def test_traineval(self):
tf.logging.set_verbosity(tf.logging.INFO)
path = PATH.decode(sys.stdout.encoding)
#2 files, 64 epochs, batchsize 32 => 2*64/32 = 4 iterations
dset = lambda: ds.dataset_with_preprocess(LISTFILE_1, path,
epochs=4,
batchsize=32,
segs_per_sample=16,
)
dset_eval = lambda: ds.dataset_with_preprocess(LISTFILE_1, path,
epochs=1,
batchsize=16,
segs_per_sample=16,
shuffle=False
)
config = self.config.replace(save_checkpoints_steps=2)
tfnet_est = TFNetEstimator(**nets.default_net(), config=config)
input_fn = lambda: dset().make_one_shot_iterator().get_next()
eval_input_fn = lambda: dset_eval().make_one_shot_iterator().get_next()
train_spec = tf.estimator.TrainSpec(input_fn)
eval_spec = tf.estimator.EvalSpec(eval_input_fn)
tf.estimator.train_and_evaluate(tfnet_est, train_spec, eval_spec)
self.assertIsNotNone(tfnet_est)
def main(_):
path = PATH.decode(sys.stdout.encoding)
dset = ds.dataset_with_preprocess(
LISTFILE_1,
path,
epochs=1,
batchsize=32,
segs_per_sample=64,
)
tfnet_est = TFNetEstimator(**nets.default_net(),
model_dir='tests/dummymodel')
tfnet_est.train(input_fn=lambda: dset.make_one_shot_iterator().get_next())
def test_train(self):
"""Runs a training pass with test data for 4 iterations
iterations of batchsize 16"""
tf.logging.set_verbosity(tf.logging.INFO)
path = PATH.decode(sys.stdout.encoding)
#2 files, 64 epochs, batchsize 32 => 2*64/32 = 4 iterations
dset = lambda: ds.dataset_with_preprocess(LISTFILE_1, path,
epochs=64,
batchsize=32,
)
tfnet_est = TFNetEstimator(**nets.default_net(), config=self.config)
tfnet_est.train(
input_fn=lambda: dset().make_one_shot_iterator().get_next())
self.assertIsNotNone(tfnet_est)
def test_loadmodel(self):
"""Test running eval from with trained model"""
tf.logging.set_verbosity(tf.logging.INFO)
path = PATH.decode(sys.stdout.encoding)
#2 files, 64 epochs, batchsize 32 => 2*64/32 = 4 iterations
dset = ds.single_file_dataset(LQ_AUDIO_FILE,
)
#RunConfig for more more printing since we are only training for very few steps
config = tf.estimator.RunConfig(log_step_count_steps=1)
tfnet_est = TFNetEstimator(**nets.default_net(), config=config,
model_dir=DUMMY_MODEL_PATH
)
preds = tfnet_est.predict(
input_fn=lambda: dset.make_one_shot_iterator().get_next())
for pred in preds:
self.assertEqual(pred.shape, (8192, 1))
def test_loadmodel(self):
"""Test running eval from with trained model"""
tf.logging.set_verbosity(tf.logging.INFO)
path = PATH.decode(sys.stdout.encoding)
#2 files, 64 epochs, batchsize 32 => 2*64/32 = 4 iterations
dset = ds.dataset_with_preprocess(
LISTFILE_1,
path,
epochs=1,
batchsize=16,
)
#RunConfig for more more printing since we are only training for very few steps
config = tf.estimator.RunConfig(log_step_count_steps=1)
tfnet_est = TFNetEstimator(**nets.default_net(),
config=config,
model_dir=DUMMY_MODEL_PATH)
tfnet_est.evaluate(
input_fn=lambda: dset.make_one_shot_iterator().get_next())
self.assertIsNotNone(tfnet_est)