aligned.exposed_model.ExposedModel

class documentation

class ExposedModel(Codable, SerializableType): (source)

Known subclasses: aligned.exposed_model.interface.ABTestModel, aligned.exposed_model.interface.CodePredictor, aligned.exposed_model.interface.DillFunction, aligned.exposed_model.interface.DillPredictor, aligned.exposed_model.interface.EnitityPredictor, aligned.exposed_model.interface.PolarsExpression, aligned.exposed_model.interface.ShadowModel

Constructors: ExposedModel.in_memory_mlflow(model_name, model_alias, reference_tag, mlflow_config), ExposedModel.mlflow_server(host, model_alias, model_name, timeout), ExposedModel.ollama_embedding(endpoint, model, input_features_versions, prompt_template, embedding_name, ...), ExposedModel.ollama_generate(endpoint, model, prompt_template, input_features_versions), ExposedModel.polars_predictor(callable, features), ExposedModel._deserialize(value)

View In Hierarchy

Undocumented

Static Method	`in_memory_mlflow`	Undocumented
Static Method	`mlflow_server`	Undocumented
Static Method	`ollama_embedding`	Undocumented
Static Method	`ollama_generate`	Undocumented
Static Method	`polars_predictor`	Undocumented
Async Method	`depends_on`	The data artefacts that the model depends on. Which is not the input features. This is useful for e.g. RAG systems, as we can describe which documents a model depends on Or something like a vector database that we assume to be up to date.
Method	`needed_configs`	Returns the config variables to set in order to use the model.
Async Method	`needed_entities`	Undocumented
Async Method	`needed_features`	Undocumented
Async Method	`potential_drift_from_model`	Undocumented
Async Method	`run_polars`	Undocumented
Method	`with_contract`	Undocumented
Method	`with_shadow`	Undocumented
Class Variable	`model_type`	Undocumented
Property	`as_markdown`	Undocumented
Property	`exposed_at_url`	Undocumented
Class Method	`_deserialize`	Undocumented
Method	`_serialize`	Undocumented

@staticmethod

def in_memory_mlflow(model_name: str, model_alias: str = 'champion', reference_tag: str = 'feature_refs', mlflow_config: MlflowConfig | None = None) -> ExposedModel: (source) ¶

Undocumented

@staticmethod

def mlflow_server(host: str, model_alias: str = 'champion', model_name: str | None = None, timeout: int = 30) -> ExposedModel: (source) ¶

Undocumented

@staticmethod

def ollama_embedding(endpoint: str, model: str, input_features_versions: str, prompt_template: str, embedding_name: str | None = None, precomputed_prompt_key: str = 'full_prompt') -> ExposedModel: (source) ¶

Undocumented

@staticmethod

def ollama_generate(endpoint: str, model: str, prompt_template: str, input_features_versions: str) -> ExposedModel: (source) ¶

Undocumented

@staticmethod

def polars_predictor(callable: Callable[[pl.DataFrame, ModelFeatureStore], Coroutine[None, None, pl.DataFrame]], features: list[FeatureReferencable] | None = None) -> ExposedModel: (source) ¶

Undocumented

async def depends_on(self) -> list[FeatureLocation]: (source) ¶

The data artefacts that the model depends on. Which is not the input features. This is useful for e.g. RAG systems, as we can describe which documents a model depends on Or something like a vector database that we assume to be up to date.

def needed_configs(self) -> list[ConfigValue]: (source) ¶

Returns the config variables to set in order to use the model.

async def needed_entities(self, store: ModelFeatureStore) -> set[Feature]: (source) ¶

overridden in aligned.exposed_model.interface.ABTestModel, aligned.exposed_model.interface.CodePredictor, aligned.exposed_model.interface.DillFunction, aligned.exposed_model.interface.DillPredictor, aligned.exposed_model.interface.EnitityPredictor, aligned.exposed_model.interface.PolarsExpression, aligned.exposed_model.interface.ShadowModel

Undocumented

async def needed_features(self, store: ModelFeatureStore) -> list[FeatureReference]: (source) ¶

Undocumented

async def potential_drift_from_model(self, old_model: ExposedModel) -> str | None: (source) ¶

Undocumented

async def run_polars(self, values: RetrievalJob, store: ModelFeatureStore) -> pl.DataFrame: (source) ¶

Undocumented

def with_contract(self, model: Model) -> ExposedModel: (source) ¶

Undocumented

def with_shadow(self, shadow_model: ExposedModel) -> ShadowModel: (source) ¶

Undocumented

model_type: str = (source) ¶

Undocumented

@property

as_markdown: str = (source) ¶

Undocumented

@property

exposed_at_url: str | None = (source) ¶

Undocumented

@classmethod

def _deserialize(cls, value: dict) -> ExposedModel: (source) ¶

Undocumented

def _serialize(self) -> dict: (source) ¶

Undocumented