import tensorflow as tf
import hypergan as hg
import hyperchamber as hc
from hypergan.generators import *
from hypergan.search.random_search import RandomSearch
from hypergan.viewer import GlobalViewer
from common import *
arg_parser = ArgumentParser("Test your gan vs a known distribution", require_directory=False)
arg_parser.parser.add_argument('--distribution', '-t', type=str, default='circle', help='what distribution to test, options are circle, modes')
arg_parser.parser.add_argument('--sequence_length', '-n', type=int, default=2, help='how many steps to look forward')
args = arg_parser.parse_args()
config = lookup_config(args)
if args.action == 'search':
config = RandomSearch({}).random_config()
class Sequence2DGenerator(BaseGenerator):
def create(self):
gan = self.gan
config = self.config
ops = self.ops
end_features = config.end_features or 2*args.sequence_length
ops.describe('custom_generator')
net = gan.encoder.sample
for i in range(2):
net = ops.linear(net, 32)
net = ops.lookup('bipolar')(net)
net = ops.linear(net, end_features)
[input.reshape(self.gan.batch_size(), -1), context['digit']], 1)
net = self.linear(net)
net = self.relu(net)
net = self.linear2(net)
net = self.tanh(net)
return net
config = lookup_config(args)
if args.action == 'search':
search = RandomSearch({
'generator': {
'class': MNISTGenerator,
'end_features': 10
},
'discriminator': {
'class': MNISTDiscriminator
}
})
config = search.random_config()
inputs = MNISTInputLoader(args.batch_size)
def setup_gan(config, inputs, args):
gan = MNISTGAN(config, inputs=inputs)
return gan
def test_random_config(self):
rs = RandomSearch({})
self.assertTrue(rs.random_config()['trainer']["class"] != None)
def test_trainers(self):
rs = RandomSearch({})
self.assertTrue(rs.trainer()["class"] != None)
def test_range(self):
rs = RandomSearch({})
self.assertTrue(isinstance(rs.range(), list))
ynet = tf.concat(axis=1, values=[x,y])
net = tf.concat(axis=0, values=[ynet, gnet])
net = ops.linear(net, 128)
net = tf.nn.tanh(net)
self.sample = net
return net
config = lookup_config(args)
if args.action == 'search':
search = RandomSearch({
'generator': {'class': MNISTGenerator, 'end_features': 10},
'discriminator': {'class': MNISTDiscriminator}
})
config = search.random_config()
mnist_loader = MNISTInputLoader(args.batch_size)
def setup_gan(config, inputs, args):
gan = hg.GAN(config, inputs=inputs, batch_size=args.batch_size)
return gan
def train(config, args):
gan = setup_gan(config, mnist_loader, args)
correct_prediction = tf.equal(tf.argmax(gan.generator.layer('fy'),1), tf.argmax(gan.inputs.y,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) * 100
metrics = [accuracy]