xai-org/x-algorithm · error · ValueError

Uknown {dataset_type=}, must be one of {DATASET_TYPES}

Error message

Uknown {dataset_type=}, must be one of {DATASET_TYPES}

What it means

The two-tower model's _make_dataset match over dataset_type exhausted all cases without a hit, so the value is not a valid DATASET_TYPES member for this family. (Message contains the typo 'Uknown'.)

Source

Thrown at phoenix/xrex/configs/xrecsys_two_tower.py:247

            return PhoenixDataset(
                hash_table=hash_table,
                path="/path/to/offline_kafka_dump",
                history_seq_len=mparams["history_seq_len"],
                candidate_seq_len=mparams["candidate_seq_len"],
                input_vocab_size=mparams["input_vocab_size"],
                hash_vocab_size=mparams["hash_vocab_size"],
                num_continuous_actions=mparams["num_continuous_actions"],
                num_negatives_per_example=mparams["num_negatives_per_example"],
                num_global_negatives_per_example=mparams["num_global_negatives_per_example"],
                num_kafka_partitions=1024,
                include_candidate_post_ids=True,
                date_range=mparams.get("date_range", None),
                global_ids_file_path=_global_ids_file_path,
                use_post_sid=_use_post_sid,
                sid_num_levels=_sid_num_levels,
            )
        case _:
            raise ValueError(f"Uknown {dataset_type=}, must be one of {DATASET_TYPES}")


def _xrecsys_two_tower_combined_base() -> dict:
    return {
        "history_seq_len": 1023,
        "enable_user_country_feature": True,
        "enable_user_language_feature": True,
        "enable_user_location_feature": True,
        "enable_user_gender_feature": True,
        "enable_user_age_feature": True,
        "enable_user_installed_apps": True,
        "candidate_seq_len": 64,
        "num_negatives_per_example": 0,
        "num_global_negatives_per_example": 64,
        "num_layers": 8,
        "emb_size": 1024,
        "emb_table_width": 1024,
        "query_heads": 16,

View on GitHub (pinned to 24c60942c5)

Solutions

  1. Set dataset_type to an exact member of DATASET_TYPES for xrecsys_two_tower
  2. Check for case-sensitive typos (must match exactly, e.g. 'aggregated_kafka')
  3. Register a custom dataset via config_registry.RANKING_DATASET_FACTORIES if needed
  4. Print the registry keys to see all accepted values at runtime

Example fix

# before
cfg = {"dataset_type": "TwoTowerKafka"}

# after
from phoenix.xrex.configs.xrecsys_two_tower import DATASET_TYPES
cfg = {"dataset_type": "aggregated_kafka"}  # exact valid member
Defensive patterns

Strategy: validation

Validate before calling

from phoenix.xrex.configs import xrecsys_two_tower

def validate_dataset_type(dataset_type: str) -> None:
    if dataset_type not in xrecsys_two_tower.DATASET_TYPES:
        raise SystemExit(
            f"Invalid dataset_type {dataset_type!r} for two_tower; valid: {xrecsys_two_tower.DATASET_TYPES}"
        )

Type guard

def is_two_tower_dataset_type(dataset_type: str) -> bool:
    return dataset_type in DATASET_TYPES

Try / catch

try:
    ds = _make_dataset(mparams, dataset_type, hash_table, config_name)
except ValueError as e:
    if "dataset_type" in str(e):
        raise SystemExit(f"Bad dataset_type; valid: {DATASET_TYPES}") from e
    raise

Prevention

When it happens

Trigger: Calling _make_dataset in xrecsys_two_tower with a dataset_type outside DATASET_TYPES and not present in RANKING_DATASET_FACTORIES — e.g. 'two_tower' (model name, not dataset type), wrong case, or a type only supported in another family.

Common situations: Confusing model name with dataset_type; copy-pasting dataset_type from gen-recs or SID-retrieval configs; version drift after dataset type renames; typos.

Understand the failure class

Background: "Invalid value" and "allowed values are" config errors: what your library rejected and how to fix it — this error's family across 41 libraries.

Related errors

AI-assisted analysis of xai-org/x-algorithm@24c60942c5 (2026-08-28). Data as JSON: /api/errors/d94f275340a4531d. Report an issue: GitHub.