Module `Modelkit.Artifact`

Portable, versioned persistence for fitted built-in pipelines.

Artifacts contain validated data only: no closures, marshalled OCaml values, commands, or training observations are serialized. The container and each component codec are independently versioned. Readers verify a declared MD5 checksum and enforce configurable bounds before allocating component payloads. The checksum detects accidental corruption; it does not authenticate or encrypt an artifact. The current 0.x format is experimental; released readers remain covered by golden compatibility tests.

Pipelines intended for persistence must use the artifact-aware stage and estimator constructors below. Encoding a fitted pipeline containing a component packaged only through Pipeline.transformer or Pipeline.estimator returns a typed artifact error.

type limits

type metadata

type 'model loaded

type regression_model =
  (Target.regression Target.t, Target.regression Target.t) Pipeline.fitted

type binary_classification_model =
  (Target.classification Target.t, Target.classification Target.t)
    Pipeline.fitted

val default_limits : limits

val limits : 
  ?max_bytes:int ->
  ?max_components:int ->
  ?max_features:int ->
  ?max_string_bytes:int ->
  ?max_metadata_entries:int ->
  unit ->
  (limits, Error.t) result

val empty_metadata : metadata

val metadata : 
  ?training_rows:int ->
  ?root_seed:Seed.t ->
  ?sample_weighted:bool ->
  ?labels:(string * string) array ->
  unit ->
  (metadata, Error.t) result

val model : 'model loaded -> 'model

val metadata_of_loaded : _ loaded -> metadata

val producer_version : _ loaded -> string

val producer_ocaml_version : _ loaded -> string option

val training_rows : metadata -> int option

val root_seed : metadata -> Seed.t option

val sample_weighted : metadata -> bool option

val labels : metadata -> (string * string) array

val simple_imputer_stage : 
  name:string ->
  Simple_imputer.t ->
  (Pipeline.transformer, Error.t) result

val standard_scaler_stage : 
  name:string ->
  Standard_scaler.t ->
  (Pipeline.transformer, Error.t) result

val variance_threshold_stage : 
  name:string ->
  Variance_threshold.t ->
  (Pipeline.transformer, Error.t) result

val linear_regression_estimator : 
  name:string ->
  Linear_regression.t ->
  ((Target.regression Target.t, Target.regression Target.t) Pipeline.estimator,
    Error.t)
    result

val ridge_regression_estimator : 
  name:string ->
  Ridge_regression.t ->
  ((Target.regression Target.t, Target.regression Target.t) Pipeline.estimator,
    Error.t)
    result

val logistic_regression_estimator : 
  name:string ->
  Logistic_regression.t ->
  ((Target.classification Target.t, Target.classification Target.t)
     Pipeline.estimator,
    Error.t)
    result

val encode_regression : 
  ?metadata:metadata ->
  ?limits:limits ->
  regression_model ->
  (bytes, Error.t) result

val encode_binary_classification : 
  ?metadata:metadata ->
  ?limits:limits ->
  binary_classification_model ->
  (bytes, Error.t) result

val decode_regression : 
  ?limits:limits ->
  bytes ->
  (regression_model loaded, Error.t) result

val decode_binary_classification : 
  ?limits:limits ->
  bytes ->
  (binary_classification_model loaded, Error.t) result

val save_regression : 
  ?metadata:metadata ->
  ?limits:limits ->
  path:string ->
  regression_model ->
  (unit, Error.t) result

Encodes completely before opening path, then writes the destination directly. Atomic replacement, signing, and encryption are transport-level responsibilities.

val save_binary_classification : 
  ?metadata:metadata ->
  ?limits:limits ->
  path:string ->
  binary_classification_model ->
  (unit, Error.t) result

The binary-classification counterpart to save_regression.

val load_regression : 
  ?limits:limits ->
  path:string ->
  unit ->
  (regression_model loaded, Error.t) result

val load_binary_classification : 
  ?limits:limits ->
  path:string ->
  unit ->
  (binary_classification_model loaded, Error.t) result

package modelkit

Module Modelkit.Artifact

Module `Modelkit.Artifact`