Config

class tinder.config.Placeholder[source]

A placeholder to denote the required parameter without default.

Example:

config = {
    'lr': 0.01,
    'gpu': Placeholder.STR
}
tinder.config.override(config)
tinder.config.bootstrap(*, logger_name='tinder', trace=True, pdb_on_error=True)[source]

Setup convinient utils to run a python script for deep learning.

Sets the ‘CUDA_DEVICE_ORDER’ environment variable to ‘PCI_BUS_ID’

Examples

./a.py lr=0.01 gpu=0,1,2

The following environments are set:

  • os.environ[‘CUDA_VISIBLE_DEVICES’] = ‘0,1,2’

  • os.environ[‘CUDA_DEVICE_ORDER’] = ‘PCI_BUS_ID’

  • os.environ[‘lr’] = ‘0.01’

Parameters

  • logger_name – setup standard python logger that is compatiable with tqdm

  • parse_args – set environment variables from command line arguments. gpu is alias for CUDA_DEVICE_DEVICES.

  • trace – use backtrace module to print stacktrace

  • pdb_on_error – enter pdb shell when an exception is raised

tinder.config.override(config)[source]

Update your config dict with command line arguments.

The original dictionary is the default values. To prevent mistakes, any command line argument not specified in the dictionary raises.

gpu is a special key. If gpu is found, os.environ[‘CUDA_VISIBLE_DEVICES’] is set accordingly.

Example:

config = {
    'lr': 0.01,
    'gpu': Placeholder.STR
}
tinder.config.override(config)

// shell
python script.py lr=0.005 gpu=0,1,2
Parameters

config (dict or SimpleNamspace) – parameter specs

Raises