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

from hypergan.samplers.base_sampler import BaseSampler 

import tensorflow as tf 

import numpy as np 

class AlphaganRandomWalkSampler(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 

self.step = 0 

self.steps = 8 

self.target = None 

def _sample(self): 

gan = self.gan 

z_t = gan.uniform_encoder.sample 

inputs_t = gan.inputs.x 

if self.z is None: 

self.z = gan.uniform_encoder.sample.eval()/2 

direct = gan.uniform_encoder.sample.eval()[0]/2 

direct = np.reshape(direct, [1, direct.shape[0]]) 

self.direction = np.tile(direct, [self.z.shape[0], 1]) 

self.input = gan.session.run(gan.inputs.x) 

if self.step > self.steps: 

self.z = np.minimum(self.z+self.direction, 1) 

self.z = np.maximum(self.z, -1) 

self.direction = gan.uniform_encoder.sample.eval() 

self.step = 0 

percent = float(self.step)/self.steps 

z_interp = self.z + self.direction*percent 

z_interp = np.minimum(z_interp, 1) 

z_interp = np.maximum(z_interp, -1) 

self.step+=1 

g=tf.get_default_graph() 

with g.as_default(): 

tf.set_random_seed(1) 

return { 

'generator': gan.session.run(gan.uniform_sample, feed_dict={z_t: z_interp, inputs_t: self.input}) 

}