Slicing functions#

giskard.slicing_function(_fn=None, row_level=True, name=None, tags: List[str] | None = None, cell_level=False)#

Decorator that registers a function as a slicing function and returns a SlicingFunction instance. It can be used for slicing datasets in a specific way during testing.

Parameters:

_fn – function to decorate. No need to provide this argument, the decorator will automatically take as input the function to decorate.
name – Optional name to use for the function when registering it.
tags – Optional list of tags to use when registering the function.
row_level – Whether to apply the slicing function row-wise (default) or on the full dataframe. If row_level is True, the slicing function will receive a row (either a Series or DataFrame), and if False, it will receive the entire dataframe.
cell_level – Whether to apply the slicing function on the cell level. If True, the slicing function will be applied to individual cells instead of rows or the entire dataframe.

Returns:

The wrapped function or a new instance of SlicingFunction.

class giskard.ml_worker.testing.registry.slicing_function.SlicingFunction(func: Callable[[...], bool] | None, row_level=True, cell_level=False)#

A slicing function used to subset data.

func#

The function used to slice the data.

Type:: SlicingFunctionType

row_level#

Whether the slicing function should operate on rows or columns.

Type:: bool

cell_level#

Whether the slicing function should operate at the cell level.

Type:: bool

params#

Additional parameters for the slicing function.

Type:: Dict

is_initialized#

Indicates if the slicing function has been initialized.

Type:: bool

Initializes a new instance of the SlicingFunction class.

Parameters:

func (SlicingFunctionType) – The function used to slice the data.
row_level (bool) – Whether the slicing function should operate on rows or the whole dataframe. Defaults to True.

execute(data: Series | DataFrame)#

Slices the data using the slicing function.

Parameters:: data (Union[pd.Series, pd.DataFrame]) – The data to slice.
Returns:: The sliced data.
Return type:: Union[pd.Series, pd.DataFrame]

upload(client: GiskardClient, project_key: str | None = None) → str#

Uploads the slicing function and its metadata to the Giskard server.

Parameters:

client (GiskardClient) – The Giskard client instance used for communication with the server.
project_key (str, optional) – The project key where the slicing function will be uploaded. If None, the function will be uploaded to the global scope. Defaults to None.

Returns:

The UUID of the uploaded slicing function.

Return type:

str

classmethod download(uuid: str, client: GiskardClient | None, project_key: str | None) → Artifact#

Downloads the artifact from the Giskard server or retrieves it from the local cache.

Parameters:

uuid (str) – The UUID of the artifact to download.
client (GiskardClient, optional) – The Giskard client instance used for communication with the server. If None, the artifact will be retrieved from the local cache if available. Defaults to None.
project_key (str, optional) – The project key where the artifact is located. If None, the artifact will be retrieved from the global scope. Defaults to None.

Returns:

The downloaded artifact.

Return type:

Artifact

Raises:

AssertionError – If the artifact metadata cannot be retrieved.
AssertionError – If the artifact is not found in the cache and the Giskard client is None.

Textual slicing#

giskard.ml_worker.testing.functions.slicing.short_comment_slicing_fn(max_words: SuiteInput | int | None = 5) → SlicingFunction#: Filter the rows where the specified ‘column_name’ contains a short comment, defined as one with at most ‘max_words’.

giskard.ml_worker.testing.functions.slicing.keyword_lookup_slicing_fn(keywords: SuiteInput | List[str] | None = None) → SlicingFunction#: Filter the rows where the specified ‘column_name’ contains at least one of the specified ‘keywords’.

giskard.ml_worker.testing.functions.slicing.positive_sentiment_analysis(column_name: SuiteInput | str | None = None, threshold: SuiteInput | float | None = 0.9) → SlicingFunction#: Filter the rows where the specified ‘column_name’ has a positive sentiment, as determined by a pre-trained sentiment analysis model.

giskard.ml_worker.testing.functions.slicing.offensive_sentiment_analysis(column_name: SuiteInput | str | None = None, threshold: SuiteInput | float | None = 0.9) → SlicingFunction#: Filter the rows where the specified ‘column_name’ has a offensive sentiment, as determined by a pre-trained sentiment analysis model.

giskard.ml_worker.testing.functions.slicing.irony_sentiment_analysis(column_name: SuiteInput | str | None = None, threshold: SuiteInput | float | None = 0.9) → SlicingFunction#: Filter the rows where the specified ‘column_name’ has a ironic sentiment, as determined by a pre-trained sentiment analysis model.

giskard.ml_worker.testing.functions.slicing.hate_sentiment_analysis(column_name: SuiteInput | str | None = None, threshold: SuiteInput | float | None = 0.9) → SlicingFunction#: Filter the rows where the specified ‘column_name’ has a hateful sentiment, as determined by a pre-trained sentiment analysis model.

giskard.ml_worker.testing.functions.slicing.emotion_sentiment_analysis(column_name: SuiteInput | str | None = None, emotion: SuiteInput | str | None = None, threshold: SuiteInput | float | None = 0.9) → SlicingFunction#: Filter the rows where the specified ‘column_name’ has an emotion matching ‘emotion’, as determined by a pre-trained sentiment analysis model. Possible emotion are: ‘optimism’, ‘anger’, ‘sadness’, ‘joy’

Numerical slicing functions#

giskard.ml_worker.testing.functions.slicing.outlier_filter(lower_bound: SuiteInput | float | None = None, upper_bound: SuiteInput | float | None = None) → SlicingFunction#: Filter rows where the specified column values fall outside the specified range.