Implementation:Kserve Kserve Alibi Parser

Knowledge Sources	Kserve_Kserve
Domains	Explainability, CLI Configuration
Last Updated	2026-02-13 00:00 GMT

Overview

Concrete tool for providing comprehensive command-line argument parsing for the Alibi explainer server with algorithm-specific configuration provided by the KServe sample code.

Description

This module defines the argument parsing infrastructure for the Alibi explainer server. It contains:

• GroupedAction -- A custom argparse.Action subclass that groups arguments into namespaces using dot-separated destination names (e.g., explainer.threshold), enabling hierarchical argument organization.
• str2bool() -- A utility function that converts string values ("yes", "true", "1", etc.) to boolean values, used for boolean command-line arguments.
• addCommonParserArgs() -- A function that adds common anchor explainer parameters (threshold, delta, tau, batch_size, coverage_samples, beam_size, stop_on_first, max_anchor_size, max_samples_start, n_covered_ex, binary_cache_size, cache_margin, verbose, verbose_every) to any subparser.
• parse_args() -- The main parsing function that creates subparsers for each ExplainerMethod (AnchorTabular, AnchorText, AnchorImages) with method-specific arguments (e.g., use_unk, sample_proba, temperature for text; p_sample for images), plus common arguments for model name, predictor host, and storage URI.

Usage

Use parse_args() as the entry point for configuring and launching the Alibi explainer container, passing in sys.argv to obtain the parsed arguments and extra explainer-specific configuration.

Code Reference

Source Location

• Repository: Kserve_Kserve
• File: docs/samples/explanation/alibi/alibiexplainer/alibiexplainer/parser.py
• Lines: 1-216

Signature

class GroupedAction(argparse.Action):
    def __call__(self, theparser, namespace, values, option_string=None):
        ...


def str2bool(v):
    ...


def addCommonParserArgs(parser):
    ...


def parse_args(sys_args):
    ...

Import

from alibiexplainer.parser import parse_args, GroupedAction, str2bool

I/O Contract

Inputs

GroupedAction.__call__()

Name	Type	Required	Description
theparser	argparse.ArgumentParser	Yes	The argument parser instance
namespace	argparse.Namespace	Yes	The current namespace being populated
values	Any	Yes	The parsed value for the argument
option_string	str	No	The option string that triggered this action

str2bool()

Name	Type	Required	Description
v	Union[bool, str]	Yes	A value to convert to boolean (accepts "yes", "true", "t", "y", "1" and their negatives)

parse_args()

Name	Type	Required	Description
sys_args	list	Yes	Command-line arguments list (typically sys.argv[1:])

Outputs

str2bool()

Name	Type	Description
result	bool	The boolean interpretation of the input string

parse_args()

Name	Type	Description
args	argparse.Namespace	Parsed arguments including model_name, predictor_host, storage_uri, and command
extra	dict	Dictionary of explainer-specific parameters extracted from the grouped namespace

Usage Examples

Basic Usage

from alibiexplainer.parser import parse_args

# Parse command-line arguments for an AnchorTabular explainer
args, extra = parse_args([
    "--model_name", "income-explainer",
    "--predictor_host", "income-predictor.default.svc.cluster.local",
    "--storage_uri", "gs://my-bucket/explainer",
    "AnchorTabular",
    "--threshold", "0.95",
    "--batch_size", "100",
])

print(args.model_name)    # "income-explainer"
print(args.command)        # "AnchorTabular"
print(extra)               # {"threshold": 0.95, "batch_size": 100}

Related Pages

• Environment:Kserve_Kserve_Python_Runtime