deep learning with multi gpu on some framework

Run neural network on multi gpu

Requiremets

• chainer==1.23.0
• conda3==4.3.8
• keras==2.0.3
• mxnet-cu80==0.9.5
• scikit-learn==0.18.1
• tensorflow-gpu==1.0.1

How to use

dataset = 'cifar10'
fm = 'tf'

# download dataset
cifar = cifar10.py if dataset == 'cifar10' else cifar100.py
python cifar

# run training
framework = {'tf': tf_pure_cifar.py, 'tfslim': tf_slim_learn_cifar.py,
             'chainer': chainer_cifar.py, 'keras': keras_cifar.py}
python framework[fm]

Code

• chainer_cifar.py => run chainer code
• chainer_model.py => note neural net model
• cifar10.py => download cifar10 datset and make it dict
• cifar100.py => download cifar100 datset and make it dict
• keras_cifar.py => run chainer code
• keras_make_parallel.py => note multi gpu processing
• keras_model.py => note neural net model
• ln_cifar_dataset.py => make synbolic link about cifar dataset
• run_experiment.py => run some test to observe gpu trianing on some env
• tf_pure_cifar.py => run neural net model with tensorflow contrib API
• tf_pure_datafeeder.py => note datafeeder(this is future work)
• tf_pure_model.py => note neural net model with the tensorflow API
• tf_pure_trainer.py => note trainer on single and multi gpu(can't run multi one)