Coverage for hypergan/trainers/multi_step_trainer.py : 21%

import tensorflow as tf 

import numpy as np 

import hyperchamber as hc 

import inspect 

from hypergan.trainers.base_trainer import BaseTrainer 

TINY = 1e-12 

class MultiStepTrainer(BaseTrainer): 

def __init__(self, gan, config, losses=[], var_lists=[], metrics=None): 

BaseTrainer.__init__(self, gan, config) 

self.losses = losses 

self.var_lists = var_lists 

self.metrics = metrics or [None for i in self.losses] 

def _create(self): 

gan = self.gan 

config = self.config 

losses = self.losses 

g_lr = config.g_learn_rate 

d_lr = config.d_learn_rate 

optimizers = [] 

for i, _ in enumerate(losses): 

loss = losses[i][1] 

var_list = self.var_lists[i] 

is_generator = losses[i][0] == 'generator' 

if is_generator: 

optimizer = self.build_optimizer(config, 'g_', config.g_trainer, self.g_lr, var_list, loss) 

else: 

optimizer = self.build_optimizer(config, 'd_', config.d_trainer, self.d_lr, var_list, loss) 

optimizers.append(optimizer) 

self.optimizers = optimizers 

if config.d_clipped_weights: 

self.clip = [tf.assign(d,tf.clip_by_value(d, -config.d_clipped_weights, config.d_clipped_weights)) for d in d_vars] 

else: 

self.clip = [] 

return None 

def _step(self, feed_dict): 

gan = self.gan 

sess = gan.session 

config = self.config 

losses = self.losses 

metrics = self.metrics 

for i, _ in enumerate(losses): 

loss = losses[i] 

optimizer = self.optimizers[i] 

metric = metrics[i] 

if(metric): 

metric_values = sess.run([optimizer] + self.output_variables(metric), feed_dict)[1:] 

if self.current_step % 100 == 0: 

print("loss " + str(i) + " "+ loss[0] + " " + self.output_string(metric) % tuple([self.current_step] + metric_values)) 

else: 

_ = sess.run(optimizer, feed_dict)