reconer.insights¶
The reconer.insights module provides more complex functionality for understanding your dataset.
It provides functions for identifying disparities in your annotations and identifying the kinds of examples and labels
that are hardest for your model to identify.
Some of the functionality in reconer.insights require a reconer.recognizer.EntityRecognizer object.
You can read more about the EntityRecognizer class here: Tutorial - Custom EntityRecognizer
API¶
reconer.insights.ents_by_label(data, use_lower=True)Get a dictionary of unique text spans by label for your data
Parameters¶
data: (List[Example]), required.
List of Examples
use_lower: (bool, optional), Defaults to True.
Use the lowercase form of the span text
Returns¶
(DefaultDict[str, List[str]]):
DefaultDict mapping label to sorted list of the unique
spans annotated for that label.
reconer.insights.get_label_disparities(data, label1, label2, use_lower=True)Identify annotated spans that have different labels in different examples
Parameters¶
data: (List[Example]), required.
Input List of Examples
label1: (str), required.
First label to compare
label2: (str), required.
Second label to compare
Returns¶
(Set[str]):
Set of all unique text spans that overlap between label1 and label2
reconer.insights.top_prediction_errors(ner, data, labels=None, k=None, exclude_fp=False, exclude_fn=False)Get a sorted list of examples your model is worst at predicting.
Parameters¶
ner: (EntityRecognizer), required.
An instance of EntityRecognizer
data: (List[Example]), required.
List of annotated Examples
labels: (List[str], optional), Defaults to None.
List of labels to get errors for. Defaults to the labels property of ner.
k: (int, optional), Defaults to None.
If set, return the top k prediction errors, otherwise the whole list.
exclude_fp: (bool, optional), Defaults to False.
Flag to exclude False Positive errors.
exclude_fn: (bool, optional), Defaults to False.
Flag to exclude False Negative errors.
Returns¶
(List[PredictionError]):
[description]