def main():
model = MnistModel()
optimizer = Optimizer()
data_source = MnistDataSource()
cfg = TrainConfig()
cfg.load(model,
optimizer,
data_source,
task_name="mnist",
framework_name='Lasagne')
saver = TrainSaver(cfg.prm['work_dir'],
cfg.prm['project_name'],
cfg.prm['model_filename_prefix'],
data_source,
task_name="mnist",
suffix="_ls")
trainer = Trainer(model=model,
optimizer=optimizer,
data_source=data_source,
saver=saver)
trainer.train(num_epoch=cfg.prm['max_num_epoch'],
epoch_tail=cfg.prm['min_num_epoch'])
def main():
model = MnistModel()
optimizer = Optimizer()
data_source = MnistDataSource()
cfg = TrainConfig()
cfg.load(model,
optimizer,
data_source,
task_name="mnist",
framework_name='Gluon')
saver = TrainSaver(cfg.prm['work_dir'],
cfg.prm['project_name'],
cfg.prm['model_filename_prefix'],
data_source,
task_name="mnist",
suffix="_gl")
#ctx = [mx.gpu(i) for i in cfg.prm['gpus']] if cfg.prm['gpus'] else mx.cpu()
ctx = mx.gpu(0) if cfg.prm['gpus'] else mx.cpu()
trainer = Trainer(model=model,
optimizer=optimizer,
data_source=data_source,
saver=saver,
ctx=ctx)
trainer.train(num_epoch=cfg.prm['max_num_epoch'],
epoch_tail=cfg.prm['min_num_epoch'])
def main():
args = parse_args()
model = MnistModel()
Converter.export_to_h5(
model=model,
checkpoint_path=os.path.join(args.checkpoint_dir, args.file_name),
dst_filepath=args.dst_filepath)
def main():
args = parse_args()
model = MnistModel()
Converter.import_from_h5(model=model,
src_filepath=args.src_filepath,
checkpoint_path=os.path.join(
args.checkpoint_dir, args.file_name))
def main():
args = parse_args(framework_name='TensorFlow')
model = MnistModel()
data_source = MnistDataSource(use_augmentation=False)
data_source.update_project_dirname(args.data_cache_dir)
Estimator.estimate(model=model,
data_source=data_source,
checkpoint_path=os.path.join(args.checkpoint_dir,
args.file_name))
def main():
args = parse_args()
model = MnistModel()
data_source = MnistDataSource(use_augmentation=False)
data_source.update_project_dirname(args.data_cache_dir)
ctx = [mx.gpu(i) for i in args.gpus] if args.gpus else mx.cpu()
Converter.import_from_h5(
model=model,
data_source=data_source,
src_filepath=args.src_filepath,
checkpoint_path=os.path.join(args.checkpoint_dir, args.prefix),
checkpoint_epoch=args.epoch,
ctx=ctx)
def main():
model = MnistModel()
optimizer = Optimizer()
data_source = MnistDataSource()
cfg = TrainConfig()
cfg.load(model,
optimizer,
data_source,
task_name="mnist",
framework_name='TFLearn')
saver = TrainSaver(cfg.prm['work_dir'],
cfg.prm['project_name'],
cfg.prm['model_filename_prefix'],
data_source=data_source,
task_name="mnist",
suffix="_tfl")
trainer = Trainer(model=model,
optimizer=optimizer,
data_source=data_source,
saver=saver)
# trainer.train(
# num_epoch=cfg.prm['max_num_epoch'],
# epoch_tail = cfg.prm['min_num_epoch'],
# dat_gaussian_blur_sigma_max=1.0,
# dat_gaussian_noise_sigma_max=0.05,
# dat_perspective_transform_max_pt_deviation=1,
# dat_max_scale_add=1.0 / (28.0 / 2),
# dat_max_translate=2.0,
# dat_rotate_max_angle_rad=0.2617994)
trainer.hyper_train(num_epoch=cfg.prm['max_num_epoch'],
epoch_tail=cfg.prm['min_num_epoch'],
bo_num_iter=cfg.prm['bo_num_iter'],
bo_kappa=cfg.prm['bo_kappa'],
bo_min_rand_num=cfg.prm['bo_min_rand_num'],
bo_results_filename='mnist_hyper.csv',
synch_file_list=cfg.prm['synch_list'],
sync_period=cfg.prm['sync_period'])
def main():
model = MnistModel()
optimizer = Optimizer()
data_source = MnistDataSource(use_augmentation=False)
cfg = TrainConfig()
cfg.load(
model,
optimizer,
data_source,
task_name="mnist",
framework_name='MXNet')
saver = TrainSaver(
cfg.prm['work_dir'],
cfg.prm['project_name'],
cfg.prm['model_filename_prefix'],
data_source,
task_name="mnist",
suffix="_mx")
ctx = [mx.gpu(i) for i in cfg.prm['gpus']] if cfg.prm['gpus'] else mx.cpu()
trainer = Trainer(
model=model,
optimizer=optimizer,
data_source=data_source,
saver=saver,
ctx=ctx)
trainer.hyper_train(num_epoch=cfg.prm['max_num_epoch'],
epoch_tail=cfg.prm['min_num_epoch'],
bo_num_iter=cfg.prm['bo_num_iter'],
bo_kappa=cfg.prm['bo_kappa'],
bo_min_rand_num=cfg.prm['bo_min_rand_num'],
bo_results_filename='mnist_hyper.csv',
synch_file_list=cfg.prm['synch_list'],
sync_period=cfg.prm['sync_period'])
os.path.join(os.path.dirname(os.path.abspath(__file__)), '..'))
import sys
sys.path.append(examples_dir)
from mnist_model import MnistModel
return MnistModel
if __name__ == '__main__':
parser = ArgumentParser(add_help=False)
parser.add_argument('--gpus', type=str, default=None)
MnistModel = import_model()
# give the module a chance to add own params
# good practice to define LightningModule speficic params in the module
parser = MnistModel.add_model_specific_args(parser)
# parse params
hparams = parser.parse_args()
# init module
model = MnistModel(hparams.batch_size, hparams.learning_rate)
# most basic trainer, uses good defaults
trainer = Trainer(
max_epochs=hparams.max_nb_epochs,
gpus=hparams.gpus,
val_check_interval=0.2,
logger=DAGsHubLogger(
), # This is the main point - use the DAGsHub logger!
default_root_dir='lightning_logs',
from mnist_model import MnistModel
if __name__ == '__main__':
mnist = MnistModel()
mnist.train()
return MnistModel
if __name__ == '__main__':
# Read parameters from a versioned file, which should also be a DVC dependency.
# This is the purest use case
hparams_from_file = read_hparams('params.yml')
# OPTIONAL:
# Allow some hyperparameters to be defined in the command line
parser = ArgumentParser(add_help=False)
parser.add_argument('--gpus', type=str, default=None, required=False)
# Parse args from command line, overriding params from file
hparams = parser.parse_args(namespace=hparams_from_file)
MnistModel = import_model()
# init module
model = MnistModel(hparams)
# most basic trainer, uses good defaults
trainer = Trainer(
max_nb_epochs=hparams.max_nb_epochs,
gpus=hparams.gpus,
val_check_interval=0.2,
logger=DAGsHubLogger(should_log_hparams=False), # This is the main point - use the DAGsHub logger!
default_save_path='lightning_logs',
)
trainer.fit(model)
from mnist_model import MnistModel
if __name__ == '__main__':
mnist = MnistModel()
mnist.gen()
mnist.show()
import tensorflow as tf
# import pandas as pd
import csv
import numpy as np
from mnist_model import MnistModel
from tensorflow.python.framework import ops
import matplotlib.pyplot as plt
data_size = 42000 # max 42k
test_size = 280 # max 28k
dim = 784
lambd = 0.01
# learning_rate = 0.001
mm = MnistModel()
train_dataset, train_labels = mm.load_csv_or_pickle("train.csv",
"train.pickle", data_size,
dim)
def split_data(X, Y, dev_size, test_size):
num_examples = X.shape[1]
train_size = num_examples - test_size - dev_size
permutation = list(np.random.permutation(num_examples))
shuffled_X = X[:, permutation]
shuffled_Y = Y[:, permutation]
X_train = shuffled_X[:, :train_size]
Y_train = shuffled_Y[:, :train_size]
X_dev = shuffled_X[:, train_size:train_size + dev_size]
Y_dev = shuffled_Y[:, train_size:train_size + dev_size]
X_test = shuffled_X[:, train_size + dev_size:]
os.path.join(os.path.dirname(os.path.abspath(__file__)), '..'))
import sys
sys.path.append(examples_dir)
from mnist_model import MnistModel
return MnistModel
if __name__ == '__main__':
parser = ArgumentParser(add_help=False)
parser.add_argument('--gpus', type=str, default=None)
MnistModel = import_model()
# give the module a chance to add own params
# good practice to define LightningModule speficic params in the module
parser = MnistModel.add_model_specific_args(parser)
# parse params
hparams = parser.parse_args()
# init module
model = MnistModel(hparams)
# most basic trainer, uses good defaults
trainer = Trainer(
max_nb_epochs=hparams.max_nb_epochs,
gpus=hparams.gpus,
val_check_interval=0.2,
logger=DAGsHubLogger(
), # This is the main point - use the DAGsHub logger!
default_save_path='lightning_logs',