'p_miss': 0.5,
'p_hint': 0.5
},
optimizer=tf.keras.optimizers.Adam())
Model3 = GAN(summary_writer=Model1.summary_writer,
hyperParams={
'p_miss': 0.5,
'p_miss': 0.5,
'alpha': 0.1,
'episode_num': 10
},
optimizer=tf.keras.optimizers.Adam(1e-4))
#%% Run - Step 1
# First train the discriminator against a random generator to increase its stability
counter = 0
train, test = Data.getPipeLine(train_rate=0.8, batch_ratio=1, repeat=2000)
test = iter(test)
for dat_train in tqdm(train):
Model1.trainWithSteps(dat_train,
randomGenerator,
Discriminator,
steps=False)
if (counter % 20 == 0):
Model1.performanceLog('<Random Generator>(train)', dat_train,
randomGenerator, Discriminator)
Model1.performanceLog('<Random Generator>(test)', test.next(),
randomGenerator, Discriminator)
counter += 1
Discriminator.save(Model1.logdir + 'Models\DiscriminatorS1E{}'.format(1))
#%% Run - Step 2
from DataModel import DataModel
from tqdm import tqdm
from os import getcwd
from VAE import VAE
from pathlib import Path
#%% Data Model
file = 'measureGenerator'
data_path = "{}\\data\\{}.csv".format(Path(getcwd()).parent, file)
Data = DataModel(data_path)
#%% Models
Dim = Data.Dim
encoder = Encoder(
compositLayers(
[Dim * 12, Dim * 6, Dim * 3, Dim * 2, Dim * 3, Dim * 6, Dim * 12, Dim],
0))
decoder = Decoder(
compositLayers(
[Dim * 12, Dim * 6, Dim * 3, Dim * 2, Dim * 3, Dim * 6, Dim * 12, Dim],
0))
Model = VAE(hyperParams={'G_train_step': 1})
#%% Run
counter = 0
for dat_train, dat_test in tqdm(
Data.getPipeLine(train_rate=0.8, batch_ratio=0.2, repeat=500)):
Model.trainWithBatch(dat_train, encoder, decoder)
# %%
