tdfs4ds.serve.dataset_code¶
The Python that builds a dataset, as the Dataset builder would (0.3.4.0).
dataset_python(...) renders the tdfs4ds.build_dataset /
tdfs4ds.build_historical_dataset call that the builder's Create runs for the
same form (the arguments come from the same parsing: panels._builder_selection,
_builder_kind, _builder_offsets), plus one explanation per argument. Pure:
no database access, so it is tested offline. Credentials are never written: the host
and user are filled in when known, the password is asked with getpass.
module-attribute
¶
_KINDS = {'current': 'A snapshot valid now: each time the view is read, it returns the values in force at that moment.', 'as_of': 'A snapshot frozen at {as_of}: the values valid at that instant, whatever is ingested later. Useful for a training set that must not move.', 'time_manager': 'A snapshot that follows the TimeManager {time_manager}: the view reads the values valid at its business date, and moving the manager (tm.update(...)) moves the dataset. This is how a scoring dataset follows a roll-out.'}
A dict literal with one key per line (easier to read and edit than pprint's).
dataset_python(*, store: str, data_domain: str, entity: str, selected_features: dict, view_name: str, schema_name: str, kind: dict, join_type: str, feature_offsets: dict | None = None, document: bool = False, columns: dict | None = None, host: str = '', username: str = '') -> dict
The code and its explanation.
Returns {"code": str, "explanation": [(argument, text), ...], "html": str};
columns ({feature: [column names]}, from preview_column_names) only
feeds the explanation.