deepchecks.nlp

Package for nlp functionality.

Modules

checks	Module importing all nlp checks.
suites	Module contains all prebuilt nlp suites.
datasets	Module for working with pre-built datasets.
metric_utils	Module containing metrics utils for nlp tasks.
utils	Utils package for nlp functionality.
text_data	The dataset module containing the tabular Dataset class and its functions.
base_checks	Module for nlp base checks.
context	Module for base nlp context.
suite	Module for base nlp suite.

Classes

class TextData

TextData wraps together the raw text data and the labels for the nlp task.

The TextData class contains metadata and methods intended for easily accessing metadata relevant for the training or validating of ML models.

Parameters

raw_textt.Sequence[str], default: None

The raw text data, a sequence of strings representing the raw text of each sample. If not given, tokenized_text must be given, and raw_text will be created from it by joining the tokens with spaces.

tokenized_textt.Sequence[t.Sequence[str]], default: None

The tokenized text data, a sequence of sequences of strings representing the tokenized text of each sample. Only relevant for task_type ‘token_classification’. If not given, raw_text must be given, and tokenized_text will be created from it by splitting the text by spaces.

labelt.Optional[TTextLabel], default: None

The label for the text data. Can be either a text_classification label or a token_classification label. If None, the label is not set.

• text_classification label - For text classification the accepted label format differs between multilabel and single label cases. For single label data, the label should be passed as a sequence of labels, with one entry per sample that can be either a string or an integer. For multilabel data, the label should be passed as a sequence of sequences, with the sequence for each sample being a binary vector, representing the presence of the i-th label in that sample.

• token_classification label - For token classification the accepted label format is the IOB format or similar to it. The Label must be a sequence of sequences of strings or integers, with each sequence corresponding to a sample in the tokenized text, and exactly the length of the corresponding tokenized text.

task_typestr, default: None

The task type for the text data. Can be either ‘text_classification’ or ‘token_classification’. Must be set if label is provided.

namet.Optional[str] , default: None

The name of the dataset. If None, the dataset name will be defined when running it within a check.

metadatat.Optional[t.Union[pd.DataFrame, str]] , default: None

Metadata for the samples. Metadata must be given as a pandas DataFrame or a path to a pandas DataFrame compatible csv file, with the rows representing each sample and columns representing the different metadata columns. If None, no metadata is set. The number of rows in the metadata DataFrame must be equal to the number of samples in the dataset, and the order of the rows must be the same as the order of the samples in the dataset. For more on metadata, see the NLP Metadata Guide.

categorical_metadatat.Optional[t.List[str]] , default: None

The names of the categorical metadata columns. If None, categorical metadata columns are automatically inferred. Only relevant if metadata is not None.

propertiest.Optional[t.Union[pd.DataFrame, str]] , default: None

The text properties for the samples. Properties must be given as either a pandas DataFrame or a path to a pandas DataFrame compatible csv file, with the rows representing each sample and columns representing the different properties. If None, no properties are set. The number of rows in the properties DataFrame must be equal to the number of samples in the dataset, and the order of the rows must be the same as the order of the samples in the dataset. In order to calculate the default properties, use the TextData.calculate_builtin_properties function after the creation of the TextData object. For more on properties, see the NLP Properties Guide.

categorical_propertiest.Optional[t.List[str]] , default: None

The names of the categorical properties columns. Should be given only for custom properties, not for any of the built-in properties. If None, categorical properties columns are automatically inferred for custom properties.

embeddingst.Optional[Union[np.ndarray, pd.DataFrame, str]], default: None

The text embeddings for the samples. Embeddings must be given as a numpy array (or a path to an .npy file containing a numpy array) of shape (N, E), where N is the number of samples in the TextData object and E is the number of embeddings dimensions. The numpy array must be in the same order as the samples in the TextData. If None, no embeddings are set.

In order to use the built-in embeddings, use the TextData.calculate_builtin_embeddings function after the creation of the TextData object. For more on embeddings, see the Text Embeddings Guide

Attributes

categorical_metadata

Return categorical metadata column names.

categorical_properties

Return categorical properties names.

embeddings

Return the embeddings of for the dataset.

label

Return the label defined in the dataset.

metadata

Return the metadata of for the dataset.

n_samples

Return number of samples in the dataset.

name

numerical_metadata

Return numeric metadata column names.

numerical_properties

Return numerical properties names.

properties

Return the properties of the dataset.

task_type

Return the task type.

text

Return sequence of raw text samples.

tokenized_text

Return sequence of tokenized text samples.

Methods

calculate_builtin_embeddings([model, ...])	Calculate the built-in embeddings of the dataset.
calculate_builtin_properties([...])	Calculate the default properties of the dataset.
cast_to_dataset(obj)	Verify Dataset or transform to Dataset.
copy([rows_to_use])	Create a copy of this Dataset with new data.
describe([n_properties_to_show, ...])	Provide holistic view of the data.
get_original_text_indexes()	Return the original indexes of the text samples.
get_sample_at_original_index(index)	Return the text sample at the original index.
has_label()	Return True if label was set.
head([n_samples, model_classes])	Return a copy of the dataset as a pandas Dataframe with the first n_samples samples.
is_multi_label_classification()	Check if the dataset is multi-label.
is_sampled(n_samples)	Return True if the dataset number of samples will decrease when sampled with n_samples samples.
label_for_display([model_classes])	Return the label defined in the dataset in a format that can be displayed.
label_for_print([model_classes])	Return the label defined in the dataset in a format that can be printed nicely.
len_when_sampled(n_samples)	Return number of samples in the sampled dataframe this dataset is sampled with n_samples samples.
sample(n_samples[, replace, random_state, ...])	Create a copy of the dataset object, with the internal data being a sample of the original data.
save_properties(path)	Save the dataset properties to csv.
set_embeddings(embeddings[, verbose])	Set the embeddings of the dataset.
set_metadata(metadata[, categorical_metadata])	Set the metadata of the dataset.
set_properties(properties[, ...])	Set the properties of the dataset.
validate_textdata_compatibility(other_text_data)	Verify that all provided datasets share same label name and task types.

__init__(raw_text: Optional[Sequence[str]] = None, tokenized_text: Optional[Sequence[Sequence[str]]] = None, label: Optional[Union[Sequence[Union[int, str, Tuple[Union[int, str]]]], Sequence[Sequence[Union[str, int]]], Sequence[None]]] = None, task_type: Optional[str] = None, name: Optional[str] = None, embeddings: Optional[Union[DataFrame, ndarray, str]] = None, metadata: Optional[DataFrame] = None, categorical_metadata: Optional[List[str]] = None, properties: Optional[DataFrame] = None, categorical_properties: Optional[List[str]] = None)

calculate_builtin_embeddings(model: str = 'miniLM', file_path: str = 'embeddings.npy', device: Optional[str] = None, long_sample_behaviour: str = 'average+warn', open_ai_batch_size: int = 500)

Calculate the built-in embeddings of the dataset.

Parameters

modelstr, default: ‘miniLM’

The model to use for calculating the embeddings. Possible values are: ‘miniLM’: using the miniLM model in the sentence-transformers library. ‘open_ai’: using the ADA model in the open_ai library. Requires an API key.

file_pathstr, default: ‘embeddings.npy’

The path to save the embeddings to.

devicestr, default: None

The device to use for calculating the embeddings. If None, the default device will be used.

long_sample_behaviourstr, default ‘average+warn’

How to handle long samples. Averaging is done as described in https://github.com/openai/openai-cookbook/blob/main/examples/Embedding_long_inputs.ipynb Currently, applies only to the ‘open_ai’ model, as the ‘miniLM’ model can handle long samples.

Options are:

• ‘average+warn’ (default): average the embeddings of the chunks and warn if the sample is too long.

• ‘average’: average the embeddings of the chunks.

• ‘truncate’: truncate the sample to the maximum length.

• ‘raise’: raise an error if the sample is too long.

• ‘nan’: return an embedding vector of nans for each sample that is too long.

open_ai_batch_sizeint, default 500

The amount of samples to send to open ai in each batch. Reduce if getting errors from open ai.

calculate_builtin_properties(include_properties: Optional[List[str]] = None, ignore_properties: Optional[List[str]] = None, include_long_calculation_properties: bool = False, ignore_non_english_samples_for_english_properties: bool = True, device: Optional[str] = None)

Calculate the default properties of the dataset.

Parameters

include_propertiesList[str], default None

The properties to calculate. If None, all default properties will be calculated. Cannot be used together with ignore_properties parameter.

ignore_propertiesList[str], default None

The properties to ignore. If None, no properties will be ignored. Cannot be used together with properties parameter.

include_long_calculation_propertiesbool, default False

Whether to include properties that may take a long time to calculate. If False, these properties will be ignored.

ignore_non_english_samples_for_english_propertiesbool, default True

Whether to ignore samples that are not in English when calculating English properties. If False, samples that are not in English will be calculated as well. This parameter is ignored when calculating non-English properties. English-Only properties WILL NOT work properly on non-English samples, and this parameter should be used only when you are sure that all the samples are in English.

deviceint, default None

The device to use for the calculation. If None, the default device will be used.

classmethod cast_to_dataset(obj: Any) → TextData

Verify Dataset or transform to Dataset.

Function verifies that provided value is a non-empty instance of Dataset, otherwise raises an exception, but if the ‘cast’ flag is set to True it will also try to transform provided value to the Dataset instance.

Parameters

obj

value to verify

Raises

DeepchecksValueError

if the provided value is not a TextData instance; if the provided value cannot be transformed into Dataset instance;

property categorical_metadata: List[str]

Return categorical metadata column names.

property categorical_properties: List[str]

Return categorical properties names.

copy(rows_to_use: Optional[Sequence[int]] = None) → TDataset

Create a copy of this Dataset with new data.

Parameters

rows_to_uset.Optional[t.List[int]] , default: None

The rows to use in the new copy. If None, the new copy will contain all the rows.

describe(n_properties_to_show: Optional[int] = 4, properties_to_show: Optional[List[str]] = None, max_num_labels_to_show: Optional[int] = 5, model_classes: Optional[List[str]] = None)

Provide holistic view of the data.

Generates the following plots: 1. Label distribution 2. Statistics about the data such as number of samples, annotation ratio, list of metadata columns, list of text properties and so on. 3. Property distribution for the text properties defined either by n_properties_to_show or properties_to_show parameter.

Parameters

n_properties_to_showint, default: 4

Number of properties to consider for generating property distribution graphs. If properties_to_show is provided, this value is ignored.

properties_to_showList[str], default: None

List of property names to consider for generating property distribution graphs. If None, all the properties are considered.

max_num_labels_to_showint, default: 5

The threshold to display the maximum number of labels on the label distribution pie chart and display rest of the labels under “Others” category.

model_classesOptional[List[str]], default: None

List of classes names to use for multi-label display. Only used if the dataset is multi-label.

Returns

Displays the Plotly Figure.

property embeddings: DataFrame

Return the embeddings of for the dataset.

get_original_text_indexes() → Sequence[int]

Return the original indexes of the text samples.

Returns

t.Sequence[int]

Original indexes of the text samples.

get_sample_at_original_index(index: int) → str

Return the text sample at the original index.

Parameters

indexint

Original index of the text sample.

Returns

str

Text sample at the original index.

has_label() → bool

Return True if label was set.

Returns

bool

True if label was set.

head(n_samples: int = 5, model_classes: Optional[list] = None) → DataFrame

Return a copy of the dataset as a pandas Dataframe with the first n_samples samples.

Parameters

n_samplesint, default 5

Number of samples to return.

model_classeslist, default None

List of classes names to use for multi-label display. Only used if the dataset is multi-label.

Returns

pd.DataFrame

A copy of the dataset as a pandas Dataframe with the first n_samples samples.

is_multi_label_classification() → bool

Check if the dataset is multi-label.

is_sampled(n_samples: Optional[int])

Return True if the dataset number of samples will decrease when sampled with n_samples samples.

property label: Union[Sequence[Union[int, str, Tuple[Union[int, str]]]], Sequence[Sequence[Union[str, int]]], Sequence[None]]

Return the label defined in the dataset.

Returns

TTextLabel

label_for_display(model_classes: Optional[list] = None) → Union[Sequence[Union[int, str, Tuple[Union[int, str]]]], Sequence[Sequence[Union[str, int]]], Sequence[None]]

Return the label defined in the dataset in a format that can be displayed.

Parameters

model_classeslist, default None

List of classes names to use for multi-label display. Only used if the dataset is multi-label.

Returns

TTextLabel

label_for_print(model_classes: Optional[list] = None) → List[str]

Return the label defined in the dataset in a format that can be printed nicely.

Parameters

model_classeslist, default None

List of classes names to use for multi-label display. Only used if the dataset is multi-label.

Returns

List[str]

len_when_sampled(n_samples: Optional[int])

Return number of samples in the sampled dataframe this dataset is sampled with n_samples samples.

property metadata: DataFrame

Return the metadata of for the dataset.

property n_samples: int

Return number of samples in the dataset.

property numerical_metadata: List[str]

Return numeric metadata column names.

property numerical_properties: List[str]

Return numerical properties names.

property properties: DataFrame

Return the properties of the dataset.

sample(n_samples: int, replace: bool = False, random_state: Optional[int] = None, drop_na_label: bool = False) → TDataset

Create a copy of the dataset object, with the internal data being a sample of the original data.

Parameters

n_samplesint

Number of samples to draw.

replacebool, default: False

Whether to sample with replacement.

random_statet.Optional[int] , default None

Random state.

drop_na_labelbool, default: False

Whether to take sample only from rows with exiting label.

Returns

Dataset

instance of the Dataset with sampled internal dataframe.

save_properties(path: str)

Save the dataset properties to csv.

Parameters

pathstr

Path to save the properties to.

set_embeddings(embeddings: ndarray, verbose: bool = True)

Set the embeddings of the dataset.

Parameters

embeddingspd.DataFrame

Embeddings to set.

verbosebool, default: True

Whether to print information about the process.

set_metadata(metadata: DataFrame, categorical_metadata: Optional[Sequence[str]] = None)

Set the metadata of the dataset.

set_properties(properties: DataFrame, categorical_properties: Optional[Sequence[str]] = None)

Set the properties of the dataset.

property task_type: Optional[deepchecks.nlp.task_type.TaskType]

Return the task type.

Returns

t.Optional[TaskType]

Task type

property text: Sequence[str]

Return sequence of raw text samples.

Returns

t.Sequence[str]

Sequence of raw text samples.

property tokenized_text: Sequence[Sequence[str]]

Return sequence of tokenized text samples.

Returns

t.Sequence[t.Sequence[str]]

Sequence of tokenized text samples.

validate_textdata_compatibility(other_text_data: TextData) → bool

Verify that all provided datasets share same label name and task types.

Parameters

other_text_dataTextData

The other dataset TextData object to compare with.

Returns

bool

True if provided dataset share same label name and task types.

class Context

Contains all the data + properties the user has passed to a check/suite, and validates it seamlessly.

Parameters

train_datasetUnion[TextData, None] , default: None

TextData object, representing data an estimator was fitted on

test_datasetUnion[TextData, None] , default: None

TextData object, representing data an estimator predicts on

with_displaybool , default: True

flag that determines if checks will calculate display (redundant in some checks).

train_predUnion[TTextPred, None] , default: None

predictions on train dataset

test_predUnion[TTextPred, None] , default: None

predictions on test dataset

train_probaUnion[TTextProba, None] , default: None

probabilities on train dataset

test_probaUnion[TTextProba, None] , default: None

probabilities on test dataset

model_classesOptional[List] , default: None

list of classes known to the model

random_state: int, default 42

A seed to set for pseudo-random functions , primarily sampling.

Attributes

common_datasets_properties

Return common for train and test datasets properties.

model

Return model if exists, otherwise raise error.

model_classes

Return ordered list of possible label classes for classification tasks.

model_name

Return the name of the model.

observed_classes

Return the observed classes in both train and test.

task_type

Return task type if model & train_dataset & label exists.

test

Return test if exists, otherwise raise error.

train

Return train if exists, otherwise raise error.

with_display

Return the with_display flag.

Methods

assert_task_type(*expected_types)	Assert task_type matching given types.
finalize_check_result(check_result, check[, ...])	Run final processing on a check result which includes validation, conditions processing and sampling footnote.
get_data_by_kind(kind)	Return the relevant Dataset by given kind.
get_scorers([scorers, use_avg_defaults])	Return initialized & validated scorers if provided or default scorers otherwise.
get_single_scorer([scorer, use_avg_defaults])	Return initialized & validated scorer if provided or a default scorer otherwise.
have_test()	Return whether there is test_dataset dataset defined.
infer_task_type(train_dataset, test_dataset)	Infer the task type.
is_multi_label_task()	Return whether the task is multi-label classification.
raise_if_multi_label_task([check])	Raise an exception if it is a multi-label classification task.
raise_if_token_classification_task([check])	Raise an exception if it is a token classification task.

__init__(train_dataset: Optional[TextData] = None, test_dataset: Optional[TextData] = None, with_display: bool = True, train_pred: Optional[Union[Sequence[int], Sequence[str], Sequence[Sequence[int]], Sequence[Sequence[str]]]] = None, test_pred: Optional[Union[Sequence[int], Sequence[str], Sequence[Sequence[int]], Sequence[Sequence[str]]]] = None, train_proba: Optional[Sequence[Sequence[float]]] = None, test_proba: Optional[Sequence[Sequence[float]]] = None, model_classes: Optional[List] = None, random_state: int = 42)

assert_task_type(*expected_types: TaskType)

Assert task_type matching given types.

If task_type is defined, validate it and raise error if needed, else returns True. If task_type is not defined, return False.

property common_datasets_properties: Dict[str, str]

Return common for train and test datasets properties.

finalize_check_result(check_result, check, dataset_kind: Optional[DatasetKind] = None)

Run final processing on a check result which includes validation, conditions processing and sampling footnote.

get_data_by_kind(kind: DatasetKind)

Return the relevant Dataset by given kind.

get_scorers(scorers: Optional[Union[Mapping[str, Union[str, Callable]], List[str]]] = None, use_avg_defaults=True) → List[DeepcheckScorer]

Return initialized & validated scorers if provided or default scorers otherwise.

Parameters

scorersUnion[List[str], Dict[str, Union[str, Callable]]], default: None

List of scorers to use. If None, use default scorers. Scorers can be supplied as a list of scorer names or as a dictionary of names and functions.

use_avg_defaultsbool, default: True

If no scorers were provided, for classification, determines whether to use default scorers that return an averaged metric, or default scorers that return a metric per class.

Returns

——-

List[DeepcheckScorer]

A list of initialized & validated scorers.

get_single_scorer(scorer: Optional[Union[Mapping[str, Union[str, Callable]], List[str]]] = None, use_avg_defaults=True) → DeepcheckScorer

Return initialized & validated scorer if provided or a default scorer otherwise.

Parameters

scorert.Union[t.Mapping[str, t.Union[str, t.Callable]], t.List[str], None], default: None

List of scorers to use. If None, use default scorers. Scorers can be supplied as a list of scorer names or as a dictionary of names and functions.

use_avg_defaultsbool, default: True

If no scorers were provided, for classification, determines whether to use default scorers that return an averaged metric, or default scorers that return a metric per class.

Returns

——-

List[DeepcheckScorer]

An initialized & validated scorer.

have_test()

Return whether there is test_dataset dataset defined.

static infer_task_type(train_dataset: TextData, test_dataset: TextData)

Infer the task type.

is_multi_label_task()

Return whether the task is multi-label classification.

property model: deepchecks.nlp.context._DummyModel

Return model if exists, otherwise raise error.

property model_classes: List

Return ordered list of possible label classes for classification tasks.

property model_name: str

Return the name of the model.

property observed_classes: List

Return the observed classes in both train and test.

raise_if_multi_label_task(check=None)

Raise an exception if it is a multi-label classification task.

raise_if_token_classification_task(check=None)

Raise an exception if it is a token classification task.

property task_type: deepchecks.nlp.task_type.TaskType

Return task type if model & train_dataset & label exists. otherwise, raise error.

property test

Return test if exists, otherwise raise error.

property train

Return train if exists, otherwise raise error.

property with_display: bool

Return the with_display flag.

class SingleDatasetCheck

Parent class for checks that only use one dataset.

Methods

add_condition(name, condition_func, **params)	Add new condition function to the check.
clean_conditions()	Remove all conditions from this check instance.
conditions_decision(result)	Run conditions on given result.
config([include_version, include_defaults])	Return check configuration (conditions' configuration not yet supported).
context_type	alias of Context
from_config(conf[, version_unmatch])	Return check object from a CheckConfig object.
from_json(conf[, version_unmatch])	Deserialize check instance from JSON string.
metadata([with_doc_link])	Return check metadata.
name()	Name of class in split camel case.
params([show_defaults])	Return parameters to show when printing the check.
remove_condition(index)	Remove given condition by index.
run(dataset[, model, with_display, ...])	Run check.
run_logic(context, dataset_kind)	Run check.
to_json([indent, include_version, ...])	Serialize check instance to JSON string.

__init__(**kwargs)

add_condition(name: str, condition_func: Callable[[Any], Union[ConditionResult, bool]], **params)

Add new condition function to the check.

Parameters

namestr

Name of the condition. should explain the condition action and parameters

condition_funcCallable[[Any], Union[List[ConditionResult], bool]]

Function which gets the value of the check and returns object of List[ConditionResult] or boolean.

paramsdict

Additional parameters to pass when calling the condition function.

clean_conditions()

Remove all conditions from this check instance.

conditions_decision(result: CheckResult) → List[ConditionResult]

Run conditions on given result.

config(include_version: bool = True, include_defaults: bool = True) → CheckConfig

Return check configuration (conditions’ configuration not yet supported).

Returns

CheckConfig

includes the checks class name, params, and module name.

context_type

alias of Context

classmethod from_config(conf: ~deepchecks.core.checks.CheckConfig, version_unmatch: ~typing.Optional[~typing.Union[typing_extensions.Literal[raise], typing_extensions.Literal[warn]]] = 'warn') → Self

Return check object from a CheckConfig object.

Parameters

confDict[Any, Any]

Returns

BaseCheck

the check class object from given config

classmethod from_json(conf: str, version_unmatch: ~typing.Optional[~typing.Union[typing_extensions.Literal[raise], typing_extensions.Literal[warn]]] = 'warn') → Self

Deserialize check instance from JSON string.

metadata(with_doc_link: bool = False) → CheckMetadata

Return check metadata.

Parameters

with_doc_linkbool, default False

whethere to include doc link in summary or not

Returns

Dict[str, Any]

classmethod name() → str

Name of class in split camel case.

params(show_defaults: bool = False) → Dict

Return parameters to show when printing the check.

remove_condition(index: int)

Remove given condition by index.

Parameters

indexint

index of condtion to remove

run(dataset: TextData, model=None, with_display: bool = True, predictions: Optional[Union[Sequence[int], Sequence[str], Sequence[Sequence[int]], Sequence[Sequence[str]]]] = None, probabilities: Optional[Sequence[Sequence[float]]] = None, model_classes: Optional[List] = None, random_state: int = 42) → CheckResult

Run check.

Parameters

dataset: TextData

Dataset representing data an estimator was fitted on

model: object, default: None

Dummy model object for compatibility with SingleDatasetBaseCheck

with_displaybool , default: True

flag that determines if checks will calculate display (redundant in some checks).

predictions: Union[TTextPred, None] , default: None

predictions on dataset

probabilities: Union[TTextProba, None] , default: None

probabilities on dataset

model_classes: Optional[List], default: None

For classification: list of classes known to the model

random_stateint, default 42

A seed to set for pseudo-random functions, primarily sampling.

Notes

The accepted formats for providing model predictions and probabilities are detailed below

Text Classification

Single Class Predictions

• predictions - A sequence of class names or indices with one entry per sample, matching the set of classes present in the labels.

• probabilities - A sequence of sequences with each element containing the vector of class probabilities for each sample. Each such vector should have one probability per class according to the class (sorted) order, and the probabilities should sum to 1 for each sample.

Multilabel Predictions

• predictions - A sequence of sequences with each element containing a binary vector denoting the presence of the i-th class for the given sample. Each such vector should have one binary indicator per class according to the class (sorted) order. More than one class can be present for each sample.

• probabilities - A sequence of sequences with each element containing the vector of class probabilities for each sample. Each such vector should have one probability per class according to the class (sorted) order, and the probabilities should range from 0 to 1 for each sample, but are not required to sum to 1.

Token Classification

• predictions - A sequence of sequences, with the inner sequence containing tuples in the following format: (class_name, span_start, span_end, class_probability). span_start and span_end are the start and end character indices of the token within the text, as it was passed to the raw_text argument. Each upper level sequence contains a sequence of tokens for each sample.

• probabilities - No probabilities should be passed for Token Classification tasks. Passing probabilities will result in an error.

Examples

Text Classification

Single Class Predictions

>>> predictions = ['class_1', 'class_1', 'class_2']
>>> probabilities = [[0.2, 0.8], [0.5, 0.5], [0.3, 0.7]]

Multilabel Predictions

>>> predictions = [[0, 0, 1], [0, 1, 1]]
>>> probabilities = [[0.2, 0.3, 0.8], [0.4, 0.9, 0.6]]

Token Classification

>>> predictions = [[('class_1', 0, 2, 0.8), ('class_2', 7, 10, 0.9)], [('class_2', 42, 54, 0.4)], []]

abstract run_logic(context, dataset_kind) → CheckResult

Run check.

to_json(indent: int = 3, include_version: bool = True, include_defaults: bool = True) → str

Serialize check instance to JSON string.

class TrainTestCheck

Parent class for checks that compare two datasets.

The class checks train dataset and test_dataset dataset for model training and test_dataset.

Methods

add_condition(name, condition_func, **params)	Add new condition function to the check.
clean_conditions()	Remove all conditions from this check instance.
conditions_decision(result)	Run conditions on given result.
config([include_version, include_defaults])	Return check configuration (conditions' configuration not yet supported).
context_type	alias of Context
from_config(conf[, version_unmatch])	Return check object from a CheckConfig object.
from_json(conf[, version_unmatch])	Deserialize check instance from JSON string.
metadata([with_doc_link])	Return check metadata.
name()	Name of class in split camel case.
params([show_defaults])	Return parameters to show when printing the check.
remove_condition(index)	Remove given condition by index.
run(train_dataset, test_dataset[, model, ...])	Run check.
run_logic(context)	Run check.
to_json([indent, include_version, ...])	Serialize check instance to JSON string.

__init__(**kwargs)

add_condition(name: str, condition_func: Callable[[Any], Union[ConditionResult, bool]], **params)

Add new condition function to the check.

Parameters

namestr

Name of the condition. should explain the condition action and parameters

condition_funcCallable[[Any], Union[List[ConditionResult], bool]]

Function which gets the value of the check and returns object of List[ConditionResult] or boolean.

paramsdict

Additional parameters to pass when calling the condition function.

clean_conditions()

Remove all conditions from this check instance.

conditions_decision(result: CheckResult) → List[ConditionResult]

Run conditions on given result.

config(include_version: bool = True, include_defaults: bool = True) → CheckConfig

Return check configuration (conditions’ configuration not yet supported).

Returns

CheckConfig

includes the checks class name, params, and module name.

context_type

alias of Context

classmethod from_config(conf: ~deepchecks.core.checks.CheckConfig, version_unmatch: ~typing.Optional[~typing.Union[typing_extensions.Literal[raise], typing_extensions.Literal[warn]]] = 'warn') → Self

Return check object from a CheckConfig object.

Parameters

confDict[Any, Any]

Returns

BaseCheck

the check class object from given config

classmethod from_json(conf: str, version_unmatch: ~typing.Optional[~typing.Union[typing_extensions.Literal[raise], typing_extensions.Literal[warn]]] = 'warn') → Self

Deserialize check instance from JSON string.

metadata(with_doc_link: bool = False) → CheckMetadata

Return check metadata.

Parameters

with_doc_linkbool, default False

whethere to include doc link in summary or not

Returns

Dict[str, Any]

classmethod name() → str

Name of class in split camel case.

params(show_defaults: bool = False) → Dict

Return parameters to show when printing the check.

remove_condition(index: int)

Remove given condition by index.

Parameters

indexint

index of condtion to remove

run(train_dataset: TextData, test_dataset: TextData, model=None, with_display: bool = True, train_predictions: Optional[Union[Sequence[int], Sequence[str], Sequence[Sequence[int]], Sequence[Sequence[str]]]] = None, test_predictions: Optional[Union[Sequence[int], Sequence[str], Sequence[Sequence[int]], Sequence[Sequence[str]]]] = None, train_probabilities: Optional[Sequence[Sequence[float]]] = None, test_probabilities: Optional[Sequence[Sequence[float]]] = None, model_classes: Optional[List] = None, random_state: int = 42) → CheckResult

Run check.

Parameters

train_dataset: Union[TextData, None] , default: None

TextData object, representing data an estimator was fitted on

test_dataset: Union[TextData, None] , default: None

TextData object, representing data an estimator predicts on

model: object, default: None

Dummy model object for compatibility with TrainTestBaseCheck

with_displaybool , default: True

flag that determines if checks will calculate display (redundant in some checks).

train_predictions: Union[TTextPred, None] , default: None

predictions on train dataset

test_predictions: Union[TTextPred, None] , default: None

predictions on test_dataset dataset

train_probabilities: Union[TTextProba, None] , default: None

probabilities on train dataset

test_probabilities: Union[TTextProba, None] , default: None

probabilities on test_dataset dataset

model_classes: Optional[List], default: None

For classification: list of classes known to the model

random_stateint, default 42

A seed to set for pseudo-random functions, primarily sampling.

Notes

The accepted formats for providing model predictions and probabilities are detailed below

Text Classification

Single Class Predictions

• predictions - A sequence of class names or indices with one entry per sample, matching the set of classes present in the labels.

• probabilities - A sequence of sequences with each element containing the vector of class probabilities for each sample. Each such vector should have one probability per class according to the class (sorted) order, and the probabilities should sum to 1 for each sample.

Multilabel Predictions

• predictions - A sequence of sequences with each element containing a binary vector denoting the presence of the i-th class for the given sample. Each such vector should have one binary indicator per class according to the class (sorted) order. More than one class can be present for each sample.

• probabilities - A sequence of sequences with each element containing the vector of class probabilities for each sample. Each such vector should have one probability per class according to the class (sorted) order, and the probabilities should range from 0 to 1 for each sample, but are not required to sum to 1.

Token Classification

• predictions - A sequence of sequences, with the inner sequence containing tuples in the following format: (class_name, span_start, span_end, class_probability). span_start and span_end are the start and end character indices of the token within the text, as it was passed to the raw_text argument. Each upper level sequence contains a sequence of tokens for each sample.

• probabilities - No probabilities should be passed for Token Classification tasks. Passing probabilities will result in an error.

Examples

Text Classification

Single Class Predictions

>>> predictions = ['class_1', 'class_1', 'class_2']
>>> probabilities = [[0.2, 0.8], [0.5, 0.5], [0.3, 0.7]]

Multilabel Predictions

>>> predictions = [[0, 0, 1], [0, 1, 1]]
>>> probabilities = [[0.2, 0.3, 0.8], [0.4, 0.9, 0.6]]

Token Classification

>>> predictions = [[('class_1', 0, 2, 0.8), ('class_2', 7, 10, 0.9)], [('class_2', 42, 54, 0.4)], []]

abstract run_logic(context) → CheckResult

Run check.

to_json(indent: int = 3, include_version: bool = True, include_defaults: bool = True) → str

Serialize check instance to JSON string.

class Suite

NLP suite to run checks of types: TrainTestCheck, SingleDatasetCheck, ModelOnlyCheck.

Methods

add(check)	Add a check or a suite to current suite.
config()	Return suite configuration (checks' conditions' configuration not yet supported).
from_config(conf[, version_unmatch])	Return suite object from a CheckConfig object.
from_json(conf[, version_unmatch])	Deserialize suite instance from JSON string.
remove(index)	Remove a check by given index.
run([train_dataset, test_dataset, ...])	Run all checks.
supported_checks()	Return tuple of supported check types of this suite.
to_json([indent])	Serialize suite instance to JSON string.

__init__(name: str, *checks: Union[BaseCheck, BaseSuite])

add(check: Union[BaseCheck, BaseSuite])

Add a check or a suite to current suite.

Parameters

checkBaseCheck

A check or suite to add.

config() → SuiteConfig

Return suite configuration (checks’ conditions’ configuration not yet supported).

Returns

SuiteConfig

includes the suite name, and list of check configs.

classmethod from_config(conf: ~deepchecks.core.suite.SuiteConfig, version_unmatch: ~typing.Optional[~typing.Union[typing_extensions.Literal[raise], typing_extensions.Literal[warn]]] = 'warn') → Self

Return suite object from a CheckConfig object.

Parameters

confSuiteConfig

the SuiteConfig object

Returns

BaseSuite

the suite class object from given config

from_json(conf: str, version_unmatch: ~typing.Optional[~typing.Union[typing_extensions.Literal[raise], typing_extensions.Literal[warn]]] = 'warn') → Self

Deserialize suite instance from JSON string.

remove(index: int)

Remove a check by given index.

Parameters

indexint

Index of check to remove.

run(train_dataset: Optional[TextData] = None, test_dataset: Optional[TextData] = None, with_display: bool = True, train_predictions: Optional[Union[Sequence[int], Sequence[str], Sequence[Sequence[int]], Sequence[Sequence[str]]]] = None, test_predictions: Optional[Union[Sequence[int], Sequence[str], Sequence[Sequence[int]], Sequence[Sequence[str]]]] = None, train_probabilities: Optional[Sequence[Sequence[float]]] = None, test_probabilities: Optional[Sequence[Sequence[float]]] = None, model_classes: Optional[List] = None, random_state: int = 42) → SuiteResult

Run all checks.

Parameters

train_dataset: Union[TextData, None] , default: None

TextData object, representing data an estimator was fitted on

test_dataset: Union[TextData, None] , default: None

TextData object, representing data an estimator predicts on

with_displaybool , default: True

flag that determines if checks will calculate display (redundant in some checks).

train_predictions: Union[TTextPred, None] , default: None

predictions on train dataset

test_predictions: Union[TTextPred, None] , default: None

predictions on test dataset

train_probabilities: Union[TTextProba, None] , default: None

probabilities on train dataset

test_probabilities: Union[TTextProba, None] , default: None

probabilities on test_dataset dataset

model_classes: Optional[List], default: None

For classification: list of classes known to the model

random_stateint, default 42

A seed to set for pseudo-random functions, primarily sampling.

Returns

SuiteResult

All results by all initialized checks

Notes

The accepted formats for providing model predictions and probabilities are detailed below

Text Classification

Single Class Predictions

• predictions - A sequence of class names or indices with one entry per sample, matching the set of classes present in the labels.

• probabilities - A sequence of sequences with each element containing the vector of class probabilities for each sample. Each such vector should have one probability per class according to the class (sorted) order, and the probabilities should sum to 1 for each sample.

Multilabel Predictions

• predictions - A sequence of sequences with each element containing a binary vector denoting the presence of the i-th class for the given sample. Each such vector should have one binary indicator per class according to the class (sorted) order. More than one class can be present for each sample.

• probabilities - A sequence of sequences with each element containing the vector of class probabilities for each sample. Each such vector should have one probability per class according to the class (sorted) order, and the probabilities should range from 0 to 1 for each sample, but are not required to sum to 1.

Token Classification

• predictions - A sequence of sequences, with the inner sequence containing tuples in the following format: (class_name, span_start, span_end, class_probability). span_start and span_end are the start and end character indices of the token within the text, as it was passed to the raw_text argument. Each upper level sequence contains a sequence of tokens for each sample.

• probabilities - No probabilities should be passed for Token Classification tasks. Passing probabilities will result in an error.

Examples

Text Classification

Single Class Predictions

>>> predictions = ['class_1', 'class_1', 'class_2']
>>> probabilities = [[0.2, 0.8], [0.5, 0.5], [0.3, 0.7]]

Multilabel Predictions

>>> predictions = [[0, 0, 1], [0, 1, 1]]
>>> probabilities = [[0.2, 0.3, 0.8], [0.4, 0.9, 0.6]]

Token Classification

>>> predictions = [[('class_1', 0, 2, 0.8), ('class_2', 7, 10, 0.9)], [('class_2', 42, 54, 0.4)], []]

classmethod supported_checks() → Tuple

Return tuple of supported check types of this suite.

to_json(indent: int = 3) → str

Serialize suite instance to JSON string.

previous

Suite.to_json

next

checks