def train_and_calc_inf_network(
i, j, k, layerSize, num_of_ephocs, learning_rate_local, batch_size,
indexes, save_grads, data_sets_org, model_type, percent_of_train,
interval_accuracy_display, calc_information, calc_information_last,
num_of_bins, interval_information_display, save_ws, rand_int, cov_net):
"""Train the network and calculate it's information"""
network_name = '{0}_{1}_{2}_{3}'.format(i, j, k, rand_int)
print('Training network - {0}'.format(network_name))
network = train_network(layerSize, num_of_ephocs, learning_rate_local,
batch_size, indexes, save_grads, data_sets_org,
model_type, percent_of_train,
interval_accuracy_display, network_name, cov_net)
network['information'] = []
if calc_information:
print('Calculating the information')
infomration = np.array([
inn.get_information(network['ws'], data_sets_org.data,
data_sets_org.labels, num_of_bins,
interval_information_display, network['model'],
layerSize)
])
network['information'] = infomration
elif calc_information_last:
print('Calculating the infomration for the last epoch')
infomration = np.array([
inn.get_information([network['ws'][-1]], data_sets_org.data,
data_sets_org.labels, num_of_bins,
interval_information_display, network['model'],
layerSize)
])
network['information'] = infomration
# If we dont want to save layer's output
if not save_ws:
network['weights'] = 0
return network
def train_and_calc_inf_network(i, j, k, layerSize, num_of_ephocs, learning_rate_local, batch_size, indexes, save_grads,
data_sets_org,
model_type, percent_of_train, interval_accuracy_display, calc_information,
calc_information_last, num_of_bins,
interval_information_display, save_ws, rand_int, cov_net):
"""Train the network and calculate it's information"""
network_name = '{0}_{1}_{2}_{3}'.format(i, j, k, rand_int)
print ('Training network - {0}'.format(network_name))
network = train_network(layerSize, num_of_ephocs, learning_rate_local, batch_size, indexes, save_grads,
data_sets_org, model_type, percent_of_train, interval_accuracy_display, network_name,
cov_net)
network['information'] = []
if calc_information:
print ('Calculating the infomration')
infomration = np.array([inn.get_information(network['ws'], data_sets_org.data, data_sets_org.labels,
num_of_bins, interval_information_display, network['model'],
layerSize)])
network['information'] = infomration
elif calc_information_last:
print ('Calculating the infomration for the last epoch')
infomration = np.array([inn.get_information([network['ws'][-1]], data_sets_org.data, data_sets_org.labels,
num_of_bins, interval_information_display,
network['model'], layerSize)])
network['information'] = infomration
# If we dont want to save layer's output
if not save_ws:
network['weights'] = 0
return network
def calc_information_last(self):
"""Calculate the information of the last epoch"""
if self.traind_network and self.save_ws:
return np.array([inn.get_information([self.ws[k][j][i][-1]], self.data_sets.data, self.data_sets.labels,
self.args.num_of_bins, self.args.interval_information_display, self.epochs_indexes)
for i in range(len(self.train_samples)) for j in
range(len(self.layers_sizes)) for k in range(self.args.num_of_repeats)])
def calc_information(self):
"""Calculate the infomration of the network for all the epochs - only valid if we save the activation values and trained the network"""
if self.traind_network and self.save_ws:
self.information = np.array([inn.get_information(self.ws[k][j][i], self.data_sets.data, self.data_sets.labels,
self.args.num_of_bins, self.args.interval_information_display, self.epochs_indexes)
for i in range(len(self.train_samples)) for j in
range(len(self.layers_sizes)) for k in range(self.args.num_of_repeats)])
else:
print ('Cant calculate the infomration of the networks!!!')
def train_and_calc_inf_network(
i, j, k, layerSize, num_of_ephocs, learning_rate_local, batch_size,
indexes, save_grads, data_sets_org, model_type, percent_of_train,
interval_accuracy_display, calc_information, calc_information_last,
num_of_bins, interval_information_display, save_ws, rand_int, cov_net):
"""Train the network and calculate it's information"""
# where is the "network" getting trained? Don't see it getting passed in...
network_name = '{0}_{1}_{2}_{3}'.format(i, j, k, rand_int)
print('Training network - {0}'.format(network_name))
# so technically everything is saved in network dict here!!
# but NOT information, only 'ws' and other things
network = train_network(layerSize, num_of_ephocs, learning_rate_local,
batch_size, indexes, save_grads, data_sets_org,
model_type, percent_of_train,
interval_accuracy_display, network_name, cov_net)
network['information'] = []
if calc_information: # default branch
# trained model with the history of training is passed into this function
print('Calculating the infomration')
infomration = np.array([
inn.get_information(network['ws'], data_sets_org.data,
data_sets_org.labels, num_of_bins,
interval_information_display, network['model'],
layerSize)
])
network['information'] = infomration
elif calc_information_last:
print('Calculating the infomration only for the last epoch')
infomration = np.array([
inn.get_information([network['ws'][-1]], data_sets_org.data,
data_sets_org.labels, num_of_bins,
interval_information_display, network['model'],
layerSize)
])
network['information'] = infomration
# ws is NOT even saved here...it's saved outside
if not save_ws:
network['weights'] = 0 # we changed save_ws to default true
return network
def train_and_calc_inf_network(i, j, k, layers, num_of_ephocs, lr_local,
batchsize, indexes, data_sets, model_type,
percent_of_train, nbins, cov):
"""Train the network and calculate its information"""
network_name = '{0}_{1}_{2}'.format(i, j, k)
print('Training network - {0}'.format(network_name))
network = train_network(layers, num_of_ephocs, lr_local, batchsize,
indexes, data_sets, model_type, percent_of_train,
network_name, cov)
network['information'] = []
print('Calculating the infomration')
infomration = np.array([
inn.get_information(network['ws'], data_sets[:, :-1],
data_sets[:, -1:], nbins, network['model'], layers)
])
network['information'] = infomration
print('Successfully calculated the infomration !!!')
return network
