Skip to main content

Optimal Asymptotic Sequential Importance Sampling

Project description

OASIS is a tool for evaluating binary classifiers when ground truth class labels are not immediately available, but can be obtained at some cost (e.g. by asking humans). The tool takes an unlabelled test set as input and intelligently selects items to label so as to provide a precise estimate of the classifier’s performance, whilst minimising the amount of labelling required. The underlying strategy for selecting the items to label is based on a technique called adaptive importance sampling, which is optimised for the classifier performance measure of interest. Currently, OASIS supports estimation of the weighted F-measure, which includes the F1-score, precision and recall.


See the Jupyter notebook under docs/tutorial/tutorial.ipynb:

>>> import oasis
>>> data = oasis.Data()
>>> data.read_h5('Amazon-GoogleProducts-test.h5')
>>> def oracle(idx):
>>>     return data.labels[idx]
>>> smplr = oasis.OASISSampler(alpha, data.preds, data.scores, oracle)
>>> smplr.sample_distinct(5000) #: query labels for 5000 distinct items
>>> print("Current estimate is {}.".format(smplr.estimate_[smplr.t_ - 1]))

License and disclaimer

The code is released under the MIT license. Please see the LICENSE file for details.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

oasis-0.1.3.tar.gz (16.6 MB view hashes)

Uploaded Source

Built Distribution

oasis-0.1.3-py2.py3-none-any.whl (25.8 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page