snorkelflow.client.nodes.get_dataset

snorkelflow.client.nodes.get_dataset(node, split='dev', batch_uid=None, combiner='AND', show_filtered_flag=False, gt_label=None, all_lfs_filter=None, no_lfs_filter=None, model_filters=None, lf_filters=None, training_set_filters=None, include_tag_uids=None, include_tag_type_uids=None, exclude_tag_uids=None, exclude_tag_type_uids=None)

Filter the dataset and return with GT labels and uids as index.

You can also filter for only data points that have a certain ground truth label or a certain predicted label by a machine learning model. This filter is handy when you want to do your own analysis on an error bucket.

Multiple conditions will be combined with AND semantics.

Examples

# Filter data where all LFs abstain.
df = sf.get_dataset(node, all_lfs_filter="UNKNOWN")
# Filter data where their GT labels are "LABEL".
df = sf.get_dataset(node, gt_label="LABEL")
# Filter data with a set of rules.
df = sf.get_dataset(
    node,
    gt_label="LABEL",
    model_filters=[(1, "LABEL"), (2, "LABEL")],
    lf_filters=[("my_lf", "LABEL")],
    training_set_filters=[(1, "LABEL")],
    combiner="AND"
)

# Generate column statistics and summaries for the dataset
df = sf.get_dataset(node)
df.describe(include="all")

Parameters

Parameters

Name	Type	Default	Info
node	int		UID of the node.
split	str	'dev'	The split to be loaded. Only "dev", "valid", and "test" splits are allowed. Default to "dev".
batch_uid	Optional[int]	None	UID of the annotation batch to filter.
combiner	str	'AND'	Combiner to apply across all filters, by default "AND"
show_filtered_flag	bool	False	If True, return all datapoints and a column called “filtered_flag” with “True” or “False” values specifying whether the data is included in the provided filters.
gt_label	Optional[str]	None	If set, include only data points where ground truth is this label string.
all_lfs_filter	Optional[str]	None	If set, include only data points where all LFs vote for the label string passed.
no_lfs_filter	Optional[str]	None	If set, include only data points where no LFs vote for the label string passed.
model_filters	Optional[List[Tuple[int, str]]]	None	Tuple where the first value is the model_id, and the second is predicted label or voting pattern. If provided, include only data points that match this pattern.
lf_filters	Optional[List[Tuple[str, str]]]	None	List of tuples where first value in tuple is LF name and second value is assigned label or voting pattern. If provided, include only data points that match this pattern.
training_set_filters	Optional[List[Tuple[int, str]]]	None	List of tuples where first value in tuple is training set ID and second value is assigned label or voting pattern. If provided, include only data points that match this pattern.
include_tag_uids	Optional[List[int]]	None	List of tag types, only data points which include this tag type will be included.
include_tag_type_uids	Optional[List[int]]	None	List of tag types, only data points which include this tag type will be included.
exclude_tag_uids	Optional[List[int]]	None	List of tag types, only data points which exclude this tag type will be included. Exclusion will supersede inclusion in case of collision.
exclude_tag_type_uids	Optional[List[int]]	None	List of tag types, only data points which exclude this tag type will be included. Exclusion will supersede inclusion in case of collision.

Returns

Returns

A Pandas DataFrame for give split filtered according to parameters

Return type

Return type

pd.DataFrame