Pipeline GUI Schema

SMF 0.5.4+ provides a dependency-free schema layer for building a Pipeline configuration GUI. The GUI itself should live outside the modeling package, but it can treat SMF as the source of truth for:

• the 7 public high-level Pipeline cards
• each Pipeline's Config dataclass fields
• field labels, descriptions, widgets, options, ranges, groups and dependencies
• run signature metadata
• result attributes
• YAML-safe config export and import
• lightweight config-only validation

Public API

from Modeling_Tool import (
    FieldMeta,
    PIPELINE_REGISTRY,
    get_pipeline_registry,
    get_pipeline_registry_schema,
    extract_pipeline_schema,
    extract_config_schema,
    extract_schema,
    config_to_dict,
    config_from_dict,
    config_to_yaml,
    config_from_yaml,
    validate_pipeline_config,
    generate_pipeline_code,
)

The same names are also exported from Modeling_Tool.Pipeline.

Registered Pipelines

get_pipeline_registry_schema() returns a JSON/YAML-friendly registry without class objects:

registry = get_pipeline_registry_schema()

for key, item in registry.items():
    print(key, item["display_name"], item["run_method"])

The registry currently contains:

key	Pipeline	Config	run_requires_data
credit_model	CreditModelPipeline	CreditModelPipelineConfig	True
feature_validation	FeatureValidationPipeline	FeatureValidationPipelineConfig	True
reject_inference	RejectInferencePipeline	RejectInferencePipelineConfig	True
score_comparison	ScoreComparisonPipeline	ScoreComparisonPipelineConfig	True
score_consistency_uat	ScoreConsistencyUATPipeline	ScoreConsistencyUATPipelineConfig	False
sample_analysis	SampleAnalysisPipeline	SampleAnalysisPipelineConfig	True
mock_sample	MockSamplePipeline	MockSamplePipelineConfig	False

run_requires_data=False means the generated code should not use data=your_dataframe. For UAT, users may still call run(offline_data=..., online_data=...) manually in code mode.

Extract One Schema

schema = extract_pipeline_schema("feature_validation")

print(schema["display_name"])
print(schema["config_class_name"])
print(schema["run_method"])

for field in schema["fields"]:
    print(field["name"], field["label"], field["widget"], field["default"])

Each field item includes:

Field	Meaning
name	Config dataclass field name.
type	Resolved type hint as text.
default	JSON/YAML-safe default value when possible.
label	Human-readable label.
description	Business or technical explanation.
widget	Suggested UI widget: text, number, select, multiselect, toggle, slider, textarea, json, hidden.
options	Allowed values for select/multiselect fields.
min_val / max_val / step	Numeric control hints.
required	Whether the GUI should mark the field as required.
group	Suggested form section.
depends_on	Conditional display rule.
yaml_serializable	Whether this field is safe for YAML export by default.
gui_editable	Whether the GUI should show it as an editable field.
advanced / expert_only	Suggested progressive disclosure flags.
nested_fields	Sub-form hints for dict-like fields such as feature_selection, woe_params, corr_params.

Extract All Schemas

all_schemas = extract_pipeline_schema()
credit_schema = all_schemas["pipelines"]["credit_model"]

extract_schema("credit_model") is a compatibility helper that returns only the fields list.

YAML Export

from Modeling_Tool.Pipeline import MockSamplePipelineConfig
from Modeling_Tool import config_to_yaml

cfg = MockSamplePipelineConfig(
    n_samples=80000,
    applied_sample=1,
)

yaml_text = config_to_yaml(cfg)
print(yaml_text)

Output shape:

pipeline: mock_sample
pipeline_class: MockSamplePipeline
config_class: MockSamplePipelineConfig
smf_version: 0.5.6
config:
  n_samples: 80000
  applied_sample: 1

config_to_yaml() requires PyYAML at runtime. SMF does not require Streamlit or any frontend dependency.

YAML Import

from Modeling_Tool import config_from_yaml

cfg = config_from_yaml(None, yaml_text)

When the first argument is None, SMF resolves the config class from the YAML pipeline or config_class. You can also pass an explicit key:

cfg = config_from_yaml("mock_sample", yaml_text)

Dict Export

from Modeling_Tool import config_to_dict

payload = config_to_dict(cfg)

By default, non-serializable fields are skipped. This is important for GUI/YAML safety.

Examples of skipped fields:

• screening_artifact
• feature_validation_result
• extra_eval_datasets
• oot_data
• ri_approved_data
• ri_approved_func
• gains_add_func
• sqlrunner
• offline_data
• online_data
• psi_reference_data

These fields remain usable in normal Python code, but should not be edited in a generic GUI form.

Config Validation

from Modeling_Tool import validate_pipeline_config

messages = validate_pipeline_config(
    "credit_model",
    {
        "target_col": "badflag",
        "warm_start_enabled": True,
        "warm_start_score_col": None,
    },
)

for msg in messages:
    print(msg)

The validator is intentionally lightweight. It catches common form mistakes before code generation, while each Pipeline's run() method remains the authoritative runtime validation.

Code Generation

from Modeling_Tool import generate_pipeline_code

code = generate_pipeline_code(
    "score_comparison",
    {
        "target_col": "badflag",
        "score_cols": ["score_old", "score_new"],
        "base_score": "score_old",
    },
)

print(code)

For MockSamplePipeline and ScoreConsistencyUATPipeline, generated code uses result = pipeline.run(). For the other five pipelines, generated code uses result = pipeline.run(data=your_dataframe).

GUI Design Guidance

• Use get_pipeline_registry_schema() to render Pipeline cards.
• Use extract_pipeline_schema(key)["fields"] to render forms.
• Hide fields with gui_editable=False unless the GUI has an explicit code-only advanced mode.
• Do not put non-serializable fields in YAML exports.
• Render dict fields with nested_fields as sub-forms when possible; otherwise use a JSON textarea.
• Treat FeatureScreeningArtifact handoff as a Python-code advanced workflow, not a YAML field.
• Keep Streamlit or web UI dependencies outside the SMF main package.