outlier-detector

Outlier Detector toolkit

This project features a set of tools for outlier detection, marking or filtering away samples as they come to your Python analysis code.

Most of the tools rely on double tailed Dixon's Q-test (https://en.wikipedia.org/wiki/Dixon%27s_Q_test).

Installation

pip install outlier-detector

TL;DR

I have a sample, and a know data distribution: is the sample an outlier?

sample = -14.5
distribution = [0.1, 1.1, 4.78, 2.0, 7.2, 5.3]

from outlier_detector.functions import is_outlier
print(is_outlier(distribution, sample))

I have a distribution and I iterate over it: is the n-th sample an outlier?

distribution = [0.1, 1.1, 4.78, 2.0, 7.2, 5.3, 8.1, -14.1, 5.4]
from outlier_detector.detectors import OutlierDetector
od = OutlierDetector(buffer_samples=5)
for sample in distribution:
    print(od.is_outlier(sample))

I have a generating object from which I pop samples; and I want only valid samples: how can I reject outliers?

distribution = [0.1, 1.1, 4.78, 2.0, 7.2, 5.3, 8.1, -14.1, 5.4]
from outlier_detector.filters import filter_outlier

class MyGen:
    def __init__(self):
        self.cursor = -1

    @filter_outlier()
    def pop(self):
        self.cursor += 1
        return distribution[self.cursor]

g = MyGen()
while True:
    try:
        r = g.pop()
        print(r)
    except IndexError:
        print('No more data')
        break

I have a generating object from which I pop samples; and I want to iterate only on valid samples: how can I reject outliers and get an iterator?

distribution = [0.1, 1.1, 4.78, 2.0, 7.2, 5.3, 8.1, -14.1, 5.4]
from outlier_detector.filters import OutlierFilter

class MyGen:
    def __init__(self):
        self.cursor = -1

    def pop(self):
        self.cursor += 1
        return distribution[self.cursor]

g = MyGen()
of = OutlierFilter()
try:
    for sample in of.filter(g.pop):
        print(sample)
except IndexError:
    print('No more data')

Documentation

The toolkit is organized so you can exploit one of the following pattern in the easiest way possible: functions for static analysis, detectors for objects with internal buffers, and filters for decorators.

For documentation see doc file