cliconfig.processing.builtin
Built-in processing classes.
Built-in classes to apply pre-merge, post-merge, pre-save and post-load modifications
to dict with processing routines process_routines
.
They are the default processing used by the config routines load_config()
and make_config()
.
- class ProcessMerge
Merge dicts just in time with
@merge_after/_before/_add
tags.Tag your key with
@merge_after
,@merge_before
or@merge_add
to load the config corresponding to the value (which is a yaml path) and merge it just before or after the current config. The merged dicts will be processed with pre-merge but not post-merge to ensure that merged configurations are recursively processed with pre-merge before applying a post-merge processing. It allows the merged configs to make references to each other (typically for copy) even without containing the merge tags itself.‘@merge_add’ merges the dict corresponding to the path by allowing ONLY new keys It is a security check when you want to add a dict completely new, the typical usage for a default config splitted in several files.
‘@merge_after’ merge the dict corresponding to the path on the current dict
‘@merge_before’ merge the current dict on the dict corresponding to the path
The processing is a pre-merge processing only and occurs before almost all of the other processings. Pre-merge order: -20.0
Examples
# config1.yaml a: b: 1 b_path@merge_after: dict2.yaml # config2.yaml a.b: 2 c_path@merge_add: config3.yaml # config3.yaml c: 3
Before merging, the config1 is interpreted as the dict:
{'a': {'b': 2, 'b_path': 'config2.yaml'}, 'c_path': 'config3.yaml', 'c': 3}
If you replace ‘@merge_after’ by ‘@merge_before’, it will be:
{'a': {'b': 1, 'b_path': 'config2.yaml'}, 'c_path': 'config3.yaml', 'c': 3}
Finally, if you replace
@merge_after
by@merge_add
, it will raises an error because the keya.b
already exists in the dict.
- class ProcessCopy
Copy a value with
@copy
tag. The copy is protected from direct updates.Tag your key with
@copy
and with value the name of the flat key to copy. The pre-merge processing removes the tag. The post-merge processing sets the value (if the copied key exists). The end-build processing checks that the key to copy exists and copy them. The pre-save processing restore the tag and the key to copy to keep the information on future loads. The post-merge and the end-build processings occur after most processings to allow the user to modify or add the copied key before the copy. Pre-merge order: 0.0 Post-merge order: 10.0 End-build order: 10.0 Pre-save order: 0.0Examples
# config.yaml a: b: 1 c@copy: a.b
Before merging, the config is interpreted as the dict
{'a': {'b': 1, 'c': 1}}
Note
The copy key is protected against any modification and will raise an error if you try to modify it but will be updated if the copied key is updated.
If the key to copy does not exist in the config on post-merge, no error is raised to let the user the possibility to add the key later via merge. However, if the key still does not exist at the end of the build (and the key was never copied), an error is raised.
- class ProcessDef
Dynamically define a value from math expression with
@def
tag.The expression can contain any parameter name of the configuration. The most usefull operators and built-in functions are supported, the random and math packages are also supported as well as some (safe) numpy, jax, tensorflow, pytorch functions. If/else statements and comprehension lists are also supported.
The pre-merge processing removes the tag. The post-merge processing sets the value while the presave processing restore the tag and the expression. The post-merge processing occurs after most processings to allow the user to modify the used keys before the calculation. Pre-merge order: 0.0 Post-merge order: 10.0 Pre-save order: 0.0
Examples
# config.yaml a: b: 1 c: 2 d@def: "(a.b + a.c) * 2 > 5"
Before merging, the config is interpreted as the dict
{'a': {'b': 1, 'c': 2}, 'd': True}
Now the parameter d is automatically updated if a.b or a.c changes while also remaining editable by it-self.
Note
Unlike @copy processing you can change the value by setting an other value or an other definition with @def.
Unlike copy processing all the keys used in expression must be in the config at post-merge.
This processing does not use
eval
and is therefore safe from malicious code.
- class ProcessTyping
Try to convert and force a type with
@type:<mytype>
tag.The type is forced forever. Allow basic types (none, any, bool, int, float, str, list, dict), nested lists, nested dicts, unions (with Union or the ‘|’ symbol) and Optional. The type description is lowercased and spaces are removed.
For instance:
@type:None|List[Dict[str, int|float]]
is valid and force the type to be None or a list containing dicts with str keys and int or float values.The processing stores the type in pre-merge and convert/check alls forced types on end-build. It restore the tag in pre-save to keep the information on future loads. The end-build processing occurs after almost all processings. Pre-merge order: 0.0 End-build order: 20.0 Pre-save order: 0.0
Note
The conversion into union type is from left to right. For instance,
param@type:List[str|float]: [True]
is converted to["True"]
.The type is not checked on pre-merge or post-merge to allow the parameter to be updated (by a copy or a merge for instance). The goal of this processing is to ensure the type at the end of the build.
Examples
in_dict = {"param@type:None|List[int|float]": None} dict1 = {param: [0, 1, 2.0]} # no error dict2 = {param: [0, 1, 2.0, 'a']} # error dict3 = {param: [0, 1, "2"]} # no error but convert to [0, 1, 2]
Merging configs with dictionaries
in_dict
anddict1
raises no error andparam
is forced to be None or a list of int or float forever. Merging config within_dict
anddict2
raises an error on post-merge due to the ‘a’ value (which is a string). Merging config within_dict
anddict3
raises no error and convert the value to [0, 1, 2].Note that removing “None|” in the type description of
param
still doesn’t raise an error in those cases because the type checking is evaluated after the merge withdict2
.
- class ProcessSelect
Select a sub-config with
@select
and delete the rest of its parent config.First look in pre-merge for a parameter tagged with
@select
containing a flat key corresponding to a sub-configurations to keep. The parent configuration is then deleted on post-merge, except the selected sub-configuration and eventually the tagged parameter (if it is in the same sub-configuration). It is also possible to select multiple keys of a same sub-configuration (meaning that the part before the last dot must be equal) by passing a list of flat keys. Pre-merge order: 0.0 Post-merge order: 0.0Examples
models: model_names@select: [models.model1, models.model3] model1: param1: 1 param2: 2 model2: param1: 3 param2: 4 model3: submodel: param: 5 model4: param: 6
Result in deleting
models.model2
(param1
andparam2
) andmodels.model4.param
, and keeping the rest.Warning
For security reasons, this processing prevents from deleting the configuration at the root, which is the case when the selected key doesn’t contain a dot. It raises an error in this case.
- class ProcessDelete
Delete the sub-configs/parameters tagged with
@delete
on pre-merge.This processing is useful to activate a processing without adding an additional parameter in the default configuration to avoid the error on merge with
allow_new_keys=False
. This processing is applied very late on pre-merge to allow the others processing to be applied before deleting the parameters. Pre-merge order: 30.0Examples
# main.yaml 1@select@delete: configs.config1 2@merge_add@delete: config1.yaml 3@merge_add@delete: config2.yaml # config1.yaml configs.config1.param: 1 configs.config1.param2: 2 # config2.yaml configs.config2.param: 3 configs.config2.param: 4
Here we want to merge two config files and select one sub-config. We use the corresponding tags but we don’t have a good name for the keys and instead of adding a new parameter in the default configuration with random names like “1”, “2”, “3”, we use the
@delete
tag to delete the keys after the pre-merge processing.Warning
The sub-config/parameter is deleted on pre-merge. Therefore, if the parameter also exists on the other configuration during merge (without the tag), this parameter will be remain as it is. This processing is more used to delete parameter that is NOT present in the default configuration.
- class ProcessNew
Allow new sub-config/parameter absent from default config with tag
@new
.The tagged sub-config/parameter and its value is stored during pre-merge and is deleted to avoid error on merge due to new key. It is then restored on post-merge. Finally, the tag is restored on pre-save for further loading. This processing is applied very late on pre-merge to allow the others processing to be applied before deleting the parameters. The post-merge processing is applied very early to allow the other processing to use this new parameter. Pre-merge order: 30.0 Post-merge order: -20.0 Pre-save order: 0.0
Disclaimer: It is preferable to have an exhaustive default configuration instead of abusing this processing to improve the readability and to avoid typos errors in the name of the parameters or their sub-configs.
Examples
# default.yaml param1: 1 # additional.yaml param2@new: 2 subconfig@new.subsubconfig: param3: 3 param4: 4
Use default.yaml as default configuration and add additional.yaml as additional configuration via CLI results on the configuration containing param1, param2 and the nested config containing param3 and param4. Without the
@new
tag, an error is raised because param2 is not present in the default configuration.Note
Tag a subconfig by adding
@new
at the end of the key containing the sub-config dict in your yaml file.When a parameter is added with this processing, it is possible to modify it later via config merge without the tag because the parameter is then present in the current configuration.
If the tagged parameter or sub-config is already present in the current configuration, no error are raised and the value is still updated on post-merge. It may no have influence in practice.
- class ProcessDict
Declare a dict instead of a sub-config with
@dict
tag.This processing can be used to declare a dict where the keys are not known in advance or will be modified. New keys are allowed in each merge and the element are still available using the dot notation like
config.subconfig.mydict.something
. The pre-merge processing remove the tag and convert the dict to a wrapped dict to prevent the flattening. The end-build processing unwrap the dicts to a normal dict. The pre-save processing restore the tag to keep the information on future loads. Pre-merge order: -30.0 End-build order: 0.0 Pre-save order: -30.0Examples
# default.yaml param1: 0 param2: 2 sweep@dict: None # additional1.yaml sweep@dict: metric: {name: accuracy, goal: maximize} method: bayes parameters: param1: {min: 0, max: 50} # additional2.yaml sweep@dict: name: "random sweep" method: random parameters: param2: {min: 0, max: 10}
The
swep
parameter is considered as a single dict object and not as a sub-config for merging.Warning
Processings are not applied in the dict keys. In particular, the tags are not used and not removed.
The tag
@dict
must be added at the key containing the dict every time you want to modify the dict.
- class PseudoDict(dict_)
Object containing a dict that dodges flattening.
- class ProcessCheckTags
Raise an error if a tag is present in a key after pre-merging processes.
This security processing is always applied after all pre-merge process and checks for ‘@’ in the keys. It raises an error if one is found.
- class DefaultProcessings
Default list of built-in processings.
To add these processings to a Config instance, use:
config.process_list += DefaultProcessings().list
- The current default processing list contains:
ProcessCheckTags: protect against ‘@’ in keys at the end of pre-merge)
ProcessMerge (@merge_all, @merge_before, @merge_after): merge multiple files into one.
ProcessCopy (@copy): persistently copy a value from one key to an other and protect it
ProcessTyping (@type:X): force the type of parameter to any type X.
ProcessSelect (@select): select sub-config(s) to keep and delete the other sub-configs in the same parent config.
ProcessDelete (@delete): delete the parameter tagged with @delete on pre-merge.