[英]How to calculate moving / running / rolling arbitrary function (e.g. kurtosis & skewness) using NumPy / SciPy
I am working on the time-series data. 我正在处理时间序列数据。 To get features from data I have to calculate moving mean, median, mode, slop, kurtosis, skewness etc. I am familiar with
scipy.stat
which provides an easy way to calculate these quantities for straight calculation. 为了从数据中获取特征,我必须计算移动平均值,中位数,
scipy.stat
, scipy.stat
,峰度,偏度等。我熟悉scipy.stat
,它提供了一种简单的方法来计算这些量以进行直接计算。 But for the moving/running part, I have explored the whole internet and got nothing. 但是对于运动/跑步部分,我已经浏览了整个互联网,但一无所获。
Surprisingly moving mean, median and mode are very easy to calculate with numpy
. 令人惊讶的是,移动平均数,中位数和
numpy
非常容易用numpy
计算。 Unfortunately, there is no built-in function for calculating kurtosis and skewness. 不幸的是,没有用于计算峰度和偏度的内置函数。 If someone can help, how to calculate moving kurtosis and skewness with scipy?
如果有人可以提供帮助,如何用scipy计算运动的峰度和偏度? Many thanks
非常感谢
Pandas offers a DataFrame.rolling()
method which can be used, in combination with its Rolling.apply()
method (ie df.rolling().apply()
) to apply an arbitrary function to the specified rolling window. 熊猫提供了一个
DataFrame.rolling()
方法,可以将其与Rolling.apply()
方法(即df.rolling().apply()
)结合使用,以将任意函数应用于指定的滚动窗口。
If you are looking for NumPy-based solution, you could use FlyingCircus (disclaimer: I am the main author of it). 如果您正在寻找基于NumPy的解决方案,则可以使用FlyingCircus (免责声明:我是它的主要作者)。
There, you could find the following: 在那里,您可以找到以下内容:
flyingcircus.extra.running_apply()
: can apply any function to a 1D array and supports weights, but it is slow; flyingcircus.extra.running_apply()
:可以将任何函数应用于一维数组并支持权重,但速度较慢; flyingcircus.extra.moving_apply()
: can apply any function supporting a axis: int
parameter to a 1D array and supports weights, and it is fast (but memory-hungry); flyingcircus.extra.moving_apply()
:可以将支持axis: int
参数的任何函数应用于一维数组并支持权重,并且速度很快(但需要大量内存); flyingcircus.extra.rolling_apply_nd()
: can apply any function supporting a axis: int|Sequence[int]
parameter to any ND array and it is fast (and memory-efficient), but it does not support weights. flyingcircus.extra.rolling_apply_nd()
:可以将支持axis: int|Sequence[int]
参数的任何函数应用于任何ND数组,并且速度快(且节省内存),但不支持权重。 Based on your requirements, I would suggest to use rolling_apply_nd()
, eg: 根据您的要求,我建议使用
rolling_apply_nd()
,例如:
import numpy as np
import scipy as sp
import flyingcircus as fc
import scipy.stats
NUM = 30
arr = np.arange(NUM)
window = 4
new_arr = fc.extra.rolling_apply_nd(arr, window, func=sp.stats.kurtosis)
print(new_arr)
# [-1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36
# -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36 -1.36
# -1.36 -1.36 -1.36]
Of course, feel free to inspect the source code , it is open source (GPL). 当然,可以随时检查源代码 ,它是开源(GPL)。
Just to get a feeling of the kind of speed we are talking about, these are the benchmarks for the solutions implemented in FlyingCircus: 只是为了了解我们正在谈论的那种速度,以下是FlyingCircus中实现的解决方案的基准:
The general approach flyingcircus.extra.running_apply()
is a couple of orders of magnitude slower than either flyingcircus.extra.rolling_apply_nd()
or flyingcircus.extra.moving_apply()
, with the first being approx. 一般方法
flyingcircus.extra.running_apply()
比flyingcircus.extra.rolling_apply_nd()
或flyingcircus.extra.moving_apply()
慢几个数量级,第一个大约是。 one order of magnitude faster than the second. 比第二个快一个数量级。 This shows the speed price for generality or support for weighting .
这显示了通用性或权重支持的速度价格。
The above plots were obtained using the scripts from here and the following code: 上面的图是使用此处的脚本和以下代码获得的:
import scipy as sp
import flyingcircus as fc
import scipy.stats
WINDOW = 4
FUNC = sp.stats.kurtosis
def my_rolling_apply_nd(arr, window=WINDOW, func=FUNC):
return fc.extra.rolling_apply_nd(arr, window, func=FUNC)
def my_moving_apply(arr, window=WINDOW, func=FUNC):
return fc.extra.moving_apply(arr, window, func)
def my_running_apply(arr, window=WINDOW, func=FUNC):
return fc.extra.running_apply(arr, window, func)
def equal_output(a, b):
return np.all(np.isclose(a, b))
input_sizes = (5, 10, 50, 100, 500, 1000, 5000, 10000, 50000, 100000)
funcs = my_rolling_apply_nd, my_moving_apply, my_running_apply
runtimes, input_sizes, labels, results = benchmark(
funcs, gen_input=np.random.random, equal_output=equal_output,
input_sizes=input_sizes)
plot_benchmarks(runtimes, input_sizes, labels, units='s')
plot_benchmarks(runtimes, input_sizes, labels, units='ms', zoom_fastest=8)
After playing around, I have come up with a solution that is purely numpy
and scipy
based. 在玩耍之后,我想出了一个纯粹基于
numpy
和scipy
的解决方案。 Of course it is using scipy.stats kurtosis
and skew
. 当然,它使用的是scipy.stats
kurtosis
和skew
。
import numpy as np
from scipy.stats import kurtosis, skew
# Window size
N = 4
# Some random data
m = np.array([2, 3, 10, 11, 0, 4, 8, 2, 5, 9])
# Running Kurtosis
def runningKurt(x, N):
# Initilize placeholder array
y = np.zeros((len(x) - (N - 1),))
for i in range(len(x) - (N - 1)):
y[i] = kurtosis(x[i:(i + N)])
return y
# Running Kurtosis
def runningSkew(x, N):
# Initilize placeholder array
y = np.zeros((len(x) - (N - 1),))
for i in range(len(x) - (N - 1)):
y[i] = skew(x[i:(i + N)])
return y
kurt = runningKurt(m, N)
print("kurtosis : ", kurt)
# kurtosis : [-1.93940828 -1.77879935 -1.61464214 -1.40236694 -1.15428571 -1.07626667 -1.42666667]
skw = runningSkew(m, N)
print("skew : ", skw)
# skew : [ 0. -0.1354179 -0.26356495 -0.13814702 0.43465076 0.32331615 -0.36514837]
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.