def compile_model(model, optimizer=Adam()):
#model.compile(optimizer=SGD(lr=C.learn_rate, momentum=0.9),
# loss=std_triplet_loss())
model.compile(optimizer=optimizer, loss=std_triplet_loss())
return 'Model compiled with lr={}'.format(Keval(optimizer.lr))
validation_steps=100)
save_loss(history, i)
if last == 0:
log('Creating base network from scratch.')
if not os.path.exists('models'):
os.makedirs('models')
base_model = create_base_network(in_dim)
else:
log('Loading model:' + save_name(last))
base_model = load_model(save_name(last))
model = tripletize(base_model)
model.compile(optimizer=SGD(lr=C.learn_rate, momentum=0.9),
loss=std_triplet_loss())
def avg(x):
return sum(x) / len(x)
vs = T.get_vectors(base_model, C.val_dir)
cents = {}
for v in vs:
cents[v] = T.centroid(vs[v])
for i in range(last + 1, last + 11):
log('Starting iteration ' + str(i) + '/' + str(last + 10) + ' lr=' +
str(C.learn_rate))
train_step(i)