def create_smaller():
# create model
model = Sequential()
model.add(
Dense(30, input_dim=60, kernel_initializer='normal',
activation='relu'))
model.add(Dense(1, kernel_initializer='normal', activation='sigmoid'))
# Compile model
model.complie(loss='binary_crossentropy',
optimizer='adam',
metrics=[accuracy])
return model
class PolicyNetwork:
def __init__(self, input_dim=0, output_dim=0, lr=0.01):
self.input_dim = input_dim
self.lr = lr
#LSTM network
self.model = Sequential()
self.model.add(
LSTM(256,
imput_shape=(1, input_dim),
return_sequenbces=True,
stateful=False,
dropout=0.5))
self.model.add(BatchNormalization())
self.add(LSTM(256, return_squence=True, stateful=False, dropout=0.5))
self.model.add(BatchNormalization())
self.add(LSTM(256, return_squence=False, stateful=False, dropout=0.5))
self.model.add(BatchNormalization())
self.model.add(Dense(output_dim))
self.model.add(Activation('sigmoid'))
self.model.complie(optmizer=sgd(lr=lr), loss='mse')
self.prob = None
def reset(self):
self.prob = None
def predict(self, sample): #Get a lot sample and return neural network
self.prob = self.model.predict(
np.array(sample).reshape(
(1, -1,
self.input_dim)))[0] #Change array for keras input type
return self.prob
def train_on_batch(self, x, y): #policy neural network learning
return self.modele.tain_on_batch(x, y)
def save_model(self, model_path): #Save policy neural network learning
if model_path is not None and self.model is not None:
self.model.save_weights(model_path, overwrite=True)
def load_model(self, model_path): #Load policy neural network learning
if model_path is not None:
self.model.load_weights(model_path)
class model:
def __init__(self):
self.model=None
def cnn_model(self):
self.model = Sequential()
self.model.add(Conv2D(32,(3,3)),strides=1,padding="same")
self.model.add(Activation("relu"))
self.model.add(Conv2D(32,(5,5)),padding="same")
self.model.add(MaxPooling2D(pool_size=(2,2)))
self.model.add(Conv2D(32,(3,3)),padding="same")
seld.model.add(Activation("relu"))
self.model.add(MaxPooling2D(pool_size=(2,2)))
self.model.add(Conv2D(64,(5,5)),padding=(2,2))
self.model.add(Activation("relu"))
self.model.add(MaxPooling2D(pool_size=(2,2)))
self.model.add(Flatten())
self.model.add(Dense(2048))
self.model.add(Activation("relu"))
self.model.add(Dropout(0.5))
self.model.add(Dense(1024))
self.model.add(Activation("relu"))
self.model.add(Dropout(0.5))
self.model.add(num_class)
self.model.add(Activation("softmax"))
self.model.summary()
def train_model(self):
sgd=SGD(lr=0.001,decay=0.000001,momentum=0.9,nesterov=True)
self.model.complie(loss="categorical_crossentropy",optimize=sgd,metrics=["accuracy"])
train_
model.add(MaxPool2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Convolution2D(64, 3, 3, border_mode='same'))
model.add(Activation('relu'))
model.add(Convolution2D(64, 3, 3))
model.add(MaxPool2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten()) # ... (*3) : p.356
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(nb_classes))
model.add(Activation('softmax'))
model.complie(loss = 'binary_crossentropy',
optimizer='rmsprop',
metrics=['accuracy'])
# 모델 트레이닝하기 ... (*4) :.p356
model.fit(X_train, y_train, batch_size=32, nb_epoch=50)
# 모델 평가하기 ... (*5)
score = model.evaluate(X_train, y_train)
print("lose :", score[0])
print("accuracy :", score[1])
# 일단은 에러, 옵션으로 다시 불러보기.. ing ...
model.add(Conv2D(128, (2, 2), padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D(2, 2)))
model.add(Conv2D(256, (2, 2), padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dropout(0.3))
model.add(Flatten())
model.add(Dense(nb_classes, activation='softmax'))
opt = Adam(lr=0.0001) # learning rate 조정
model.complie(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
# 2. 데이터셋 불러오기, 라벨링
from keras.preprocessing.image import ImageDataGenerator
train_data = ImageDataGenerator(rescale=1./255) # 모든 픽셀을 0~1 사이의 숫자로 변환
val_data = ImageDataGenerator(rescale=1./255)
test_data = ImageDataGenerator(rescale=1./255)
train_set = train_data.flow_from_directory('C:/Users/User/Desktop/dataset/train', # train set 저장 경로
target_size=(100,100), # image 크기 재설정
batch_size=1, # 이미지를 몇 개씩 가져와 처리할지 설정
class_mode='categorical') # 다중분류 시 자동으로 라벨링
val_set = test_data.flow_from_directory('C:/Users/User/Desktop/dataset/test', # validation set 저장 경로
from keras.models import Sequential from keras.layers.core import Dense, Activation model = Sequential() model.add(Dense(2,1,init='uniform',activation='linear')) model.complie(loss='mse',optimizer='rmsprop')
from keras.models import Sequential
from keras.layers import Dense
impoer numpy as np
x = np.array([1, 2, 3, 4, 5])
y = np.array([1, 2, 3, 4, 5])
model = Sequential()
model.add(Dense(5, input_dim=1, activation='relu'))
model.add(Dense(3))
model.add(Dense(1))
model.complie(lostt='mse', optimizer='adam')
model.fit(x, y, epochs=100, batch_size=1)
loss, acc = model.evaluate(x, y, batch_size=1)
print('acc : ', acc)
### Schritt 6 Vollvernetztes Hidden Layer hinzufügen
# output_dim = Anzahl der Neuronen im Hidden Layer
# Anzahl der Neuronen im Hidden Layer = keine Daumenregel, aber Zahl zwischen Inout Neuronen und Output-Neuronen
# Anzahl der Neuronen im Hidden Layer = 128
classifier.add(Dense(units = 128, activation = 'relu'))
### Schritt 7 - Outpu Layer hinzufügen
### Anzahl der Output NEuronen nur noch 1 da binäres Problem
### nicht mehr Relu sundern Sigmoid, da wir einen binären Outout haben, Relu geht von 0 bis unendlich, sigmoid geht von 0 bnis 1 = WSK für die jeweilige Klasse
classifier.add(Dense(units = 1, activation = 'sigmoid'))
### Schritt 8 - Kompilieren des CNNs
### Fehlerfunktion = binäre Crossentropy, da binäres Problem
### Adam Optimizer
classifier.complie(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])
### Schritt 9 - Data Augmentation mittels Keras' ImageDataGenerator
# Um Overfitting zu verhindern und die Datenmenge künstlich zu erhöhen verwendet man Data Augmentation
# 1. kreiert man viele Batches mit den Bildern
# 2. verwendet man zufällige Transformationen auf eine zufällige Selektion der Bilder in jedem Batch
# Das Modell wird durch die Randomisierung nie das selbe Bild in zwei oder mehr Batches vorfinden
# flow_from_directory = Pfad der Trainings und Testdasten anegeben
# rescale = obligatroisch
# Weiter Data Augmentation Methoden https://keras.io/preprocessing/image/
# Rescale Data auf Werte zwischen 1 und 0
# rescale=1./255 da Pixel Werte zwischen 1 und 255 annehmen können
from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale = 1./255,
shear_range = 0.2,
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D(2, 2)))
model.add(Conv2D(256, (2, 2), padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dropout(0.3))
model.add(Flatten())
model.add(Dense(nb_classes, activation='softmax'))
# model.add(Dense(nb_classes, activation='sigmoid'))
opt = Adam(lr=0.0001) # learning rate 조정
model.complie(optimizer=opt, loss='binary_crossentropy', metrics=['accuracy'])
# 2. 데이터셋 불러오기, 라벨링
from keras.preprocessing.image import ImageDataGenerator
train_data = ImageDataGenerator(rescale=1./255) # 모든 픽셀을 0~1 사이의 숫자로 변환
val_data = ImageDataGenerator(rescale=1./255)
test_data = ImageDataGenerator(rescale=1./255)
train_set = train_data.flow_from_directory('C:/Users/User/Desktop/dataset/train', # train set 저장 경로
target_size=(100,100), # image 크기 재설정
batch_size=1, # 이미지를 몇 개씩 가져와 처리할지 설정
class_mode='binary') # 다중분류 시 자동으로 라벨링
val_set = test_data.flow_from_directory('C:/Users/User/Desktop/dataset/test', # validation set 저장 경로
