels = get_dimensions(dataset)
model_input = InputLayer(input_shape=(img_rows, img_cols, img_channels))
save_dir = './Model/%s_models/' % dataset + model_indicator + '/'
# loop over block
block_input = model_input.output
for i in range(nb_block):
weight_dir = save_dir + '%d_' % i + '%d'
block_output = construct_switching_block(
input=block_input,
nb_channels=structure[i],
channel_definition=blocks_definition[i],
weights=weight_dir)
block_input = block_output
# construct Model object
model = Model(input=model_input.input, output=block_output)
return model
def q_learning_keras(env, num_episodes=2000):
# create the keras model
model = Sequential()
model.add(InputLayer(batch_input_shape=(1, 4)))
model.add(Dense(30, activation='sigmoid'))
model.add(Dense(3, activation='linear'))
model.compile(loss='mse', optimizer='adam', metrics=['mae'])
# now execute the q learning
y = 0.95
eps = 0.5
decay_factor = 0.999
r_avg_list = []
for i in range(num_episodes):
s = env.reset()
eps *= decay_factor
if i % 100 == 0:
print("Episode {} of {}".format(i + 1, num_episodes))
done = False
r_sum = 0
while not done:
if np.random.random() < eps:
a = np.random.randint(0, 3)
else:
a = np.argmax(model.predict(np.array([s])))
new_s, r, done = env.DO(a)
target = r + y * np.amax(model.predict(np.array([new_s])))
target_vec = model.predict(np.array([s]))[0]
target_vec[a] = target
model.fit(np.array([s]), target_vec.reshape(-1, 3), epochs=1, verbose=0)
s = new_s
r_sum += r
r_avg_list.append(r_sum)
plt.plot(r_avg_list)
plt.ylabel('Average reward per game')
plt.xlabel('Number of games')
plt.show()
return model
def basic_transferlearning(self, train, train_label, val, val_label,
input_shape, model_name):
vgg_model = self.import_vgg_model(input_shape)
train_features = self.get_bottleneck_features(
vgg_model, self.scale_bottleneck(train))
val_features = self.get_bottleneck_features(vgg_model,
self.scale_bottleneck(val))
input_shape = vgg_model.output_shape[1]
model = Sequential()
model.add(InputLayer(input_shape=(input_shape, )))
model.add(Dense(512, activation='relu', input_dim=input_shape))
model.add(Dropout(0.3))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.3))
if (self.total_classes):
model.add(Dense(self.total_classes, activation="softmax"))
else:
model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy',
optimizer=optimizers.RMSprop(lr=1e-4),
metrics=['accuracy'])
model.summary()
history = model.fit(x=train_features,
y=train_label,
validation_data=(val_features, val_label),
batch_size=self.batch_size,
epochs=100,
verbose=1)
model.save("basic_transferlearning_" + model_name + ".h5")
return history
def createNetwork(self, input_layer=None):
model = Sequential()
model.add(input_layer or InputLayer(Config.input_shape))
# convolutional part
for l in self.conv_layers:
if type(l) is ConvLayer:
model.add(
Conv2D(l.filters, (l.kernel_size, l.kernel_size),
padding='same'))
model.add(Activation(l.activation))
elif type(l) is MaxPoolLayer:
# check if pooling is possible
if model.output_shape[1] >= l.pool_size and model.output_shape[
2] >= l.pool_size:
model.add(
MaxPooling2D(pool_size=(l.pool_size, l.pool_size)))
else:
raise TypeError("unknown type of layer")
# dense part
model.add(Flatten())
for l in self.dense_layers:
model.add(Dense(l.size))
model.add(Activation(l.activation))
if l.dropout > 0:
model.add(Dropout(l.dropout))
# final part
model.add(Dense(self.noutputs))
if Config.task_type == "classification":
model.add(Activation('softmax'))
self.nparams = model.count_params()
return model
def build(self):
model = Sequential()
model.add(InputLayer(input_shape=(INPUT_COUNT, 1), name="in"))
encoder = Sequential(name="encoder")
add_pool_convolution(encoder, 4)
add_pool_convolution(encoder, 4)
add_pool_convolution(encoder, 8)
add_pool_convolution(encoder, 8)
add_pool_convolution(encoder, 32)
add_pool_convolution(encoder, 4)
add_pool_convolution(encoder, 4)
add_pool_convolution(encoder, 8)
add_pool_convolution(encoder, 8)
add_pool_convolution(encoder, 128)
model.add(encoder)
decoder = Sequential(name="decoder")
add_upsampling_convolution(decoder, 128)
add_upsampling_convolution(decoder, 8)
add_upsampling_convolution(decoder, 8)
add_upsampling_convolution(decoder, 4)
add_upsampling_convolution(decoder, 4)
add_upsampling_convolution(decoder, 32)
add_upsampling_convolution(decoder, 8)
add_upsampling_convolution(decoder, 8)
add_upsampling_convolution(decoder, 4)
add_upsampling_convolution(decoder, 4)
add_convolution(decoder, 1)
model.add(decoder)
model.compile(optimizer="adam", loss="mse", metrics=["acc"])
self.model = model
self.build_encoder()
self.build_decoder()
def create_model_1d(cfg):
pool_size = cfg['pool_size'] # size of pooling area for max pooling
kernel_size = cfg['kernel_size'] # convolution kernel size
input_shape = (23, 3 * (2 * cfg['feature_context'] + 1))
# Keras Model
model = Sequential()
model.add(InputLayer(batch_input_shape=(None, input_shape[0], input_shape[1]), name='input'))
model.add(GaussianNoise(stddev=cfg['gaussian_noise']))
for i in range(cfg['num_cnn_layers']):
model.add(Conv1D(filters=cfg['filters'], kernel_size=kernel_size,
padding=cfg['padding'],
kernel_initializer='he_normal',
kernel_regularizer=regularizers.l2(cfg['weight_decay'])))
model.add(Activation(activation=cfg['activation']))
if cfg['batch_normalization']:
model.add(BatchNormalization())
model.add(MaxPooling1D(pool_size=pool_size))
model.add(Dropout(cfg['dropout']))
model.add(Flatten())
for i in range(cfg['num_ff_layers']):
model.add(Dense(cfg['ff_layer_size']))
model.add(Activation(cfg['activation']))
if cfg['batch_normalization']:
model.add(BatchNormalization())
model.add(Dropout(cfg['dropout']))
model.add(Dense(cfg['output_dim']))
model.add(BatchNormalization())
# Optional softmax layer
if cfg['task'] == 'classification':
model.add(Softmax())
return model
def createCDNet(summary=False, output_layers=8):
print("Start Initialzing Neural Network!")
model = Sequential()
model.add(InputLayer(input_shape=(2, 75)))
model.add(LSTM(256, return_sequences=True))
model.add(Dropout(0.25))
model.add(BatchNormalization())
model.add(Dense(256))
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(LSTM(128, return_sequences=True))
model.add(Dropout(0.25))
# model.add(BatchNormalization())
model.add(Dense(128))
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(Dense(64, activation='softmax'))
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(Dense(32, activation='softmax'))
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(Flatten())
model.add(Dense(output_layers, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adadelta', metrics=['accuracy'])
if summary:
print(model.summary())
return model
def create_model(self, input_shape):
model = Sequential()
model.add(InputLayer(input_shape=input_shape))
model.add(
Conv2D(64, (3, 3), activation='relu',
data_format='channels_first'))
model.add(MaxPooling2D(pool_size=(2, 2), data_format='channels_first'))
model.add(BatchNormalization())
model.add(Dropout(0.2))
model.add(
Conv2D(128, (3, 3),
activation='relu',
data_format='channels_first'))
model.add(MaxPooling2D(pool_size=(2, 2), data_format='channels_first'))
model.add(BatchNormalization())
model.add(Dropout(0.2))
model.add(
Conv2D(256, (3, 3),
activation='relu',
data_format='channels_first'))
model.add(MaxPooling2D(pool_size=(2, 2), data_format='channels_first'))
model.add(BatchNormalization())
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(7, activation='linear'))
model.summary()
model.compile(loss='mse',
optimizer=Adam(lr=1e-4),
metrics=['accuracy'])
return model
def Bi_RNN2(word_index, embedding_matrix, embedding_dim=50):
model = Sequential()
model.add(InputLayer(input_shape=(49, )))
model.add(
Embedding(len(word_index) + 1,
embedding_dim,
weights=[embedding_matrix],
input_length=49,
name="embedding",
trainable=True))
model.add(SpatialDropout1D(0.4))
model.add(Bidirectional(CuDNNLSTM(64, return_sequences=True)))
model.add(Bidirectional(CuDNNLSTM(32)))
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['acc'])
# try using different optimizers and different optimizer configs
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['acc'])
return model
def __init__(self, name, state_shape, n_actions, epsilon=0, reuse=False):
"""A simple DQN agent"""
self.n_actions = n_actions
self.name = name
self.filename = '%s.h5' % self.name
with tf.variable_scope(name, reuse=reuse):
self.model = Sequential()
self.model.add(InputLayer(state_shape))
self.model.add(Dense(256, activation='relu'))
# self.model.add(Dense(192, activation='relu'))
self.model.add(Dense(128, activation='relu'))
self.model.add(Dense(64, activation='relu'))
self.model.add(Dense(n_actions, activation='linear'))
# prepare a graph for agent step
self.state_t = tf.placeholder('float32', [
None,
] + list(state_shape))
self.qvalues_t = self.get_symbolic_qvalues(self.state_t)
self.weights = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,
scope=name)
self.epsilon = epsilon
def create_model():
from keras.layers import (ELU, Activation, BatchNormalization, Dense,
Dropout, InputLayer, Flatten, Reshape, LSTM,
TimeDistributed)
from keras.models import Sequential
from keras.utils import plot_model
from PIL import Image
""" ----------------------------- VANILLA ARTIFICIAL NEURAL NETWORK ----------------------------- """
model = Sequential()
model.add(InputLayer(input_shape=(1, )))
model.add(Dense(10, activation='relu'))
model.add(Dense(60, activation='relu'))
model.add(Reshape((10, 6)))
model.add(Dense(10, activation='softmax'))
# model.compile(loss='mse', optimizer='adam', metrics=['mae'])
# model = Sequential()
# model.add(LSTM(100, input_shape=(10,6), return_sequences=True))
# model.add(LSTM(500, return_sequences=True))
# model.add(LSTM(250, return_sequences=True))
# model.add(LSTM(125, return_sequences=True))
# model.add(LSTM(10))
# model.add(Dropout(rate=0.25))
# model.add(Dense(1))
# model.add(Activation('softmax'))
model.compile(loss='mean_squared_error',
optimizer='adam',
metrics=['accuracy'])
plot_model(model, 'model.png', show_layer_names=True, show_shapes=True)
Image.open('model.png').show()
return model
def main(verbose):
(x_train, y_train), _ = mnist.load_data()
x_train = x_train / 255
x_train = x_train[..., None]
y_train = to_categorical(y_train)
model = Sequential([
InputLayer([28, 28, 1]),
Conv2D(32, 5, padding="same", activation="relu"),
MaxPool2D(),
Conv2D(64, 5, padding="same", activation="relu"),
MaxPool2D(),
Flatten(),
Dense(512, activation="relu"),
Dropout(0.5),
Dense(10, activation="softmax")
])
model.compile("adam", "categorical_crossentropy")
# measure
model.fit(x_train,
y_train,
batch_size=32,
epochs=1,
verbose=verbose,
shuffle=False)
start = time.time()
model.fit(x_train,
y_train,
batch_size=32,
epochs=1,
verbose=verbose,
shuffle=False)
print(f"Elapsed time: {time.time()-start}")
def modeloDNN(args,
mode="regression",
num_hidden_dense_layers=2,
hidden_dense_layers_units=20,
length=50,
nfeatures=7):
if mode == "classification":
last_activation = "softmax"
last_units = 3
else:
last_activation = "linear"
last_units = 1
# print(f'modeloCNNBasico... conv_size: {conv_size}')
model = Sequential()
model.add(InputLayer(input_shape=(length, nfeatures)))
for i in range(num_hidden_dense_layers):
model.add(Dense(hidden_dense_layers_units, activation=args.activation))
model.add(BatchNormalization())
model.add(Dense(hidden_dense_layers_units, activation=args.activation))
model.add(Flatten())
model.add(Dense(last_units, activation=last_activation))
return model
def nsfsrcnn(x, d=56, s=12, m=4, scale=3, pos=1):
"""Build an FSRCNN model, but change deconv position.
See https://arxiv.org/abs/1608.00367
"""
model = Sequential()
model.add(InputLayer(input_shape=x.shape[-3:]))
c = x.shape[-1]
f1 = [5, 1] + [3] * pos
n1 = [d, s] + [s] * pos
f2 = [3] * (m - pos - 1) + [1]
n2 = [s] * (m - pos - 1) + [d]
f3 = 9
n3 = c
for ni, fi in zip(n1, f1):
model.add(
Conv2D(ni,
fi,
padding='same',
kernel_initializer='he_normal',
activation='relu'))
model.add(
Conv2DTranspose(s,
3,
strides=scale,
padding='same',
kernel_initializer='he_normal'))
for ni, fi in zip(n2, f2):
model.add(
Conv2D(ni,
fi,
padding='same',
kernel_initializer='he_normal',
activation='relu'))
model.add(Conv2D(n3, f3, padding='same', kernel_initializer='he_normal'))
return model
def fsrcnn(x, d=56, s=12, m=4, scale=3):
"""Build an FSRCNN model.
See https://arxiv.org/abs/1608.00367
"""
model = Sequential()
model.add(InputLayer(input_shape=x.shape[-3:]))
c = x.shape[-1]
f = [5, 1] + [3] * m + [1]
n = [d, s] + [s] * m + [d]
for ni, fi in zip(n, f):
model.add(
Conv2D(ni,
fi,
padding='same',
kernel_initializer='he_normal',
activation='relu'))
model.add(
Conv2DTranspose(c,
9,
strides=scale,
padding='same',
kernel_initializer='he_normal'))
return model
def create_cnn_model(): cnn = Sequential() # Add input layer cnn.add(InputLayer(input_shape=input_shape)) # Normalization cnn.add(BatchNormalization()) # Conv + Maxpooling cnn.add(Conv2D(conv_kernels_1, (kernel_size, kernel_size), padding="same",activation="relu")) cnn.add(MaxPooling2D((pool_size,pool_size))) # Dropout cnn.add(Dropout(drop_prop_1)) cnn.add(Conv2D(conv_kernels_2, (kernel_size, kernel_size))) cnn.add(MaxPooling2D((pool_size,pool_size))) # Dropout cnn.add(Dropout(drop_prop_2)) cnn.add(Conv2D(conv_kernels_2, (kernel_size, kernel_size))) cnn.add(MaxPooling2D((pool_size,pool_size))) # Cannot add more layers as the size is less than kernel_size #cnn.add(Conv2D(conv_kernels_2, (kernel_size, kernel_size))) #cnn.add(MaxPooling2D((pool_size,pool_size))) cnn.add(Flatten()) cnn.add(Dense(dense_size, activation='relu')) cnn.add(Dropout(drop_prop_2)) cnn.add(Dense(dense_size//2, activation='relu')) cnn.add(Dropout(drop_prop_2)) cnn.add(Dense(n_classes, activation='softmax')) cnn.compile(loss='categorical_crossentropy', metrics=['accuracy'],optimizer="Adam") return cnn
def vgg9CNN(input_shape, outclass, sigma='sigmoid'):
model = None
model = Sequential()
model.add(InputLayer(input_shape=input_shape))
for i in [64, 96, 128]:
model.add(Conv2D(i, (3, 3), padding='valid'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Conv2D(i, (3, 3), padding='valid'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(3, 3), strides=(2, 2)))
model.add(Flatten())
for i in range(2):
model.add(Dense(1024, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(outclass))
model.add(Activation(sigma))
return model
def train(train_generator,train_size,input_num,dims_num):
print("Start Train Job! ")
start=time.time()
inputs=InputLayer(input_shape=(input_num,dims_num),batch_size=batch_size)
layer1=Dense(100,activation="relu")
layer2=Dense(20,activation="relu")
flatten=Flatten()
layer3=Dense(2,activation="softmax",name="Output")
optimizer=Adam()
model=Sequential()
model.add(inputs)
model.add(layer1)
model.add(Dropout(0.5))
model.add(layer2)
model.add(Dropout(0.5))
model.add(flatten)
model.add(layer3)
call=TensorBoard(log_dir=log_dir,write_grads=True,histogram_freq=1)
model.compile(optimizer,loss="categorical_crossentropy",metrics=["accuracy"])
model.fit_generator(train_generator,steps_per_epoch=train_size//batch_size,epochs=epochs_num,callbacks=[call])
# model.fit_generator(train_generator, steps_per_epoch=5, epochs=5, callbacks=[call])
model.save(model_dir)
end=time.time()
print("Over train job in %f s"%(end-start))
def build_model():
model = Sequential()
model.add(RandomRotation(factor=0.45))
model.add(RandomFlip(mode='horizontal'))
model.add(InputLayer(input_shape=(224, 224, 1)))
model.add(Conv2D(1, (2, 2), activation='relu', padding='same'))
model.add(Conv2D(32, (2, 2), strides=2, activation='relu', padding='same'))
model.add(Conv2D(64, (2, 2), strides=2, activation='relu', padding='same'))
model.add(Conv2D(128, (2, 2), strides=2, activation='relu',
padding='same'))
model.add(
Conv2DTranspose(128, (2, 2),
strides=2,
activation='relu',
padding='same'))
model.add(
Conv2DTranspose(64, (2, 2),
strides=2,
activation='relu',
padding='same'))
model.add(
Conv2DTranspose(2, (2, 2),
strides=2,
activation='relu',
padding='same'))
return model
def init_model(self):
self.model = Sequential()
self.model.add(InputLayer(input_shape=(self.state_num,
*self.env_size)))
self.model.add(
Convolution2D(16,
4,
4,
border_mode='same',
activation='relu',
subsample=(2, 2)))
self.model.add(
Convolution2D(32,
2,
2,
border_mode='same',
activation='relu',
subsample=(1, 1)))
self.model.add(
Convolution2D(32,
2,
2,
border_mode='same',
activation='relu',
subsample=(1, 1)))
self.model.add(Flatten())
self.model.add(Dense(128, activation='relu'))
self.model.add(Dense(self.n_actions, activation='linear'))
optimizer = RMSprop if not self.use_graves else RMSpropGraves
self.model.compile(loss=loss_func,
optimizer=optimizer(lr=self.learning_rate),
metrics=['accuracy'])
self.target_model = copy.copy(self.model)
def q_learning_with_keras(env, num_episodes=500):
model = Sequential()
model.add(InputLayer(batch_input_shape=(1, 5)))
model.add(Dense(10, activation='sigmoid'))
model.add(Dense(2, activation='linear'))
model.compile(loss='mse', optimizer='adam', metrics=['mae'])
y = 0.95
eps = 0.5
decay_factor = 0.999
r_avg_list = []
for i in range(num_episodes):
s = env.reset()
eps *= decay_factor
if i % 100 == 0:
print("Episode {} of {}".format(i + 1, num_episodes))
done = False
r_sum = 0
while not done:
if np.random.random() < eps:
a = np.random.randint(0, 2)
else:
a = np.argmax(model.predict(np.identity(5)[s:s + 1]))
new_s, r, done, _ = env.step(a)
target = r + y * np.max(
model.predict(np.identity(5)[new_s:new_s + 1]))
target_vec = model.predict(np.identity(5)[s:s + 1])[0]
target_vec[a] = target
model.fit(np.identity(5)[s:s + 1],
target_vec.reshape(-1, 2),
epochs=1,
verbose=0)
s = new_s
r_sum += r
r_avg_list.append(r_sum / 1000)
return model
def build(self):
print("############### BUILDING MODEL ####################")
model = Sequential()
model.add(InputLayer(input_shape=(LAYERS, self.height, self.width)))
if self.width > 5 or self.height > 5:
model.add(Conv2D(32, (3, 3), strides=(1, 1), padding='same'))
model.add(Activation('relu'))
model.add(Flatten())
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dense(ACTIONS))
filename = self.filebase + ".hd5"
if os.path.isfile(filename):
print("loading prior weights")
model.load_weights(filename)
adam = Adam(lr=LEARNING_RATE)
model.compile(loss='mse', optimizer=adam)
self.model = model
print("############### DONE ####################")
