cleanlab
is a data-centric AI package for data quality and machine learning with messy, real-world data and labels.
cleanlab
clean
s lab
els and supports finding, quantifying, and learning with label issues in datasets. See datasets cleaned with cleanlab
at labelerrors.com.
Check out the: documentation, examples, and installation instructions
cleanlab
is powered by confident learning, published in this paper | blog.
Get started with tutorials
- (Easiest) Improve a simple classifier from 60% ---> 80% accuracy on the Iris dataset:
- (Comprehensive) Image classification with noisy labels
- Run Cleanlab on 4 datasets using 9 different classifiers/models:
- Find label errors in MNIST, ImageNet, CIFAR-10/100, Caltech-256, QuickDraw, Amazon Reviews, IMDB, 20 Newsgroups, AudioSet:
- Demystifying Confident Learning (the theory and algorithms underlying
cleanlab
):
News! (2021) -- cleanlab
finds pervasive label errors in the most common ML test sets (click to learn more)
- Dec 2021
🎉 NeurIPS published the label errors paper (Northcutt, Athalye, & Mueller, 2021). - Apr 2021
🎉 Journal of AI Research published the confident learning paper (Northcutt, Jiang, & Chuang, 2021). - Mar 2021
😲 cleanlab
used to find and fix label issues in 10 of the most common ML benchmark datasets, published in: NeurIPS 2021. Along with the paper (Northcutt, Athalye, & Mueller, 2021), the authors launched labelerrors.com where you can view the label issues in these datasets.
News! (2020) -- cleanlab
adds support for all OS, achieves state-of-the-art, supports co-teaching, and more (click to learn more)
- Dec 2020
🎉 cleanlab
supports NeurIPS workshop paper (Northcutt, Athalye, & Lin, 2020). - Dec 2020
🤖 cleanlab
supports PU learning. - Feb 2020
🤖 cleanlab
now natively supports Mac, Linux, and Windows. - Feb 2020
🤖 cleanlab
now supports Co-Teaching (Han et al., 2018). - Jan 2020
🎉 cleanlab
achieves state-of-the-art on CIFAR-10 with noisy labels. Code to reproduce: examples/cifar10. This is a great place to see how to use cleanlab on real datasets (with predicted probabiliteis already precomputed for you).
Past release notes and future features planned is available here.
cleanlab
So fresh, so cleanlab
finds and cleans label issues in any dataset using state-of-the-art algorithms to find label issues, characterize noise, and learn in spite of it. cleanlab
is fast: its built on optimized algorithms and parallelized across CPU threads automatically. cleanlab
is powered by provable guarantees of exact noise estimation and label error finding in realistic cases when model output probabilities are erroneous. cleanlab
supports multi-label, multiclass, sparse matrices, etc. By default, cleanlab
requires no hyper-parameters.
cleanlab
implements the family of theory and algorithms called confident learning with provable guarantees of exact noise estimation and label error finding (even when model output probabilities are noisy/imperfect).
cleanlab
supports many classification tasks: multi-label, multiclass, sparse matrices, etc.
cleanlab
is:
- backed-by-theory - Provable perfect label error finding in some realistic conditions.
- fast - Non-iterative, parallelized algorithms (e.g. < 1 second to find label issues in ImageNet with pre-computed probabilities)
- general - Works with any dataset, model, and framework, e.g., Tensorflow, PyTorch, sklearn, xgboost, etc.
Find label issues with PyTorch, Tensorflow, sklearn, xgboost, etc. in 1 line of code
# Compute pred_probs (n x m matrix of predicted probabilities) on your own, with any classifier.
# Be sure you compute probs in a holdout/out-of-sample manner (e.g. via cross-validation)
# Here is an example that shows in detail how to compute pred_probs on CIFAR-10:
# https://github.com/cleanlab/examples/tree/master/cifar10
# Now finding label issues is trivial with cleanlab... its one line of code.
# label issues are ordered by likelihood of being an error. First index is most likely error.
from cleanlab.filter import find_label_issues
ordered_label_issues = find_label_issues(
labels=numpy_array_of_noisy_labels,
pred_probs=numpy_array_of_predicted_probabilities,
return_indices_ranked_by='normalized_margin', # Orders label issues
)
CAUTION: Predicted probabilities from your model must be out-of-sample! You should never provide predictions on the same datapoints used to train the model, as these will be overfit and unsuitable for finding label-errors. To obtain out-of-sample predicted probabilities for every datapoint in your dataset, you can use cross-validation. Alternatively, you can train your model on a separate dataset and you are only evaluating labels in data that was previously held-out.
Pre-computed out-of-sample predicted probabilities for CIFAR-10 train set are available: here.
Learning with noisy labels in 3 lines of code
from cleanlab.classification import LearningWithNoisyLabels
from sklearn.linear_model import LogisticRegression
# Wrap around any classifier. Yup, you can use sklearn/pyTorch/Tensorflow/FastText/etc.
lnl = LearningWithNoisyLabels(clf=LogisticRegression())
lnl.fit(X=X_train_data, labels=train_noisy_labels)
# Estimate the predictions you would have gotten by training with *no* label issues.
predicted_test_labels = lnl.predict(X_test)
Check out these examples and tests (includes how to use other types of models).
Learn cleanlab in 5min
New to cleanlab
? Try out these easy tutorials:
- Simple example of learning with noisy labels on the Iris dataset (multiclass classification).
- Learning with noisy labels on CIFAR
cleanlab
with any model (Tensorflow, PyTorch, sklearn, xgboost, etc.)
Use All of the features of the cleanlab
package work with any model. Yes, any model. Feel free to use PyTorch, Tensorflow, caffe2, scikit-learn, mxnet, etc. If you use a scikit-learn classifier, all cleanlab
methods will work out-of-the-box. It’s also easy to use your favorite model from a non-scikit-learn package, just wrap your model into a Python class that inherits the sklearn.base.BaseEstimator
:
from sklearn.base import BaseEstimator
class YourFavoriteModel(BaseEstimator): # Inherits sklearn base classifier
def __init__(self, ):
pass
def fit(self, X, y, sample_weight=None):
pass
def predict(self, X):
pass
def predict_proba(self, X):
pass
def score(self, X, y, sample_weight=None):
pass
# Now you can use your model with `cleanlab`. Here's one example:
from cleanlab.classification import LearningWithNoisyLabels
lnl = LearningWithNoisyLabels(clf=YourFavoriteModel())
lnl.fit(train_data, train_labels_with_errors)
Here’s a compliant PyTorch MNIST CNN class
Want to see a working example?As you can see here, technically you don’t actually need to inherit from sklearn.base.BaseEstimator
, as you can just create a class that defines .fit()
, .predict()
, and .predict\_proba()
, but inheriting makes downstream scikit-learn applications like hyper-parameter optimization work seamlessly. See cleanlab.classification.LearningWithNoisyLabels() for a fully compliant model.
Note, some libraries exists to do this for you. For PyTorch, check out the skorch
Python library which will wrap your pytorch
model into a scikit-learn
compliant model.
Installation
Python 3.6+ are supported. Linux, macOS, and Windows are supported.
Stable release (pip):
$ pip install cleanlab # Using pip
Stable release (conda):
$ conda install -c cleanlab cleanlab # Using conda
Developer release:
$ pip install git+https://github.com/cleanlab/cleanlab.git
To install from the source code (enabling you to make modifications locally):
$ conda update pip # if you use conda
$ git clone https://github.com/cleanlab/cleanlab.git
$ cd cleanlab
$ pip install -e .
Confident Learning paper
Reproducing Results inState of the Art Learning with Noisy Labels in CIFAR
A step-by-step guide to reproduce these results is available here. This guide is also a good tutorial for using cleanlab on any large dataset. You'll need to git clone
confidentlearning-reproduce which contains the data and files needed to reproduce the CIFAR-10 results.
Comparison of confident learning (CL), as implemented in cleanlab
, versus seven recent methods for learning with noisy labels in CIFAR-10. Highlighted cells show CL robustness to sparsity. The five CL methods estimate label issues, remove them, then train on the cleaned data using Co-Teaching.
Observe how cleanlab (i.e. the CL method) is robust to large sparsity in label noise whereas prior art tends to reduce in performance for increased sparsity, as shown by the red highlighted regions. This is important because real-world label noise is often sparse, e.g. a tiger is likely to be mislabeled as a lion, but not as most other classes like airplane, bathtub, and microwave.
Find label issues in ImageNet
Use cleanlab
to identify ~100,000 label errors in the 2012 ILSVRC ImageNet training dataset: examples/imagenet.
Label issues in ImageNet train set found via cleanlab
. Label Errors are boxed in red. Ontological issues in green. Multi-label images in blue.
Find Label Errors in MNIST
Use cleanlab
to identify ~50 label errors in the MNIST dataset: examples/mnist.
Top 24 least-confident labels in the original MNIST train dataset, algorithmically identified via cleanlab
. Examples are ordered left-right, top-down by increasing self-confidence (predicted probability that the given label is correct), denoted conf in teal. The most-likely correct label (with largest predicted probability) is in green. Overt label errors highlighted in red.
cleanlab
Performance across 4 Data Distributions and 9 Classifiers
cleanlab
is a general tool that can learn with noisy labels regardless of dataset distribution or classifier type: examples/classifier_comparison.
Each sub-figure above depicts the decision boundary learned using cleanlab.classification.LearningWithNoisyLabels in the presence of extreme (~35%) label errors (circled in green). Label noise is class-conditional (not uniformly random). Columns are organized by the classifier used, except the left-most column which depicts the ground-truth data distribution. Rows are organized by dataset.
Each sub-figure depicts accuracy scores on a test set (with correct non-noisy labels) as decimal values:
- LEFT (in black): The classifier test accuracy trained with perfect labels (no label errors).
- MIDDLE (in blue): The classifier test accuracy trained with noisy labels using
cleanlab
. - RIGHT (in white): The baseline classifier test accuracy trained with noisy labels.
As an example, the table below is the noise matrix (noisy channel) *P(s | y) characterizing the label noise for the first dataset row in the figure. s represents the observed noisy labels and y represents the latent, true labels. The trace of this matrix is 2.6. A trace of 4 implies no label noise. A cell in this matrix is read like: "Around 38% of true underlying '3' labels were randomly flipped to '2' labels in the observed dataset."
p(label︱y) |
y=0 | y=1 | y=2 | y=3 |
---|---|---|---|---|
label=0 | 0.55 | 0.01 | 0.07 | 0.06 |
label=1 | 0.22 | 0.87 | 0.24 | 0.02 |
label=2 | 0.12 | 0.04 | 0.64 | 0.38 |
label=3 | 0.11 | 0.08 | 0.05 | 0.54 |
ML Research with cleanlab
cleanlab
Core Package Components
- cleanlab/classification.py - LearningWithNoisyLabels() class for learning with noisy labels.
- cleanlab/count.py - Estimates and fully characterizes all variants of label noise.
- cleanlab/noise_generation.py - Generate mathematically valid synthetic noise matrices.
- cleanlab/filter.py - Finds the examples with label issues in a dataset.
- cleanlab/rank.py - Rank every example in a dataset with various label quality scores.
Many methods have default parameters not covered here. Check out the method docstrings and our full documentation.
For additional details/notation, refer to the Confident Learning paper.
Methods to Standardize Research with Noisy Labels
cleanlab
supports a number of functions to generate noise for benchmarking and standardization in research. This next example shows how to generate valid, class-conditional, unformly random noisy channel matrices:
# Generate a valid (necessary conditions for learnability are met) noise matrix for any trace > 1
from cleanlab.noise_generation import generate_noise_matrix_from_trace
noise_matrix=generate_noise_matrix_from_trace(
K=number_of_classes,
trace=float_value_greater_than_1_and_leq_K,
py=prior_of_y_actual_labels_which_is_just_an_array_of_length_K,
frac_zero_noise_rates=float_from_0_to_1_controlling_sparsity,
)
# Check if a noise matrix is valid (necessary conditions for learnability are met)
from cleanlab.noise_generation import noise_matrix_is_valid
is_valid=noise_matrix_is_valid(
noise_matrix,
prior_of_y_which_is_just_an_array_of_length_K,
)
For a given noise matrix, this example shows how to generate noisy labels. Methods can be seeded for reproducibility.
# Generate noisy labels using the noise_marix. Guarantees exact amount of noise in labels.
from cleanlab.noise_generation import generate_noisy_labels
s_noisy_labels = generate_noisy_labels(y_hidden_actual_labels, noise_matrix)
# This package is a full of other useful methods for learning with noisy labels.
# The tutorial stops here, but you don't have to. Inspect method docstrings for full docs.
Fully characterize label noise and uncertainty in your dataset.
s denotes a random variable that represents the observed, noisy label and y denotes a random variable representing the hidden, actual labels. Both s and y take any of the m classes as values. The cleanlab
package supports different levels of granularity for computation depending on the needs of the user. Because of this, we support multiple alternatives, all no more than a few lines, to estimate these latent distribution arrays, enabling the user to reduce computation time by only computing what they need to compute, as seen in the examples below.
Throughout these examples, you’ll see a variable called confident_joint. The confident joint is an m x m matrix (m is the number of classes) that counts, for every observed, noisy class, the number of examples that confidently belong to every latent, hidden class. It counts the number of examples that we are confident are labeled correctly or incorrectly for every pair of obseved and unobserved classes. The confident joint is an unnormalized estimate of the complete-information latent joint distribution, Ps,y.
The label flipping rates are denoted P(s | y), the inverse rates are P(y | s), and the latent prior of the unobserved, true labels, p(y).
Most of the methods in the cleanlab package start by first estimating the confident_joint. You can learn more about this in the confident learning paper.
Option 1: Compute the confident joint and predicted probs first. Stop if that’s all you need.
from cleanlab.count import estimate_latent
from cleanlab.count import estimate_confident_joint_and_cv_pred_proba
# Compute the confident joint and the n x m predicted probabilities matrix (pred_probs),
# for n examples, m classes. Stop here if all you need is the confident joint.
confident_joint, pred_probs = estimate_confident_joint_and_cv_pred_proba(
X=X_train,
labels=train_labels_with_errors,
clf=logreg(), # default, you can use any classifier
)
# Estimate latent distributions: p(y) as est_py, P(s|y) as est_nm, and P(y|s) as est_inv
est_py, est_nm, est_inv = estimate_latent(
confident_joint,
labels=train_labels_with_errors,
)
Option 2: Estimate the latent distribution matrices in a single line of code.
from cleanlab.count import estimate_py_noise_matrices_and_cv_pred_proba
est_py, est_nm, est_inv, confident_joint, pred_probs = estimate_py_noise_matrices_and_cv_pred_proba(
X=X_train,
labels=train_labels_with_errors,
)
Option 3: Skip computing the predicted probabilities if you already have them.
# Already have pred_probs? (n x m matrix of predicted probabilities)
# For example, you might get them from a pre-trained model (like resnet on ImageNet)
# With the cleanlab package, you estimate directly with pred_probs.
from cleanlab.count import estimate_py_and_noise_matrices_from_probabilities
est_py, est_nm, est_inv, confident_joint = estimate_py_and_noise_matrices_from_probabilities(
labels=train_labels_with_errors,
pred_probs=pred_probs,
)
Completely characterize label noise in a dataset:
The joint probability distribution of noisy and true labels, P(s,y), completely characterizes label noise with a class-conditional m x m matrix.
from cleanlab.count import estimate_joint
joint = estimate_joint(
labels=noisy_labels,
pred_probs=probabilities,
confident_joint=None, # Provide if you have it already
)
PU learning with cleanlab:
Positive-Unlabeled learning (in which your data only contains a few positively labeled examples with the rest unlabeled) is just a special case of LearningWithNoisyLabels when one of the classes has no error. P
stands for the positive class and is assumed to have zero label errors and U
stands for unlabeled data, but in practice, we just assume the U
class is a noisy negative class that actually contains some positive examples. Thus, the goal of PU learning is to (1) estimate the proportion of negatively labeled datapoints that actually belong to the positive class (seefraction\_noise\_in\_unlabeled\_class
in the last example), (2) find the errors (see last example), and (3) train on clean data (see first example below). cleanlab
does all three, taking into account that there are no label errors in whichever class you specify as positive.
There are two ways to use cleanlab
for PU learning. We'll look at each here.
Method 1. If you are using the cleanlab classifier LearningWithNoisyLabels(), and your dataset has exactly two classes (positive = 1, and negative = 0), PU
learning is supported directly in cleanlab
. You can perform PU learning like this:
from cleanlab.classification import LearningWithNoisyLabels
from sklearn.linear_model import LogisticRegression
# Wrap around any classifier. Yup, you can use sklearn/pyTorch/Tensorflow/FastText/etc.
pu_class = 0 # Should be 0 or 1. Label of class with NO ERRORS. (e.g., P class in PU)
lnl = LearningWithNoisyLabels(clf=LogisticRegression(), pulearning=pu_class)
lnl.fit(X=X_train_data, labels=train_noisy_labels)
# Estimate the predictions you would have gotten by training with *no* label errors.
predicted_test_labels = lnl.predict(X_test)
Method 2. However, you might be using a more complicated classifier that doesn't work well with LearningWithNoisyLabels (see this example for CIFAR-10). Or you might have 3 or more classes. Here's how to use cleanlab for PU learning in this situation. To let cleanlab know which class has no error (in standard PU learning, this is the P class), you need to set the threshold for that class to 1 (1 means the probabilitythat the labels of that class are correct is 1, i.e. that class has no error). Here's the code:
import numpy as np
# K is the number of classes in your dataset
# pred_probs are the cross-validated predicted probabilities.
# s is the array/list/iterable of noisy labels
# pu_class is a 0-based integer for the class that has no label errors.
thresholds = np.asarray([np.mean(pred_probs[:, k][s == k]) for k in range(K)])
thresholds[pu_class] = 1.0
Now you can use cleanlab however you were before. Just be sure to pass in this thresholds parameter wherever it applies. For example:
# Uncertainty quantification (characterize the label noise
# by estimating the joint distribution of noisy and true labels)
cj = compute_confident_joint(s, pred_probs, thresholds=thresholds, )
# Now the noise (cj) has been estimated taking into account that some class(es) have no error.
# We can use cj to find label errors like this:
indices_of_label_issues = find_label_issues(s, pred_probs, confident_joint=cj, )
# In addition to label issues, we can find the fraction of noise in the unlabeled class.
# First we need the inv_noise_matrix which contains P(y|s) (proportion of mislabeling).
_, _, inv_noise_matrix = estimate_latent(confident_joint=cj, labels=s, )
# Because inv_noise_matrix contains P(y|s), p (y = anything | labels = pu_class) should be 0
# because the prob(true label is something else | example is in pu_class) is 0.
# What's more interesting is p(y = anything | s is not put_class), or in the binary case
# this translates to p(y = pu_class | s = 1 - pu_class) because pu_class is 0 or 1.
# So, to find the fraction_noise_in_unlabeled_class, for binary, you just compute:
fraction_noise_in_unlabeled_class = inv_noise_matrix[pu_class][1 - pu_class]
Now that you have indices\_of\_label\_errors
, you can remove those label issues and train on clean data (or only remove some of the label issues and iteratively use confident learning / cleanlab to improve results).
Citation and Related Publications
cleanlab
isn't just a github, it's based on peer-reviewed research. Here are the relevant papers to cite if you use this package:
@article{northcutt2021confidentlearning,
title={Confident Learning: Estimating Uncertainty in Dataset Labels},
author={Curtis G. Northcutt and Lu Jiang and Isaac L. Chuang},
journal={Journal of Artificial Intelligence Research (JAIR)},
volume={70},
pages={1373--1411},
year={2021}
}
If you use this package for binary classification or PU learning, please also cite the rankpruning paper:
@inproceedings{northcutt2017rankpruning,
author={Northcutt, Curtis G. and Wu, Tailin and Chuang, Isaac L.},
title={Learning with Confident Examples: Rank Pruning for Robust Classification with Noisy Labels},
booktitle = {Proceedings of the Thirty-Third Conference on Uncertainty in Artificial Intelligence},
series = {UAI'17},
year = {2017},
location = {Sydney, Australia},
numpages = {10},
url = {http://auai.org/uai2017/proceedings/papers/35.pdf},
publisher = {AUAI Press},
}
Other Resources
- Blogpost: Introduction to Confident Learning
- NeurIPS 2021 paper: Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks
Join our Community
Have ideas for the future of cleanlab? How are you using cleanlab? Join the discussion.
Have code improvements for cleanlab? Submit a code pull request.
Do you have an issue with cleanlab? Submit an issue.
License
Copyright (c) 2017-2022 Cleanlab Inc.
cleanlab is free software: you can redistribute it and/or modify it under the terms of the GNU Affero General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
cleanlab is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
See GNU Affero General Public LICENSE for details.