Implementation:Interpretml Interpret Debug Utilities

Knowledge Sources	Interpretml Interpret
Domains	Machine_Learning, Interpretability
Last Updated	2026-02-07 12:00 GMT

Overview

The develop module provides debug and development utilities for the InterpretML package, including functions for enabling debug mode, collecting system diagnostics, configuring logging, and managing internal development options.

Description

This module serves as the development/debugging interface for the InterpretML package:

• debug_mode: Activates package-wide debug mode. It registers a log file, writes system diagnostics, and loads native libraries in debug mode. Can only be called once per session.

• register_log: Registers a logging handler that writes to a file or stream (such as sys.stderr or sys.stdout). Uses WatchedFileHandler for file-based logging with a formatted output including timestamp, filename, line number, and function name.

• debug_info: Collects comprehensive debug information including the package version, show server status, static system info, and dynamic system info. Returns a dictionary.

• print_debug_info: Pretty-prints the debug information dictionary as formatted JSON.

• static_system_info: Gathers machine architecture, OS version, Python version, CPU count, and memory totals using the platform and psutil packages.

• dynamic_system_info: Gathers runtime metrics including CPU utilization, CPU frequency, virtual memory usage, and swap memory usage.

• get_option / set_option: Get and set internal development options stored in the _develop_options dictionary. These options control advanced EBM training behavior such as intercept rounds, learning rate scaling, categorical handling, acceleration flags, and boosting options.

Usage

Use debug_mode when diagnosing issues with InterpretML or when the package authors request diagnostic information. Use register_log to capture detailed logs during model training. Use get_option/set_option for advanced tuning of internal EBM development parameters (not recommended for general use).

Code Reference

Source Location

• Repository: Interpretml_Interpret
• File: python/interpret-core/interpret/develop.py

Signature

def debug_mode(log_filename="log.txt", log_level="INFO", native_debug=True):

def register_log(filename, level="DEBUG"):

def debug_info():

def print_debug_info(file=None):

def static_system_info():

def dynamic_system_info():

def get_option(name):

def set_option(name, value):

Import

from interpret.develop import debug_mode, register_log, debug_info, print_debug_info
from interpret.develop import get_option, set_option

I/O Contract

debug_mode

Name	Type	Required	Description
log_filename	str or stream	No	Filepath to log to, or sys.stderr/sys.stdout (default "log.txt")
log_level	str	No	Logging level such as "INFO" or "DEBUG" (default "INFO")
native_debug	bool	No	Whether to load debug versions of native libraries (default True)

Name	Type	Description
handler	logging.Handler	The logging handler that was registered

register_log

Name	Type	Required	Description
filename	str or stream	Yes	Filepath to log to, or sys.stderr/sys.stdout
level	str	No	Logging level (default "DEBUG")

Name	Type	Description
handler	logging.Handler	The logging handler that was registered

debug_info

Name	Type	Description
debug_dict	dict	Dictionary containing version, server status, static and dynamic system info

get_option / set_option

Name	Type	Required	Description
name	str	Yes	Name of the development option
value	varies	Yes (set_option only)	Value to set for the option

Development Options

The module exposes the following internal options via get_option/set_option:

Option	Default	Description
n_intercept_rounds_initial	25	Initial intercept calibration rounds
n_intercept_rounds_final	100	Final intercept calibration rounds
intercept_learning_rate	0.25	Learning rate for intercept updates
cat_l2	0.0	L2 regularization for categorical features
learning_rate_scale	1.0	Scaling factor for the learning rate
allow_float_interactions	False	Whether to allow floating-point interaction terms
purify_boosting	False	Whether to purify during boosting
purify_result	False	Whether to purify the final result
randomize_initial_feature_order	True	Randomize initial feature ordering
randomize_greedy_feature_order	True	Randomize feature order during greedy selection

Usage Examples

Enable Debug Mode

from interpret.develop import debug_mode

handler = debug_mode(log_filename="interpret_debug.txt", log_level="DEBUG")

Print System Diagnostics

from interpret.develop import print_debug_info

print_debug_info()

Register Custom Log Handler

import sys
from interpret.develop import register_log

handler = register_log(sys.stderr, level="INFO")

Related Pages

• Principle:Interpretml_Interpret_Debug_And_Logging