cougar

cougar

A python C-extension for rolling window aggregations. Try to support more methods than bottleneck and run faster than pandas. Currently this is only a weekend project, feel free to contribute.

Installation

pip install cougar

Usage

>>> import numpy as np
>>> import cougar as cg


>>> arr = np.random.randn(8)
>>> arr
array([-2.31572505, -1.78462521,  0.17355123,  0.77365821,  0.81431295,
        1.56188616,  0.74933881,  0.06184727])
>>> cg.rolling_mean(arr, 3)
array([        nan,         nan, -1.30893301, -0.27913859,  0.58717413,
        1.04995244,  1.04184597,  0.79102408])
>>> cg.rolling_rank(arr, 5, min_count=1)
array([ 0. ,  1. ,  1. ,  1. ,  1. ,  1. , -0.5, -1. ])

Supported Methods

Method	Time Complexity
sum	$O(1)$
mean	$O(1)$
std	$O(1)$
var	$O(1)$
skew	$O(1)$
kurt	$O(1)$
min	$O(1)$
max	$O(1)$
argmin	$O(1)$
argmax	$O(1)$
median	$O(\log n)$
rank	$O(\log n)$
quantile	$O(\log n)$

Benchmark

python benchmark.py

            Cougar vs. Bottleneck             
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┓
┃ Method       ┃ Cougar       ┃ Bottleneck   ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━┩
│ sum          │ 2.99         │ 1.00         │
│ mean         │ 3.08         │ 1.00         │
│ std          │ 2.10         │ 1.00         │
│ var          │ 2.15         │ 1.00         │
│ max          │ 2.84         │ 1.00         │
│ min          │ 1.23         │ 1.00         │
│ argmax       │ 1.40         │ 1.00         │
│ argmin       │ 1.16         │ 1.00         │
│ median       │ 1.33         │ 1.00         │
│ rank         │ 1.00         │ 15.02        │
└──────────────┴──────────────┴──────────────┘