Project Architecture

Note

Open the diagram in a new tab to see the full details.

UML Class Diagram

@startuml skinparam BackgroundColor transparent skinparam Shadowing false ' Diagram setup hide empty members left to right direction set namespaceSeparator none skinparam linetype polyline skinparam linetype ortho package foreshadow.utils { class check_df << (M,lemonchiffon) >> } package foreshadow.logging { class ForeshadowLogger } package foreshadow.intents { abstract class BaseIntent { list engineering_pipeline list preprocessing_pipeline resolve_intent() } class DropIntent class NumericalIntent class CategoricalIntent class TextIntent BaseIntent <|-- DropIntent BaseIntent <|-- NumericalIntent BaseIntent <|-- CategoricalIntent BaseIntent <|-- TextIntent note "Config for intentless transformations (cleaner) are placed in\nBaseIntent's specification" as N1 } package foreshadow.transformers.core { abstract class SmartTransformer { bool fixed_fit log_decision() } class ParallelProcessor class SigCopy class DropFeature class wrap_transformer << (M,lemonchiffon) >> wrap_transformer o-- SigCopy } package foreshadow.transformers.smart { SmartTransformer <|-- Cleaner SmartTransformer <|-- Engineerer SmartTransformer <|-- Scaler SmartTransformer <|-- Imputer SmartTransformer <|-- CategoricalEncoder SmartTransformer <|-- TextEncoder SmartTransformer <|-- Reducer } package foreshadow.transformers.internal { class FancyImpute class UncommonRemover class BoxCox FancyImpute o-- Imputer CategoricalEncoder o-- UncommonRemover Scaler o-- BoxCox class DaysSince class NumericalFeatuerizer class CategoricalFeatuerizer Engineerer o-- DaysSince Engineerer o-- NumericalFeaturizer Engineerer o-- CategoricalEncoder class ToString class SplitDate class FinancialCleaner Cleaner o-- ToString Cleaner o-- SplitDate Cleaner o-- FinancialCleaner class Boruta class Hypothesis Reducer o-- Boruta Reducer o-- Hypothesis } package foreshadow.transformers.external { note "All sklearn transformers are mirrored\nand pandas wrapped here." as N3 } ' foreshadow.transformers.smart <|-r- foreshadow.transformers.internal package foreshadow.config { class ConfigManager { json framework_config json user_config json local_config } } package foreshadow.tuners { class TunerWrapper { BaseEstimator tuner_type } } package foreshadow.core { abstract class BaseFeatureMapper { split_columns() join_columns() } class Foreshadow class DataPreparer { bool is_y_var } class FeatureCleaner class IntentResolver class FeatureEngineerer class FeaturePreprocessor class FeatureReducer class ColumnInfoSharer class SerializerMixin << (X,peru) >> Foreshadow "0..2" o-- DataPreparer Foreshadow "0..1" o-- sklearn.RandomizedSearchCV DataPreparer o-- FeatureCleaner DataPreparer o-- IntentResolver DataPreparer "0..1" o-- FeatureEngineerer DataPreparer o-- FeaturePreprocessor DataPreparer "0..1" o-- FeatureReducer SerializerMixin <|-- DataPreparer SerializerMixin <|-- FeatureCleaner SerializerMixin <|-- IntentResolver SerializerMixin <|-- FeatureEngineerer SerializerMixin <|-- FeaturePreprocessor SerializerMixin <|-- FeatureReducer BaseFeatureMapper <|-- DataPreparer BaseFeatureMapper <|-- FeatureCleaner BaseFeatureMapper <|-- IntentResolver BaseFeatureMapper <|-- FeatureEngineerer BaseFeatureMapper <|-- FeaturePreprocessor BaseFeatureMapper <|-- FeatureReducer } package foreshadow.estimators { class MetaEstimator class AutoEstimator MetaEstimator "0..1" o-- AutoEstimator MetaEstimator o-- DataPreparer Foreshadow "0..1" o-- MetaEstimator } package sklearn.base { class TransformerMixin << (X,peru) >> class BaseEstimator } @enduml

UML Sequence Diagrams

Main Sequence Diagram

@startuml skinparam Shadowing false participant User User -> Foreshadow: ~__init__() note over Foreshadow, DataPreparer Foreshadow and DataPreparer are wrappers for an sklearn Pipeline. Fit and transform calls originate from sklearn logic and not the objects themselves. end note note over Foreshadow shadow.steps = Pipeline([ ('t', DataPreparer()), ('m', LogisticRegression()), ]) end note note over DataPreparer ci = ColumnInfoSharer() dp.steps = Pipeline([ ('c', FeatureCleaner(share=ci)), ('i', IntentResolver(share=ci)), ('e', FeatureEngineerer(share=ci)), ('p', FeaturePreprocessor(share=ci)), ('r', FeatureReducer(share=ci)), ]) end note User -> Foreshadow: fit(X, y) Foreshadow -> DataPreparer: fit(X, y) DataPreparer -> ColumnInfoSharer: ~__init__() DataPreparer -> FeatureCleaner: fit_transform(X, y) FeatureCleaner -> ColumnInfoSharer: set_info(column, 'tags', values) FeatureCleaner --> DataPreparer: X DataPreparer -> IntentResolver: fit_transform(X, y) IntentResolver -> ColumnInfoSharer: set_info(column, 'intent', value) IntentResolver --> DataPreparer: X DataPreparer -> FeatureEngineerer: fit_transform(X, y) FeatureEngineerer -> ColumnInfoSharer: get_info(column, 'tags') return col_tag_list FeatureEngineerer --> DataPreparer: X DataPreparer -> FeaturePreprocessor: fit_transform(X, y) FeaturePreprocessor -> ColumnInfoSharer: get_info(column, 'intent') return col_intent FeaturePreprocessor --> DataPreparer: X DataPreparer -> FeatureReducer: fit_transform(X, y) FeatureReducer -> ColumnInfoSharer: get_info(column, 'intent') return col_intent FeatureReducer --> DataPreparer: X Foreshadow -> LogisticRegression: fit(X, y) LogisticRegression -> Foreshadow: self @enduml

Hyperparameter Optimization Sequence Diagram

@startuml skinparam BackgroundColor transparent skinparam Shadowing false participant User User -> Foreshadow: ~__init__() note over Foreshadow pipeline = Pipeline([ ('dp', DataPreparer()), ('lr', LogisticRegression()), ]) end note Foreshadow -> TunerWrapper: ~__init__(pipeline, RandomizedSearchCV) User -> Foreshadow: fit(X, y) Foreshadow -> TunerWrapper: fit(X, y) TunerWrapper -> RandomizedSearchCV ++: fit_pipelines(X, y) return best_pipeline TunerWrapper -> RandomizedSearchCV ++: fit_params(X, y) return best_pipeline_params TunerWrapper --> Foreshadow: self Foreshadow --> User: self @enduml