Skip to content

Latest commit

 

History

History
66 lines (37 loc) · 2.92 KB

README.md

File metadata and controls

66 lines (37 loc) · 2.92 KB
Raw

Confidence Tool

Introduction

Welcome to the confidence tool page.

The confidence tool measures how confident a given classifier is about each prediction it makes. The tool can be implemented with a single line of code, and may be applied in a number of ways that will increase the capabilities of your original machine learning model.
If you just like to see how it can be applied in practice, keep reading and check out our Basic Tour in a jupyter notebook.

To actually understand how it works, check the reference for more details.

Applications

One could use this tool to:

  • Quantifying the confidence of each prediction.
  • Increasing model robustness, by filtering marginal classification events due to noise in data.
  • Operating the classifier (virtually) up to any accuracy level.
  • Binning examples according to the confidence of their prediction.

Check how to achieve each of these applications here.

The tool is very versatile and may also be used as an ensemble framework, or even as a semi supervised learning strategy for automatic label assigment (self learning).

Limitations

This tool is only compatible to classifiers that are able to output class probabilities for each prediction. Currently, the confidence tool relies on scikit-learn models that have a predict_proba() attribute. The majority is listed below:

	sklearn.svm.SVC(probability=True)
	sklearn.naive_bayes.GaussianNB()
	sklearn.ensemble.ExtraTreesClassifier()
	sklearn.ensemble.GradientBoostingClassifier()
	sklearn.linear_model.SGDClassifier(loss='modified_huber')
	sklearn.neighbors.KNeighborsClassifier()
	sklearn.gaussian_process.GaussianProcessClassifier()

If your classifier is not from scikit-learn, ensure that it is able to output arrays of class probabilities for each example. In this case you will also have to change the load_models script in the source code, to accommodate your model.

Installation

  1. Ensure that you are operating with Python 3 or above.

  2. Install the dependencies:

    Via pip:

     $ pip install pandas numpy matplotlib

    Via conda:

     $ conda install pandas numpy matplotlib

  3. Install the confidence tool:

     $ pip install -i https://test.pypi.org/simple/ confidence-tool

References

Please cite us:

A. P. Frade, P. McCabe and R. I. Cooper. “Increasing the performance, trustworthiness and practical value of machine learning models: a case study predicting hydrogen bond network dimensionalities from molecular diagrams”. 2020. CrystEngComm. DOI: 10.1039/D0CE00111B

Access the paper here.