Coverage for hypergan/samplers/autoencode_sampler.py : 17%

from hypergan.samplers.base_sampler import BaseSampler 

import tensorflow as tf 

import numpy as np 

class AutoencodeSampler(BaseSampler): 

def __init__(self, gan, samples_per_row=8): 

BaseSampler.__init__(self, gan, samples_per_row) 

self.z = None 

self.y = None 

self.x = None 

def _sample(self): 

gan = self.gan 

inputs_t = gan.inputs.x 

z_t = gan.encoder.sample 

if self.z is None: 

print("GAN IS", gan, gan.encoder) 

self.input = gan.session.run(inputs_t) 

self.z = gan.session.run(z_t, feed_dict={inputs_t: self.input}) 

destination = self.z[1] 

origin = self.z[0] 

for i in range(0, np.shape(self.z)[0], self.samples_per_row): 

last = i+self.samples_per_row-1 

multiple = np.linspace(0, 1, self.samples_per_row-4) 

for j in range(i+2, last-1): 

percent = (j - (i))/float((last) - (i+1)) 

self.z[j] = self.z[i]*(1.0-percent) + (self.z[last])*percent 

self.z[i+1] = self.z[i] 

self.z[last-1] = self.z[last] 

output = gan.session.run(gan.generator.sample, feed_dict={z_t: self.z}) 

for i in range(0, np.shape(self.z)[0], self.samples_per_row): 

last = i+self.samples_per_row-1 

output[i] = self.input[i] 

output[last] = self.input[last] 

g=tf.get_default_graph() 

with g.as_default(): 

tf.set_random_seed(1) 

return { 

'generator': output 

}