def test_deepar_init():
config = process_config(config_path)
model = deeparsys.DeepARSysModel(config)
model.delete_checkpoints()
create_dirs([
config.summary_dir, config.checkpoint_dir, config.plots_dir,
config.output_dir
])
assert os.path.exists(config.summary_dir)
assert os.path.exists(config.output_dir)
assert os.path.exists(config.checkpoint_dir)
data = data_generator.DataGenerator(config)
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
logger = Logger(sess, config)
trainer = deeparsys.DeepARSysTrainer(sess, model, data, config, logger)
trainer.eval_step()
trainer.train_step()
model.load(sess)
trainer.train()
trainer.eval_step()
def test_base_model():
base_config = Bunch(config)
base_model = base.BaseModel(base_config)
base_model.init_saver()
base.BaseModel.gaussian_likelihood(1.)
create_dirs([base_config.summary_dir, base_config.checkpoint_dir])
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
logger = Logger(sess, base_config)
logger.summarize(0, summaries_dict={})
def test_deepar_init():
config = process_config(config_path)
create_dirs([config.summary_dir, config.checkpoint_dir])
model = deepar.DeepARModel(config)
# model.delete_checkpoints()
data = data_generator.DataGenerator(config)
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
logger = Logger(sess, config)
trainer = deepar.DeepARTrainer(sess, model, data, config, logger)
trainer.train()
os.path.join("experiments", args.experiment.strip()))
print(config_file)
else:
config_file = args.config
print('ok2')
config = process_config(config_file)
except:
print("missing or invalid arguments")
exit(0)
model = deepar.DeepARModel(config)
data = data_generator.DataGenerator(config)
create_dirs([
config.summary_dir, config.checkpoint_dir, config.plots_dir,
config.output_dir
])
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
logger = Logger(sess, config)
trainer = deepar.DeepARTrainer(sess, model, data, config, logger)
if trainer.config.from_scratch:
model.delete_checkpoints()
model.load(sess)
trainer.train(verbose=True)
samples = trainer.sample_on_test()