A lab for bad labels. Learn more here.

This repository contains general tricks that may help you find bad, or noisy, labels in your dataset. The hope is that this repository makes it easier for folks to quickly check their own datasets before they invest too much time and compute on gridsearch.

You can install the tool via pip or conda.

Install with pip

python -m pip install doubtlab

Install with conda

conda install -c conda-forge doubtlab

Doubtlab allows you to define "reasons" for a row of data to deserve another look. These reasons can form a pipeline which can be used to retreive a sorted list of examples worth checking again.

from sklearn.datasets import load_iris
from sklearn.linear_model import LogisticRegression

from doubtlab.ensemble import DoubtEnsemble
from doubtlab.reason import ProbaReason, WrongPredictionReason

# Let's say we have some dataset/model already
X, y = load_iris(return_X_y=True)
model = LogisticRegression(max_iter=1_000)
model.fit(X, y)

# Next we can add reasons for doubt. In this case we're saying
# that examples deserve another look if the associated proba values
# are low or if the model output doesn't match the associated label.
reasons = {
    'proba': ProbaReason(model=model),
    'wrong_pred': WrongPredictionReason(model=model)
}

# Pass these reasons to a doubtlab instance.
doubt = DoubtEnsemble(**reasons)

# Get the ordered indices of examples worth checking again
indices = doubt.get_indices(X, y)
# Get dataframe with "reason"-ing behind the sorting
predicates = doubt.get_predicates(X, y)

The library implemented many "reasons" for doubt.

• RandomReason: assign doubt randomly, just for sure
• OutlierReason: assign doubt when the model declares a row an outlier

• ProbaReason: assign doubt when a models' confidence-values are low for any label
• WrongPredictionReason: assign doubt when a model cannot predict the listed label
• ShortConfidenceReason: assign doubt when the correct label gains too little confidence
• LongConfidenceReason: assign doubt when a wrong label gains too much confidence
• DisagreeReason: assign doubt when two models disagree on a prediction
• CleanlabReason: assign doubt according to cleanlab
• MarginConfidenceReason: assign doubt when there's a small difference between the top two class confidences

• AbsoluteDifferenceReason: assign doubt when the absolute difference is too high
• RelativeDifferenceReason: assign doubt when the relative difference is too high
• StandardizedErrorReason: assign doubt when the absolute standardized residual is too high

It is early days for the project. The project should be plenty useful as-is, but we prefer to be honest. Feedback and anecdotes are very welcome!

• The cleanlab project was an inspiration for this one. They have a great heuristic for bad label detection but I wanted to have a library that implements many. Be sure to check out their work on the labelerrors.com project.
• My former employer, Rasa, has always had a focus on data quality. Some of that attitude is bound to have seeped in here. Be sure to check the Conversation Driven Development approach and Rasa X if you're working on virtual assistants.
• My current employer, Explosion, has a neat labelling tool called Prodigy. I'm currently investigating how tools like doubtlab might lead to better labels when combined with this (very like-able) annotation tool.