Researchers Release Cleanlab 2.0: An Open-Source Python Framework For Machine Learning And Analytics With Messy, Real-World Data

This Article Is Based On The Cleanlab article 'cleanlab 2.0: Automatically Find Errors in ML Datasets'. All Credit For This Research Goes To The Cleanlab Researchers. 👏👏👏

Please Don't Forget To Join Our ML Subreddit

Data preparation is the most time-consuming and hectic process in data science and machine learning, accounting for 80% of the labor. Messy data is a serious issue that costs businesses trillions of dollars every year.

Model performance can be harmed by data errors (for example, mislabeled samples in the training set) and dataset-level concerns like overlapping classes. Most test set errors are ubiquitous even in gold-standard benchmark datasets. This can cause data scientists to deploy worse models.

Although physically analyzing and cleaning up individual data points sounds tiresome, it frequently gives a significantly bigger payback than experimenting with advanced modeling approaches. Automatically identifying the tiny fraction of noise can considerably lessen the discomfort in this process.

Cleanlab researchers has recently open-sourced Cleanlab 2.0 Python package that detects flaws in datasets, assesses dataset quality, trains trustworthy models with noisy data, and assists in curating high-quality datasets, all with only a few lines of code. It is a data-centric tool that provides clean labels during training. 

Engineers used Cleanlab at:

  • Google to clean and train robust models on speech data
  • Amazon to estimate how often the Alexa device does not wake
  • Wells Fargo to train reliable financial prediction models and at Microsoft, Tesla, Facebook, and other companies after the release of version 1.0 last year. 

Cleanlab 2.0 was entirely redesigned for all data scientists, ML datasets, and models. Quantum information theory inspired the theory that underpins cleanlab’s algorithms.

Some noticeable features of Cleanlab are as follows:

  1. Locate and resolve database issues
  2. Locate and resolve ontological dataset issues
  3. With a theoretically-proven health score, one can keep track of the total dataset quality.
  4. In actual circumstances with faulty models, including verifiable guarantees of precise noise estimates and label error discovery
  5. Optimized and parallel-threaded code
  6. Detect label errors in the data automatically.
  7. Each sample in the dataset is given a label quality score.
  8. In a single line of code, users can find label faults or train noise-resistant models. Cleanlab does not require any hyper-parameters by default.
  9. Any dataset and model may be used 
  10. At the dataset level, examine for overlapping classes to combine and/or eliminate.

Offering a framework to streamline data-centric AI, cleanlab assists data scientists and ML developers with the remaining 80%. Cleanlab helps machine learning and analytics processes with messy real-world data by identifying and resolving errors at the example, class, and dataset levels, evaluating and tracking overall dataset quality, and providing cleaned data for machine learning pipelines.

Source: https://cleanlab.ai/blog/cleanlab-2/

Github: https://github.com/cleanlab/cleanlab

Research: https://cleanlab.ai/research/

Tuutorials: https://docs.cleanlab.ai/v2.0.0/tutorials/