• Introduction
• Imports
• Sections
  • Hierarchical nesting
  • Replication
  • Commands
  • Applications
• Placeholders

Paramparse is a lightweight argparse wrapper to allow hierarchically nested class-based parameters suitable for automatic code analysis / intellisense in advanced IDEs like Pycharm.
It also provides a unified parameter specification protocol that can be used to provide parameter values through both text files and command line.

Please refer to the included parameter skeleton of a large multi object tracking project for which the functionality included in this package was originally developed. It provides an excellent example of a highly modular project with deeply nested and shared modules.

For example, this is one of the deeper instances of module nesting in this example:

Params->Trainer->Target->Templates->Siamese->SiamFC->DesignParams

Parameter for this configuration can be provided as:

trainer.target.templates.siamese.siamfc.design.exemplar_sz=150

Specifying multiple parameters for a deeply nested module can quickly become cumbersome especially from command line. The package thus provides a way to group parameters from the same module using the @ identifier. An example is provided in example/cfg/params.cfg. Note that the indentation used in this file is only for ease of human parsing and is not needed as this system of grouping also works from command line. Example commands are in example/commands.md.

The @ identifier specifies a prefix pf to be added to all subsequent arguments so that arg_name is then treated as pf.arg_name.
Assuming pf=arg1.arg2, following flavors of @ identifier usage are supported:

Usage of the package is very simple and involves calling paramparse.process as demonstrated in example/main.py.

More detailed usage examples are available in Deep MDP for which this parser was originally designed
Another repo that uses it extensively is IPSC Prediction.

It also provides three converter functions from_parser, from_dict and from_function that can create a parameter class compatible with this package from existing parameters in argparse.ArgumentParser and dict formats or using a function's keyword arguments respectively. The generated class code is either writen to a python source file whose name can be specified as the second argument (defaults to Params.py) or copied to clipboard if to_clipboard=1 is provided (requires pyperclip).

The process function does type inference from the default value of each param but also supports extracting the type from restructuredText/pycharm type docstring (as generated by the converter functions) if it is provided.

Note : paramparse uses the reserved parameter cfg to specify paths to text files containing parameter values. If an existing argparse or dict object to be converted into paramparse class already has a cfg field used for some other purpose, it will conflict with the parser so please rename this field before or after converting but before running paramparse.process.

Usage of converter functions is demonstrated in example/utils_demo.py

Run python3 main.py --h from the example folder to see the hierarchical help documentation generated by argparse.

Apart from the hierarchical nesting and parameter grouping, an important utility of paramparse is in the class based representation that allows automated code analysis, navigation and refactoring in IDEs like Pycharm that is not possible when using vanilla argparse.ArgumentParser or dict.

A cfg can be imported from within another using the %import% keyword followed by the relative path of that cfg with respect to the importing cfg.

For example, the line

%import% dir3/imported.cfg

in dir1/dir2/importing.cfg will import dir1/dir2/dir3/imported.cfg.

Importing a CFG is equivalent to copying and pasting all the lines from the imported CFG into the importing CFG at the place where the %import% line is.

Recursive imports are supported and circular imports will raise an assertion error.

Sections provide a means to read a cfg file selectively rather than in its entirety.
Each section is a named group of consecutive lines within a cfg file that is preceded and succeeded by a line starting with two or more #:

• preceding line specifies the name of the section
• succeeding line can either specify the name of the next section or simply mark the end of the current section if it is nameless
• lines without an explicit section are considered to be in an implicit common section that is always read while parsing a cfg (in addition to the specified sections)

Sections can be nested to any number of hierarchical levels

• level at which a section lies is determined by the number of # preceding its name
  • ## denotes level 1, ### denotes level 2 and so on
  • lines starting with a single # are regarded as comments and ignored
• each section at level 2 or greater (i.e. with 3 or more # preceding its name) is considered to be nested within the nearest section before it that is one level above it
• a section can have multiple children
• a section can only have a single parent
• a section can be included in a command only if all its ancestors have also been included
• sections nested within different hierarchies in the same cfg file can have identical names

Multiple sections can be started in a single line by specifying their names separated by commas

• all lines in such a multi-name section (inluding any section hierarchies therein) are replicated for each one of the names
• this can reduce redundancy when specifying sections that differ only in ways that can be derived from their names using placeholders
• long lists of numeric names can be specified succinctly using range() and irange() placeholders that can be used with 1, 2 or 3 arguments similar to the python range function, with irange being inclusive with respect to its upper bound
  • multiple disjointed ranges can be joined with each other as well as with manually specified lists using + operator, e,g. range(3, 6)+irange(10, 13)+21,22 = 3,4,5,10,11,12,13,21,22

Sections are specified in a command by following the name of the cfg file with a colon (:) and a list of section names also separated with colons

• name of a parent section should precede that of its children
• if the cfg file has multiple identically-named sections in different hierarchies and only one of them is to be read (while reading all ancestors of more than one of them), its name should be separated from its parent by a hyphen (-) instead of a :
• if a cfg file needs to be read again with a different section hierarchy, name of the root section of this hierarchy can be prefixed by ++ or +++
  • ++ causes the common sections to be read again in addition to the new section hierarchy
  • +++ causes the common sections to be be skipped so that only the new section hierarchy is read

Sections can have many uses:

• each section (or hierarchy thereof) can be used to specify the parameter values needed to run the script in a given configuration (or variants thereof)
• sections can be given short and intuitive names to quickly figure out exactly which configuration any given command corresponds to
• sections can make it very easy to switch between configurations with minimal changes to the commands
• hierarchical nesting of sections makes the relations between the different configurations explicit and easier to keep track of
• placeholders allow the names and values of parameters within a given section to be derived from the names of the section and its parents.

Placeholders are reserved keywords that or variables will be replaced by their respective texts while parsing a cfg file. Paramparse supports the following placeholders: