Config

class tfsnippet.utils.Config

Bases: object

Base class for defining config values.

Derive sub-classes of Config, and define config values as public class attributes. For example:

class YourConfig(Config):
    max_epoch = 100
    learning_rate = 0.01
    activation = ConfigField(str, default='leaky_relu',
                             choices=['sigmoid', 'relu', 'leaky_relu'])
    l2_regularization = ConfigField(float, default=None)

config = YourConfig()

When an attribute is assigned, it will be validated by:

1. If the attribute is defined as a ConfigField, then its validate() will be used to validate the value.
2. If the attribute is not None, and a validator is registered for type(value), then an instance of this type of validator will be used to validate the value.
3. Otherwise if the attribute is not defined, or is None, then no validation will be performed.

For example:

config.l2_regularization = 5e-4  # okay
config.l2_regularization = 'xxx'  # raise an error
config.activation = 'sigmoid'  # okay
config.activation = 'tanh'  # raise an error
config.new_attribute = 'yyy'  # okay

The config object also implements dict-like interface:

# to test whether a key exists
print(key in config)

# to iterate through all config values
for key in config:
    print(key, config[key])

# to set a config value
config[key] = value

You may get all the config values of a config object as dict:

print(config_to_dict(config))

Or you may get the default values of the config class as dict:

print(config_defaults(YourConfig))
print(config_defaults(config))  # same as the above line

Methods Summary

to_dict()	Get the config values as a dict.
update(key_values)	Update the config values from key_values.

Methods Documentation

to_dict()

Get the config values as a dict.

Returns:The config values. Return type:dict[str, any]

update(key_values)

Update the config values from key_values.

Parameters:key_values – A dict, or a sequence of (key, value) pairs.