API Reference

The reference pages list the public classes and functions of each subpackage with their signatures, defaults, and docstrings. They are generated from the SMF source by mkdocstrings, so they describe the source the site was built from (0.9.1, see the release notes).

Docstrings

Parameter names, order, and defaults come straight from the code. Every public class, function, and method documents all of its parameters in the NumPy docstring layout (Parameters, Returns, Raises), and a test in the pytest suite (test_public_docstring_parameters.py) keeps the docstrings in step with the signatures. The User Guides explain the behavior in more depth, with runnable examples.

Subpackages

Import path	Reference page	Main contents
Modeling_Tool.Core	Core	Binning, model registry (save_model, load_model), sample-weight helpers, ODPS client, parallel apply engine, ProcCompare, encryption, JSON conversion, utilities
Modeling_Tool.WOE	WOE	WOE_Master, MonotoneWOEBinner, WOE transforms, WOE plots, engine adapter
Modeling_Tool.Feature	Feature	PSI, IV/KS insights, correlation filter, feature_screen, distribution analysis
Modeling_Tool.Model	Model	LRMaster, GradientBoostingModel (LightGBM, XGBoost, CatBoost), backward elimination
Modeling_Tool.Eval	Eval	Gains tables, PerformanceEvaluator, score comparison, ROC/KS/PR/lift functions and plots
Modeling_Tool.Sample	Sample	Splitting, sampling, balancing, reject inference, distribution adaptation
Modeling_Tool.Explainability	Explainability	ModelExplainer (SHAP, Owen value, PDP, ICE, ALE, LIME), coalition structure
Modeling_Tool.UAT	UAT	UATConsistencyChecker, UATConfig, comparison helpers
Modeling_Tool.Pipeline	Pipeline	The seven one-click pipelines with their config and result classes, config schema helpers
ExcelMaster	ExcelMaster	Cursor-based Excel writer, preset formats, report templates
Report	Report	Report_Tool: performance, WOE plot, and variable-importance sheets

Top-Level Names

Modeling_Tool.__all__ lists 123 names, and you can import any of them directly:

from Modeling_Tool import WOE_Master, GradientBoostingModel, PerformanceEvaluator

Area	Names
Samples	SampleSplitter, StratifiedSampler, SampleBalancer, select_sample_seed, RejectInferenceFactory, RejectInferrer, ParcelingInferrer, HardCutoffInferrer, FuzzyAugmentInferrer, SimpleAugmentInferrer, DistributionAdaptation
WOE	WOE_Master, MonotoneWOEBinner, WOEEngineAdapter, as_woe_engine, is_monotonic, woe_transform, woe_transformation, plot_woe, save_mapping_table, load_mapping_table, get_overall_woe_table, SMF_MISSING_BIN
Feature screening	PSICalculator, calculate_psi_within_dataset, VarExtractionInsights, CorrelationFilter, feature_screen, feature_screen_from_dataframe, FeatureScreenConfig, FeatureScreenResult, fit_screening_woe_engine, screen_config_from_mapping, weighted_feature_screen, WeightedScreenResult, DistributionShiftAnalyzer, DistributionPlotter, proc_means_odps
Models	LRMaster, GradientBoostingModel, LightGBMModel, XGBoostModel, CatBoostModel, lgbm_quick_train, xgbm_quick_train, catboost_quick_train, FeatureSelectionAnalyzer, BackwardVariableEliminator
Evaluation	PerformanceEvaluator, GainsTableCalculator, Model_Evaluation_Tool, EvaluationPipeline, cross_risk, get_gains_table_by_cust_metrics, calc_lift_apt, evaluate_performance, comparison_performance
Explainability	ModelExplainer, build_coalition_structure, CREDIT_PRIOR_GROUPS
Binning and utilities	Binning, super_binning, SlopeCalculator, DataFrameProcessor, FilePathManager, DateTimeUtils, WOEIVCalculator, TextEncryptor, get_feature_names, scoring
Model files	save_model, load_model, load_model_metadata
ODPS	ODPSRunner, ParallelODPSManager, ParallelODPSConfig, pull_attributes_in_batch
Parallel apply	parallel_apply, ParallelApplyEngine, ParallelApplyConfig, ParallelApplyResult
Dataset comparison	proc_compare, ProcCompareEngine, ProcCompareConfig, ProcCompareResult
Pipelines	CreditModelPipeline, FeatureValidationPipeline, RejectInferencePipeline, ScoreComparisonPipeline, ScoreConsistencyUATPipeline, SampleAnalysisPipeline, MockSamplePipeline, each with a <Name>Config and a <Name>Result class
Pipeline schema	FieldMeta, PipelineRegistryEntry, PIPELINE_REGISTRY, get_pipeline_registry, get_pipeline_registry_schema, get_config_field_meta, extract_config_schema, extract_pipeline_schema, extract_schema, config_to_dict, config_from_dict, config_to_yaml, config_from_yaml, validate_pipeline_config, generate_pipeline_code
Package metadata	__version__, __author__

Lazy names

import Modeling_Tool stays light. The model, explainability, pipeline, and ODPS names load their subpackage the first time you access them, and the optional libraries behind them (LightGBM, XGBoost, CatBoost, shap, lime, pyodps) load only when you use them. Because ODPSRunner is in __all__, from Modeling_Tool import * and from Modeling_Tool.Core import * import pyodps and fail without the odps extra. Import the names you need explicitly.

Names Outside __all__

These names can be imported from Modeling_Tool but are not part of __all__, so a star import skips them:

• get_gains_table, get_perf_summary, calc_roc, calc_pr, calc_equid_dist, calc_equid_pct, calc_fixed_pct
• backward_lgbm, backward_xgbm
• FeatureScreeningArtifact, screen_result_to_summary, run_modeling_from_validation
• ParallelODPSPuller, an alias of ParallelODPSManager

Names That Need a Subpackage Import

Many public names are not exported at the top level. Import them from their subpackage:

from Modeling_Tool.Feature import proc_means_by_grp
from Modeling_Tool.Core import calc_woe, calc_iv, parse_sql_file
from Modeling_Tool.Eval import plot_roc_curve, summarize_pct
from Modeling_Tool.Eval.evaluate_model import summarize_roc
from Modeling_Tool.UAT import UATConfig, UATConsistencyChecker
from Modeling_Tool.WOE import mapping_woe
from ExcelMaster.ExcelMaster import ExcelMaster
from Report.Report_Tool import single_model_perf

The pages above list each subpackage's names in the module that defines them.

Docstring Style

SMF uses NumPy-style docstrings: a one-line summary followed by Parameters, Returns, and Examples sections. mkdocstrings renders the sections as tables (docstring_style: numpy and docstring_section_style: table in mkdocs.yml), shows members in source order, and hides names that start with _.

Sample Weight API

The training and evaluation functions take sample weights as native arguments. The keyword depends on the function: weight_col (a column name) for LRMaster.fit, PerformanceEvaluator, GainsTableCalculator, and the DataFrame-based APIs; sample_weight (an array) for GradientBoostingModel.fit and the low-level calc_roc and evaluate_performance. The FAQ lists the keyword of every function. For metric semantics, see Model Training: Sample Weights and Model Evaluation: Sample-Weighted Evaluation.