Can't instantiate a LTRModelConfig from JSON (e.g. via exported LTRModelConfig.to_dict())

[aarondelgiudice]

The LTRModelConfig class provides the to_dict() method as easy way to export the object as a JSON:

with open(<filepath>, "w") as f:
    json.dump(ltr_config.to_dict(), f, indent=4)

The exported JSON can be checked into version control and referenced later.

But the class doesn't support a way to instantiate a new LTRModelConfig object from an exported JSON:

with open(<filepath>, "r") as f:
    config_dict = json.load(f)

def load_ltr_config(config_dict: dict) -> LTRModelConfig:
    feature_extractors = []

    for q in config_dict.get("learning_to_rank", {}).get("feature_extractors", {}):
        q = q.get("query_extractor", {})
        feature_name = q.get("feature_name")
        query = q.get("query")

        if feature_name and query:
            feature_extractors.append(
                QueryFeatureExtractor(feature_name=feature_name, query=query)
            )

    return LTRModelConfig(feature_extractors=feature_extractors)

A new method from_dict() would allow users to more easily import a LTR model config from JSON:

@classmethod
    def from_dict(cls, d: Mapping[str, Any]) -> "LTRModelConfig":
        if TYPE_LEARNING_TO_RANK not in d:
            raise ValueError(
                f"Invalid LTR model config, missing '{TYPE_LEARNING_TO_RANK}' key"
            )

        feature_extractors = []
        for feature_extractor in d[TYPE_LEARNING_TO_RANK]["feature_extractors"]:
            if "query_extractor" in feature_extractor:
                fe = feature_extractor["query_extractor"]
                feature_extractors.append(
                    QueryFeatureExtractor(
                        feature_name=fe["feature_name"],
                        query=fe["query"],
                        default_score=fe.get("default_score"),
                    )
                )
            else:
                raise ValueError(
                    f"Unknown feature extractor type: {list(feature_extractor.keys())}"
                )

        return cls(feature_extractors=feature_extractors)