snorkelflow.client.datasources.create_datasource

snorkelflow.client.datasources.create_datasource(dataset, path, file_type, uid_col=None, split='train', datasource_ds=None, reader_kwargs=None, credential_kwargs=None, scheduler=None, load_to_model_nodes=False, sync=True)

Create a data source.

Parameters

Parameters

Name	Type	Default	Info
dataset	Union[str, int]		Name or UID of the dataset to create the data source in.
path	str		Path to the data source (e.g. minio, s3, http).
file_type	str		File type (csv or parquet).
uid_col	Optional[str]	None	Name of the UID column in the data source. The values in this column must be unique non-negative integers that are not duplicated across files. If not specified, we will generate a SnorkelFlow ID column.
split	str	'train'	Split of the dataset to add data source to (train, valid, or test).
datasource_ds	Optional[str]	None	Datestamp of the data source in YYYY-MM-DD format.
reader_kwargs	Optional[Dict[str, Any]]	None	Dictionary of keyword arguments to pass to Dask read functions.
credential_kwargs	Optional[Dict[str, Any]]	None	Dictionary of credentials for specific data connectors.
scheduler	Optional[str]	None	Dask scheduler (threads, client, or group) to use.
load_to_model_nodes	bool	False	Load datasource in all tasks in dataset?.
sync	bool	True	Poll job status and block until complete?.

Returns

Returns

UID of the created data source if sync mode used. Otherwise, return job_id

Return type

Return type

Union[str, int]