rollingrank 0.3.0

fast rolling rank for numpy

rollingrank

rollingrank is a fast implementation of rolling rank transformation (described as the following code).

import pandas as pd

# x is numpy array
def rollingrank(x, window=None):
    def to_rank(x):
        # result[i] is the rank of x[i] in x
        return np.sum(np.less(x, x[-1]))
    return pd.Series(x).rolling(window).apply(to_rank).values

Motivation

Rolling rank is a good tool to create features for time series prediction. However, rolling rank was not easy to use in python. There were no exact methods to do it. The simple implementation using pandas and numpy is too slow.

Performance

Implementation	Complexity
rollingrank	O(n * log(w))
pandas rolling + numpy	O(n * w)

n: input length w: rolling window size

Install

pip install rollingrank

Example

import numpy as np
import rollingrank

x = np.array([0.1, 0.2, 0.3, 0.25, 0.1, 0.2, 0.3])
y = rollingrank.rollingrank(x, window=3)
print(y)
# [nan nan  2.  1.  0.  1.  2.]

y = rollingrank.rollingrank(x, window=3, pct=True)
print(y)
# [nan nan 1.  0.5 0.  0.5 1. ]

rci

RCI is also implemented because fast implementation is not found.

https://docs.anychart.com/Stock_Charts/Technical_Indicators/Mathematical_Description#rank_correlation_index

Kaggle Example

https://www.kaggle.com/bakuage/rollingrank-example

Development

test

python -m unittest discover tests

build/upload

python setup.py sdist
twine upload --repository pypitest dist/*
twine upload --repository pypi dist/*

TODO

• support axis