print("=" * 50)
print("Number of Training Examples: %d/%d" %
(len(train_loader.sampler), len(indices)))
print("Number of Validation Examples: %d/%d" %
(len(valid_loader.sampler), len(indices)))
print("Number of Test Examples: %d/%d" %
(len(test_loader.sampler), len(indices)))
print("=" * 50, "\n")
data_loaders = {"train": train_loader, "valid": valid_loader}
###
latent_dim = 2048
CAE = Model(z_dim=latent_dim)
CAE = nn.DataParallel(CAE).cuda()
### Training
n_epochs = 200
optimiser = optim.Adam(CAE.parameters(),
lr=0.0001,
weight_decay=1e-6,
amsgrad=True)
###
best_val_loss = np.infty
print("\n")
print("Data Statistics")
print("=" * 50)
print("Number of Training Examples: %d/%d" %
(len(train_loader.sampler), len(indices)))
print("Number of Validation Examples: %d/%d" %
(len(valid_loader.sampler), len(indices)))
print("Number of Test Examples: %d/%d" %
(len(test_loader.sampler), len(indices)))
print("=" * 50, "\n")
data_loaders = {"train": train_loader, "valid": valid_loader}
###
CAE = Model(z_dim=2048)
CAE = nn.DataParallel(CAE).cuda()
CAE.load_state_dict(torch.load("./models/BroadCAE.pth"))
###
NN = NeuralNet()
NN = nn.DataParallel(NN).cuda()
NN.load_state_dict(torch.load("./models/PretrainedNeuralNetwork.pth"))
###
SilenceCorpus=None)
encoder = train_corpus.encoder
decoder = train_corpus.decoder
num_classes = len(decoder) - 1
# set_graph Graph
batch_size = batch_parameters.batch_size
is_training = cfg.is_training
max_gradient = cfg.max_gradient
training_iters = train_corpus.len
cnn_model = Model(cfg)
graph = tf.Graph()
with graph.as_default():
# tf Graph input
tf.set_random_seed(cfg.tf_seed)
with tf.name_scope("Input"):
x = tf.placeholder(tf.float32,
shape=batch_parameters.input_shape,
name="input")
y = tf.placeholder(tf.int64, shape=(None, ), name="input")
keep_prob = tf.placeholder(tf.float32, name="dropout")
with tf.variable_scope('logit'):
logits = cnn_model.calc_logits(x, keep_prob, num_classes)
predictions = tf.nn.softmax(logits)
cfg = Config()
silence_corpus = SoundCorpus(cfg.soundcorpus_dir, mode='silence')
test_corpus = SoundCorpus(cfg.soundcorpus_dir,
mode='own_test',
fn='own_test.p.soundcorpus.p')
silence_classifier = SilenceDetector()
batch = [item for item in test_corpus]
decoder = silence_corpus.decoder
encoder = silence_corpus.encoder
num_classes = len(decoder) - 1
model = Model(cfg)
# set_graph Graph
# batch_size = cfg.batch_size
is_training = cfg.is_training
graph = tf.Graph()
with graph.as_default():
# tf Graph input
tf.set_random_seed(3)
with tf.name_scope("Input"):
x = tf.placeholder(tf.float32, shape=(None, 99, 13, 3), name="input")
# x.set_shape([batch_size, size])
#y = tf.placeholder(tf.int64, shape=(None,), name="input")
keep_prob = tf.placeholder(tf.float32, name="dropout")
print("=" * 50)
print("Number of Training Examples: %d/%d" %
(len(train_loader.sampler), len(indices)))
print("Number of Validation Examples: %d/%d" %
(len(valid_loader.sampler), len(indices)))
print("Number of Test Examples: %d/%d" %
(len(test_loader.sampler), len(indices)))
print("=" * 50, "\n")
data_loaders = {"train": train_loader, "valid": valid_loader}
###
latent_dim = 2048
CAE = Model(z_dim=latent_dim)
CAE = nn.DataParallel(CAE).cuda()
CAE.load_state_dict(torch.load("./models/BroadCAE.pth"))
CAE.eval()
###
n_epochs = 200
optimiser = optim.SGD(NN.parameters(), lr=1e-1, weight_decay=1e-3)
best_val_loss = np.infty