def tune(n_times=25):
for i in range(1, n_times):
print("Running tuning iteration number {}".format(i))
hyperparameter_setting = hyperparameters(batch_size=(10, 100),
num_units=(16, 150),
embedding_size=(40, 100),
learning_rate=(.00001, 0.1))
hyperparameter_setting.batch_size = sample_randomly(
hyperparameter_setting.batch_size,
use_log_scale=False,
sample_int=True,
from_list=False)
hyperparameter_setting.num_units = sample_randomly(
hyperparameter_setting.num_units,
use_log_scale=False,
sample_int=True,
from_list=False)
hyperparameter_setting.embedding_size = sample_randomly(
hyperparameter_setting.embedding_size,
use_log_scale=False,
sample_int=True,
from_list=False)
hyperparameter_setting.learning_rate = sample_randomly(
hyperparameter_setting.learning_rate,
use_log_scale=True,
sample_int=False,
from_list=False)
print("The configuration is ", hyperparameter_setting.__dict__)
train(resume_training=False,
total_epochs=8,
test_after_n_iter=10,
disp_loss_after_n_iter=10,
merge_summary_after_n_iter=10,
hyperparameters=hyperparameter_setting)
from os.path import join, expanduser
import tensorboard_logger
from tensorboard_logger import *
from hyperparameters import hyperparameters
vox_dir ='vox'
global_step = 0
global_epoch = 0
use_cuda = torch.cuda.is_available()
if use_cuda:
cudnn.benchmark = False
use_multigpu = None
hparams = hyperparameters()
print(hparams)
fs = hparams.sample_rate
use_assistant = 1
def train(model, train_loader, val_loader, optimizer,
init_lr=0.002,
checkpoint_dir=None, checkpoint_interval=None, nepochs=None,
clip_thresh=1.0, assistant=None):
model.train()
if use_cuda:
model = model.cuda()
linear_dim = model.linear_dim