config_argparse
Config library based on standard ArgumentParser.
Usage
Simple:
from config_argparse import Config
class MyConfig(Config):
weight = 10.0 # -> .add_argument('--weight', type=float, default=10.0)
layers = [10, 20, 30] # -> .add_argument('--layers', type=int, nargs='+', default=[10, 20, 30])
MyConfig()
# MyConfig:
# weight = 10.0
# layers = [10, 20, 30]
MyConfig().parse_args(['--weight', '30.5', '--layers', '1', '2'])
# MyConfig:
# weight = 30.5
# layers = [1, 2]
MyConfig().parse_args(['--help'])
# usage: MyConfig [-h] [--weight WEIGHT] [--layers LAYERS [LAYERS ...]]
#
# optional arguments:
# -h, --help show this help message and exit
# --weight WEIGHT
# --layers LAYERS [LAYERS ...]You can use int, float, str, bool, and list of them.
config_argparse.Config automatically generates argparse.ArgumentParser internally and parse arguments.
All the values defined as class variables are copied to the instance by copy.deepcopy.
Nested:
This library supports nested config (it generates multiple ArgumentParser internally).
from config_argparse import Config
class Nest(Config):
a = 10
class MyConfig(Config):
a = 'str'
nest = Nest() # -> .add_argument('--nest.a', type=int, default=10, dest=`a in nest`)
MyConfig()
# MyConfig:
# a = str
# nest = Nest:
# a = 10
MyConfig().parse_args(['--nest.a', '33'])
# MyConfig:
# a = str
# nest = Nest:
# a = 33
MyConfig().parse_args(['--help'])
# usage: MyConfig [-h] [--a A] [--nest]
#
# optional arguments:
# -h, --help show this help message and exit
# --a A
# --nest
MyConfig().parse_args(['--nest.help'])
# usage: Nest [-h] [--nest.a A]
#
# optional arguments:
# -h, --help show this help message and exit
# --nest.a AInherited:
You can merge multiple configs by inheriting them.
from config_argparse import Config
class ParentA(Config):
a = 10
class ParentB(Config):
b = 10
class MyConfig(ParentA, ParentB):
c = 'str'
MyConfig()
# MyConfig:
# a = 10
# b = 10
# c = str
MyConfig().parse_args(['--a', '33'])
# MyConfig:
# a = 33
# b = 10
# c = strDynamic:
You can dynamically build nested configs according to the values in the parent.
from config_argparse import Config, Value, DynamicConfig
class ConfigA(Config):
a = 1
class ConfigB(Config):
b = 1
config_map = {
'a': ConfigA,
'b': ConfigB,
}
class MyConfig(Config):
lr = 0.1
epoch = 100
model = Value('a', choices=list(config_map.keys()))
model_cfg = DynamicConfig(lambda c: config_map[c.model]()) # pass function which returns instance of Config
args = MyConfig()
print(args)
# MyConfig:
# lr = 0.1
# epoch = 100
# model = a
# model_cfg = None # default value of DynamicConfig is None (if auto_load=False)
args = MyConfig().parse_args([])
print(args)
# MyConfig:
# lr = 0.1
# epoch = 100
# model = a
# model_cfg = ConfigA:
# a = 1
args = MyConfig().parse_args(['--lr', '1', '--model', 'b', '--model_cfg.b', '100'])
print(args)
# MyConfig:
# lr = 1.0
# epoch = 100
# model = b
# model_cfg = ConfigB:
# b = 100DynamicConfig also supports list of Configs.
class ConfigA(Config):
a = 1
class MyConfig(Config):
models = ['a', 'b', 'c']
# pass function which returns list of Config
models_cfg = DynamicConfig(lambda c: list(map(lambda x: ConfigA(), c.models)))
args = MyConfig()
print(args)
# MyConfig:
# models = ['a', 'b', 'c']
# models_cfg = None
print(args.parse_args([]))
# MyConfig:
# models = ['a', 'b', 'c']
# models_cfg = [ConfigA:
# a = 1, ConfigA:
# a = 1, ConfigA:
# a = 1]Value:
You can use some ArgumentParser's features such as required, choices by config_argparse.Value.
from config_argparse import Config, Value
class MyConfig(Config):
weight = Value(type=float, required=True)
dim = Value('a', choices=['a', 'b', 'c'])
MyConfig().parse_args([])
# MyConfig: error: the following arguments are required: --weight
MyConfig().parse_args(['--weight', '1.0', '--dim', 'foo'])
# MyConfig: error: argument --dim: invalid choice: 'foo' (choose from 'a', 'b', 'c')Convert to/from dict
Default values can be set by passing dict like object to Config::__init__.
Config::todict creates serializable dict (if the config is already parsed).
import json
from config_argparse import Config
class MyConfig(Config):
weight = 10.0 # -> .add_argument('--weight', type=float, default=10.0)
layers = [10, 20, 30] # -> .add_argument('--layers', type=int, nargs='+', default=[10, 20, 30])
c = MyConfig({'weight': 1.0})
c.weight
# -> 1.0
c.todict()
# -> {'weight': 1.0, 'layers': [10, 20, 30]}
