def gather_data(self, step, t_error, gann, cman: CustomCaseManager):
if (not step % self.interval) or (step + 1 == gann.options.steps):
self.append_error(step, t_error,
self.t_err) # Standard training error
if len(cman.get_validation_cases()) > 0:
self.append_error(step,
gann.do_testing(cman.get_validation_cases()),
self.v_err) # Validation error
self.updated = True
def get_glass_options(session_tracker):
cases = DataSets.Glass()
lrate = 0.001
cman = CustomCaseManager(cases, None, cfrac=1, vfrac=0.1, tfrac=0.1)
params = {
'net_dims': calc_net_dims(cases[0], [80, 80]),
'h_activation_function': tf.nn.relu,
'o_activation_function': tf.nn.softmax,
'cost_function': tf.losses.softmax_cross_entropy,
'learning_rate': lrate,
'weight_range': [-.5, .5],
'optimizer': tf.train.AdamOptimizer(lrate, 0.9, 0.999),
'case_manager': cman,
'minibatch_size': 75,
'steps': 1000,
'vint': 10,
'session_tracker': session_tracker,
# MAP options
'map_case_count': 10,
'map_case_func': cman.get_testing_cases
}
# Options used throughout the program
return Options(**params)
def get_bit_counter_options(session_tracker):
cases = DataSets.Bit_Counter(500, 15)
lrate = 0.1
cman = CustomCaseManager(cases, None, cfrac=1, vfrac=0.1, tfrac=0.1)
params = {
'net_dims': calc_net_dims(cases[0], [80, 80]),
'h_activation_function': tf.nn.relu,
'o_activation_function': tf.nn.softmax,
'cost_function': tf.losses.mean_squared_error,
'learning_rate': lrate,
'weight_range': [-.2, .2],
'optimizer': tf.train.AdagradOptimizer(lrate, 0.001),
'case_manager': cman,
'minibatch_size': 100,
'steps': 10000,
'vint': 100,
'session_tracker': session_tracker,
# MAP options
'map_case_count': 10,
'map_case_func': cman.get_testing_cases
}
# Options used throughout the program
return Options(**params)