Time Series Forecasting Models (TSFM)

classDiagram
    %% Exceptions
    class TSFMError {
        <<Exception>>
    }
    
    class InvalidInputError {
        <<Exception>>
        +__init__(message: str)
    }
    
    Exception <|-- TSFMError
    TSFMError <|-- InvalidInputError
    
    %% Output Classes
    class ForecastOutput {
        -df_preds: DataFrame
        -meta: dict[str, Any]
        +rmsfe: DataFrame
        +mae: DataFrame
        +me: DataFrame
        +metric(name: Literal) DataFrame
        +summary(digits: int) None
        +plot_actual_vs_pred(horizon: int, ax: Axes, start: Timestamp, end: Timestamp, return_ax: bool) Axes
        -_agg_mean(s: Series, name: str, post: Callable) DataFrame
        -_summary(digits: int) str
    }
    
    %% Base Model (Abstract)
    class Model {
        <<Abstract>>
        +registry: ClassVar[dict]
        +name: property
        +__init_subclass__(cls, name: str, **kwargs)
        +build(name: str, **kwargs) Model
        +get_backbone() Any*
        +_pred(df: DataFrame, y: str, X: list, ctx_len: int, horizon: int, oos_start: str)* DataFrame
        +pred(df: DataFrame, y: str, X: list, ctx_len: int, horizon: int, oos_start: str) ForecastOutput
    }
    
    ABC <|-- Model
    
    %% ARModel
    class ARModel {
        +get_backbone() None
        +_pred(df: DataFrame, y: str, X: list, ctx_len: int, horizon: int, oos_start: str) DataFrame
    }
    
    Model <|-- ARModel : name="armodel"
    ARModel ..> InvalidInputError : raises
    ARModel ..> ForecastOutput : creates
    
    %% Moirai Model
    class Moirai {
        +get_backbone() MoiraiModule
        +get_model(ctx_len: int, horizon: int, n_covariates: int) MoiraiForecast
        +_pred(df: DataFrame, y: str, X: list, ctx_len: int, horizon: int, oos_start: str) DataFrame
    }
    
    Model <|-- Moirai : name="moirai"
    Moirai ..> ForecastOutput : creates
    
    %% Moirai2 Model
    class Moirai2 {
        +get_backbone() Moirai2Module
        +get_model(ctx_len: int, horizon: int, n_covariates: int) Moirai2Forecast
        +_pred(df: DataFrame, y: str, X: list, ctx_len: int, horizon: int, oos_start: str) DataFrame
    }
    
    Model <|-- Moirai2 : name="moirai2"
    Moirai2 ..> ForecastOutput : creates
    
    %% Chronos Model
    class Chronos {
        +get_backbone() BaseChronosPipeline
        +_pred(df: DataFrame, y: str, X: list, ctx_len: int, horizon: int, oos_start: str) DataFrame
    }
    
    Model <|-- Chronos : name="chronos"
    Chronos ..> ForecastOutput : creates
    
    %% Chronos2 Model
    class Chronos2 {
        +get_backbone() Chronos2Pipeline
        +_pred(df: DataFrame, y: str, X: list, ctx_len: int, horizon: int, oos_start: str) DataFrame
    }
    
    Model <|-- Chronos2 : name="chronos2"
    Chronos2 ..> ForecastOutput : creates
    
    %% Data Module Functions
    class DataModule {
        <<module: data.py>>
        +infer_freq(df: DataFrame) str
        +generator(n: int, phi: float, beta0: float, beta1: float, sigma: float, trend: float, season_period: int, season_ampl: float, seed: int) DataFrame
        +split_is_oos(df: DataFrame, test_frac: float) tuple[DataFrame, DataFrame]
    }
    
    %% ARModel Helper Functions
    class ARModelHelpers {
        <<module: armodel.py>>
        +create_lags(var: Series, lags: int) DataFrame
        +get_design_matrix(ys: Series, y_lags: int, horizon: int) tuple[Series, DataFrame]
        +fit(ys: Series, y_lags: int, horizon: int) RegressionResultsWrapper
        +pred(ys: Series, y_lags: int, horizon: int, oos_start: str) Series
        +build_oos_panel(ys: Series, y_lags: int, horizon: int, oos_start: str, freq: str) DataFrame
        -_to_period_end(idx: Index, freq: str) DatetimeIndex
    }
    
    ARModel ..> ARModelHelpers : uses
    
    %% Moirai Helper Functions
    class MoiraiHelpers {
        <<module: moirai.py>>
        +make_fc_df(forecasts: list, y_true_series: Series, freq: str) DataFrame
        +prepare_data(df: DataFrame, y: str, X: list, horizon: int, oos_start: str, freq: str) TestData
        +MODEL_ID: str
    }
    
    Moirai ..> MoiraiHelpers : uses
    
    %% Moirai2 Helper Functions
    class Moirai2Helpers {
        <<module: moirai2.py>>
        +make_fc_df(forecasts: list, y_true_series: Series, freq: str) DataFrame
        +prepare_data(df: DataFrame, y: str, X: list, horizon: int, oos_start: str, freq: str) TestData
        +MODEL_ID: str
    }
    
    Moirai2 ..> Moirai2Helpers : uses
    
    %% Chronos Helper Functions
    class ChronosHelpers {
        <<module: chronos.py>>
        +prepare_data(df: DataFrame, y: str, X: list, ctx_len: int, oos_start: str, freq: str) DataFrame
        +make_fc_df(forecasts: DataFrame, y_true: DataFrame, horizon: int, freq: str) DataFrame
        +BOLT_MODEL_ID: str
    }
    
    Chronos ..> ChronosHelpers : uses
    
    %% Chronos2 Helper Functions
    class Chronos2Helpers {
        <<module: chronos2.py>>
        +prepare_data(df: DataFrame, y: str, X: list, ctx_len: int, oos_start: str, freq: str) DataFrame
        +make_fc_df(forecasts: DataFrame, y_true: DataFrame, horizon: int, freq: str) DataFrame
        +MODEL_ID: str
    }
    
    Chronos2 ..> Chronos2Helpers : uses
    
    %% Output Helper Functions
    class OutputHelpers {
        <<module: outputs.py>>
        +get_horizon_groupby(df: DataFrame, freq: str) Index
    }
    
    ForecastOutput ..> OutputHelpers : uses
    
    %% Relationships with external libraries
    Model ..> ForecastOutput : returns
    Model ..> DataModule : uses infer_freq
    
    %% Package level
    class TSFMPackage {
        <<module: __init__.py>>
        +ARModel
        +Model
        +Moirai
        +Moirai2
        +Chronos
        +Chronos2
        +generator
    }
    
    TSFMPackage ..> ARModel : exports
    TSFMPackage ..> Model : exports
    TSFMPackage ..> Moirai : exports
    TSFMPackage ..> Moirai2 : exports
    TSFMPackage ..> Chronos : exports
    TSFMPackage ..> Chronos2 : exports
    TSFMPackage ..> DataModule : exports generator

    %% Notes
    note for Model "Registry pattern:\nSubclasses auto-register\nby name parameter\nAuto-infers frequency from data"
    
    note for ForecastOutput "Contains forecast results\nwith MultiIndex[cutoff, oos_date]\nFrequency-aware metrics\nand visualization"
    
    note for ARModel "Autoregressive model\nusing statsmodels OLS\nwith HAC standard errors\nNo covariate support"
    
    note for Moirai "Pretrained foundation model\nfrom Salesforce\nmodel: moirai-1.1-R-small\nSupports covariates"
    
    note for Moirai2 "Pretrained foundation model\nfrom Salesforce\nmodel: moirai-2.0-R-small\nuses uni2ts library\nSupports covariates"
    
    note for Chronos "Amazon Chronos-Bolt model\nchronos-bolt-small\nNo covariate support\nFrequency-aware"
    
    note for Chronos2 "Amazon Chronos 2.0 model\namazon/chronos-2\nSupports covariates\nFrequency-aware"
    
    note for DataModule "infer_freq:\nAuto-detects frequency\nfrom DatetimeIndex\nSupports M, Q, W, D, etc."