def main():
max_iter = 150000
set_name = 'mnist'
latent_size = 128
class_num = DATASET_CLASS_COUNT[set_name]
batch_size = 100
num_labeled = 100
l2 = False
temp = 1
share_encoder = True
model = Model(set_name,
latent_size,
class_num,
share_encoder=share_encoder,
temp=temp,
l2=l2)
data = Data(set_name,
batch_size,
num_labeled=num_labeled,
label_map_index=0)
# lr = scheduler.CustomSchedule(latent_size, 2000)
lr = 1e-4
opt = tf.keras.optimizers.Adam(lr, beta_1=0.9, beta_2=0.98, epsilon=1e-9)
name = 'adjNL{}Share{}Tmp{}l2{}LS{}'.format(num_labeled, share_encoder,
temp, l2, latent_size)
time_string = name + '_' + strftime("%d%b-%H%M", gmtime())
result_path = os.path.join(ROOT_PATH, 'result', set_name)
save_path = os.path.join(result_path, 'model', time_string)
summary_path = os.path.join(result_path, 'log', time_string)
if not os.path.exists(result_path):
os.makedirs(result_path)
if not os.path.exists(save_path):
os.makedirs(save_path)
writer = tf.summary.create_file_writer(summary_path)
checkpoint = tf.train.Checkpoint(opt=opt, model=model)
for i in range(max_iter):
with writer.as_default():
train_loss = step_train(model, data, opt, i)
if i == 0:
print(model.summary())
if (i + 1) % 200 == 0:
print('Step: {}, Loss: {}'.format(i, train_loss.numpy()))
test_acc = step_val(model, data, i)
print('Hook: {}, acc: {}'.format(i, test_acc.numpy()))
if (i + 1) % 5000 == 0:
save_name = os.path.join(save_path, '_' + str(i))
checkpoint.save(file_prefix=save_name)
def test_padding_sketch_filter(self):
# GIVEN
sketch_string = "Root1(3) Root(3) Sel(0) N(2) Filter(2)"
sketch = [eval(x) for x in sketch_string.split(' ')]
expected_padding_string = "Root1(3) Root(3) Sel(0) N(2) A(0) C(0) T(0) A(0) C(0) T(0) A(0) C(0) T(0) Filter(2) A(0) C(0) T(0) V(0)"
basic_model = BasicModel()
# WHEN
actual_padding = basic_model.padding_sketch(sketch)
# THEN
self.assertEqual(expected_padding_string,
' '.join([str(x) for x in actual_padding]))
def main():
charset_size = len(CHARSET)
max_word_len = 32
build_params = {
"max_word_len": max_word_len,
"charset_size": charset_size,
"lstm_layers_sizes": (512, 512)
}
model = BasicModel(build_params, "data/logs/run_3/train",
"data/logs/run_3/validation",
"data/logs/run_3/checkpoints",
"data/logs/run_3/best_model")
dataset = Dataset("data/dataset/train_set.csv",
"data/dataset/validation_set.csv",
"data/dataset/test_set.csv", max_word_len)
train_params = {"lstm_dropout": 0.5}
model.train(dataset, **train_params)
def main():
conf = parser.parse_args()
time_string = strftime("%a%d%b%Y-%H%M%S", gmtime())
result_path = os.path.join(ROOT_PATH, 'result', conf.set_name)
task_name = conf.task_name
save_path = os.path.join(result_path, 'model',
task_name + '_' + time_string)
summary_path = os.path.join(result_path, 'log',
task_name + '_' + time_string)
if not os.path.exists(result_path):
os.makedirs(result_path)
if not os.path.exists(save_path):
os.makedirs(save_path)
model = BasicModel(conf)
data = load_data(conf, training=True)
data_iter = iter(data)
opt = tf.keras.optimizers.Adam(1e-4)
writer = tf.summary.create_file_writer(summary_path)
checkpoint = tf.train.Checkpoint(actor_opt=opt, model=model)
for i in range(conf.max_iter):
batch_1 = next(data_iter)
batch_2 = next(data_iter)
with writer.as_default():
loss = step_train(conf, batch_1, batch_2, model, opt, i)
if i == 0:
print(model.summary())
if i % 50 == 0:
print('iter {}, loss {}'.format(i, loss))
if i % 5000 == 0 and i > 0:
save_name = os.path.join(save_path, '_' + str(i))
checkpoint.save(file_prefix=save_name)
def step_train(model: Model, data: Data, opt: tf.keras.optimizers.Optimizer,
step):
s_d, u_d = data.next_train()
shape = np.prod(DATASET_SHAPE[data.set_name][1:])
batch_feed = [
tf.reshape(s_d[0], [-1, shape]),
tf.reshape(u_d[0], [-1, shape]), s_d[1], u_d[1]
]
batch_feed = [tf.identity(i) for i in batch_feed]
summary_step = -1 if step % 50 > 0 else step
with tf.GradientTape() as tape:
x, bbn, context_ind, feat, pred, adj = model(batch_feed,
training=True,
step=summary_step)
img = tf.concat([batch_feed[0], batch_feed[1]], axis=0)
loss = model.obj(img,
s_d[1],
x,
bbn,
context_ind,
feat,
adj,
pred,
step=summary_step)
gradient = tape.gradient(loss, sources=model.trainable_variables)
opt.apply_gradients(zip(gradient, model.trainable_variables))
if summary_step >= 0:
acc = eval.acc(s_d[1], pred)
err = 1 - acc
ll = tf.concat([batch_feed[2], batch_feed[3]], axis=0)
if tf.shape(ll).shape[0] < 2:
ll = tf.one_hot(ll, DATASET_CLASS_COUNT[data.set_name])
sim = tf.expand_dims(
tf.expand_dims(tf.matmul(ll, ll, transpose_b=True), 0), -1)
tf.summary.image('gt/adj', sim, step=step, max_outputs=1)
tf.summary.scalar('train/acc', acc, step=step)
tf.summary.scalar('train/err', err, step=step)
return loss
def setUp(self):
self.model = BasicModel()
self.model._hp.parse()
self.mp = MovieQAPath()