def __init__(self, num_items=3):
self.num_items = num_items
self.purple, self.blue, self.orange, self.pu_bl, self.pu_or, self.bl_pu, self.bl_or, self.or_pu, self.or_bl, \
self.pu_hand, self.bl_hand, self.or_hand = getImages()
self.env_model = MDN(num_components=NUM_COMPONENTS,
in_dim=LATENT_DIM + 4,
out_dim=LATENT_DIM,
model_path="models/env_model_0002.h5")
self.encoder = load_model("models/encoder_2001.h5")
self.dqn_model = load_model('models/controller_0002.h5')
self.decoder = load_model("models/decoder_2001.h5")
self.r_model = load_model("models/r_model_0002.h5")
self.s_bar = None
import random
from sklearn.manifold import TSNE
from keras.models import load_model
from load_process_images import getImages
latent_dim = 4
IMAGE_WIDTH = 64
IMAGE_HEIGHT = 64
CHANNELS = 3
decoder = load_model('models/decoder_1001.h5')
encoder = load_model('models/encoder_1001.h5')
imgs_list = []
for i in getImages():
imgs_list.append(i)
imgs = getImages(True)
im_size = 64
while True:
c = random.choice(range(0, imgs.shape[0]))
img = imgs[c, :, :]
print(c)
#img = imgs[117]
img = img.reshape((1, 64, 64, 3))
encoded = np.asarray(encoder.predict(img))
#encoded_logvar = encoded[1, :, :] #store log(var) vector for later
#encoded = encoded[0, :, :] #get just means
def autoencode_images():
images = getImages(return_single=True)
encoder = load_model('models/encoder_1001.h5')
return np.asarray(encoder.predict(images))
activation='relu')(h_decoded)
deconv = Conv2DTranspose(64, (5, 5), strides=(2, 2), activation='relu')(deconv)
deconv = Conv2DTranspose(32, (6, 6), strides=(2, 2), activation='relu')(deconv)
decoded_mean = Conv2DTranspose(3, (6, 6), strides=(2, 2),
activation='sigmoid')(deconv)
encoder = Model(img_input, z, name='encoder')
decoder = Model(decoder_input, decoded_mean, name='decoder')
decoder.summary()
reconstructed = decoder(z)
ae = Model(img_input, reconstructed, name='vae')
opt = RMSprop(lr=0.00025)
ae.compile(optimizer='adam', loss=vae_loss)
ae.summary()
x_train = y_train = x_test = y_test = getImages(return_single=True)
try:
history = ae.fit(x_train,
x_train,
shuffle=True,
epochs=epochs,
batch_size=batch_size,
validation_data=(x_test, x_test))
finally:
sss = 0
encoder.save('models/encoder_1001.h5')
decoder.save('models/decoder_1001.h5')
ae.save('models/ae_1001.h5')
n = 15
h_decoded = Reshape((1, 1, 1024))(h_decoded)
deconv = Conv2DTranspose(128, (5, 5), strides= (2,2), activation='relu')(h_decoded)
deconv = Conv2DTranspose(64, (5, 5), strides= (2,2), activation='relu')(deconv)
deconv = Conv2DTranspose(32, (6, 6), strides= (2,2), activation='relu')(deconv)
decoded_mean = Conv2DTranspose(3, (6, 6), strides= (2,2), activation='sigmoid')(deconv)
encoder = Model(img_input, [z_mean, z_log_var], name='encoder')
decoder = Model(decoder_input, decoded_mean, name='decoder')
decoder.summary()
reconstructed = decoder(z)
vae = Model(img_input, reconstructed, name='vae')
opt = RMSprop(lr=0.00025)
vae.compile(optimizer='adam', loss=vae_loss)
vae.summary()
x_train = y_train = x_test = y_test = getImages(return_single=True, use_all=False, val=False)
try:
history = vae.fit(x_train, x_train,
shuffle=True,
epochs=epochs,
batch_size=batch_size,
validation_data=(x_test, x_test))
finally:
sss = 0
encoder.save('models/encoder_2001.h5')
decoder.save('models/decoder_2001.h5')
vae.save('models/vae_2001.h5')
n = 15
def __init__(self, num_items=3, use_all=True, val=False):
self.state = None
self.num_items = num_items
self.purple, self.blue, self.orange, self.pu_bl, self.pu_or, self.bl_pu, self.bl_or, self.or_pu, self.or_bl,\
self.pu_hand, self.bl_hand, self.or_hand = getImages(return_single=False ,use_all=use_all, val=val)