def get_dump_path(params):
"""
Create a directory to store the experiment.
"""
if fs_tracker:
return fs_tracker.get_model_directory()
if os.environ.get('MODELS_PATH'):
return os.environ.get('MODELS_PATH')
MODELS_PATH = DEFAULT_MODELS_PATH
assert os.path.isdir(MODELS_PATH)
# create the sweep path if it does not exist
sweep_path = os.path.join(MODELS_PATH, params.name)
if not os.path.exists(sweep_path):
subprocess.Popen("mkdir %s" % sweep_path, shell=True).wait()
# create a random name for the experiment
chars = 'abcdefghijklmnopqrstuvwxyz0123456789'
while True:
exp_id = ''.join(random.choice(chars) for _ in range(10))
dump_path = os.path.join(MODELS_PATH, params.name, exp_id)
if not os.path.isdir(dump_path):
break
# create the dump folder
if not os.path.isdir(dump_path):
subprocess.Popen("mkdir %s" % dump_path, shell=True).wait()
return dump_path
def test_get_model_directory_noargs(self):
testExperiment = 'testExperiment'
testPath = os.path.join(os.path.expanduser('~'),
'.studioml/experiments', testExperiment,
'modeldir')
os.environ['STUDIOML_EXPERIMENT'] = testExperiment
self.assertTrue(testPath == fs_tracker.get_model_directory())
def test_get_model_directory_args(self):
experimentName = 'testExperiment'
modelDir = fs_tracker.get_model_directory(experimentName)
self.assertTrue(modelDir == os.path.join(
os.path.expanduser('~'),
'.studioml/experiments/testExperiment/modeldir'))
model = make_parallel(model, no_gpus)
model.compile(loss='categorical_crossentropy', optimizer='adam')
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(60000, 784)
x_test = x_test.reshape(10000, 784)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
# convert class vectors to binary class matrices
y_train = to_categorical(y_train, 10)
y_test = to_categorical(y_test, 10)
checkpointer = ModelCheckpoint(fs_tracker.get_model_directory() +
'/checkpoint.{epoch:02d}-{val_loss:.2f}.hdf')
tbcallback = TensorBoard(log_dir=fs_tracker.get_tensorboard_dir(),
histogram_freq=0,
write_graph=True,
write_images=False)
model.fit(x_train,
y_train,
validation_data=(x_test, y_test),
epochs=int(sys.argv[1]),
batch_size=batch_size * no_gpus,
callbacks=[checkpointer, tbcallback])
# Let's train the model with a TensorFlow optimizer:
mnist_data = input_data.read_data_sets('MNIST_data', one_hot=True)
global_step = tf.Variable(0, name='global_step', trainable=False)
train_step = tf.train.GradientDescentOptimizer(
0.5).minimize(loss, global_step=global_step)
# Initialize all variables
init_op = tf.global_variables_initializer()
saver = tf.train.Saver()
sess.run(init_op)
logger = logging.get_logger('train_mnist')
logger.setLevel(10)
# Run training loop
with sess.as_default():
while True:
batch = mnist_data.train.next_batch(50)
train_step.run(feed_dict={img: batch[0],
labels: batch[1]})
sys.stdout.flush()
saver.save(
sess,
os.path.join(
fs_tracker.get_model_directory(),
"ckpt"),
global_step=global_step)
time.sleep(1)
from keras.layers import Dense from keras.models import Sequential from studio import fs_tracker import os model = Sequential() model.add(Dense(2, input_shape=(2, ))) model.set_weights([[[2, 0], [0, 2]]]) model.save(os.path.join(fs_tracker.get_model_directory(), 'weights.h5'))
model.add(Dense(10, activation='softmax'))
model.summary()
batch_size = 128
no_epochs = int(sys.argv[1]) if len(sys.argv) > 1 else 10
lr = 0.01
print('learning rate = {}'.format(lr))
print('batch size = {}'.format(batch_size))
print('no_epochs = {}'.format(no_epochs))
model.compile(loss='categorical_crossentropy',
optimizer=optimizers.SGD(lr=lr),
metrics=['accuracy'])
print("Saving checkpoints to {}".format(fs_tracker.get_model_directory()))
checkpointer = ModelCheckpoint(fs_tracker.get_model_directory() +
'/checkpoint.{epoch:02d}-{val_loss:.2f}.hdf')
tbcallback = TensorBoard(log_dir=fs_tracker.get_tensorboard_dir(),
histogram_freq=0,
write_graph=True,
write_images=True)
model.fit(x_train,
y_train,
validation_data=(x_test, y_test),
epochs=no_epochs,
callbacks=[checkpointer, tbcallback],
batch_size=batch_size)
# We define the placeholder for the labels, and the loss function we will use:
labels = tf.placeholder(tf.float32, shape=(None, 10))
loss = tf.reduce_mean(categorical_crossentropy(labels, preds))
# Let's train the model with a TensorFlow optimizer:
mnist_data = input_data.read_data_sets('MNIST_data', one_hot=True)
global_step = tf.Variable(0, name='global_step', trainable=False)
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(
loss, global_step=global_step)
# Initialize all variables
init_op = tf.global_variables_initializer()
saver = tf.train.Saver()
sess.run(init_op)
logger = logging.getLogger('train_mnist')
logger.setLevel(10)
# Run training loop
with sess.as_default():
while True:
batch = mnist_data.train.next_batch(50)
train_step.run(feed_dict={img: batch[0], labels: batch[1]})
sys.stdout.flush()
saver.save(sess,
os.path.join(fs_tracker.get_model_directory(), "ckpt"),
global_step=global_step)
time.sleep(1)
