如何规范化seaborn distplot？

Question

出于重现性原因，数据集和重现性原因，我在[此处][1]共享它。

这是我正在做的 - 从第 2 列开始，我正在读取当前行并将其与前一行的值进行比较。如果它更大，我会继续比较。如果当前值小于上一行的值，我想将当前值（较小）除以先前值（较大）。据此，代码如下：

这给出了以下图表。

sns.distplot(quotient, hist=False, label=protname)

从图中可以看出

quotient_times

Data-V 的商为 0.8，如果 quotient_times 为大于 3.

我想对这些值进行归一化，以便我们有 y-axis 个介于 0 和 1 之间的第二个绘图值。我们如何在 Python 中做到这一点？

Answer 1

前言

据我了解，默认情况下，seaborn distplot 会进行 kde 估计。如果您想要归一化的 distplot 图，可能是因为您假设图的 Y 应该在 [0;1] 之间。如果是这样，一个堆栈溢出问题引发了kde estimators showing values above 1.

的问题

引用 one answer:

a continous pdf (pdf=probability density function) never says the value to be less than 1, with the pdf for continous random variable, function p(x) is not the probability. you can refer for continuous random variables and their distrubutions

引用 importanceofbeingernest 的第一条评论：

The integral over a pdf is 1. There is no contradiction to be seen here.

据我所知，CDF (Cumulative Density Function) 的值应该在 [0; 1].

注意：所有可能的连续拟合函数为on SciPy site and available in the package scipy.stats

也许也看看 probability mass functions ?

如果你真的想对同一张图进行归一化，那么你应该收集绘制函数（选项 1）或函数定义（选项 2）的实际数据点，然后自己对它们进行归一化并再次绘制。

选项 1

import numpy as np
import matplotlib
import matplotlib.pyplot as plt
import seaborn as sns
import sys

print('System versions          : {}'.format(sys.version))
print('System versions          : {}'.format(sys.version_info))
print('Numpy versqion           : {}'.format(np.__version__))
print('matplotlib.pyplot version: {}'.format(matplotlib.__version__))
print('seaborn version          : {}'.format(sns.__version__))

protocols = {}

types = {"data_v": "data_v.csv"}

for protname, fname in types.items():
    col_time,col_window = np.loadtxt(fname,delimiter=',').T
    trailing_window = col_window[:-1] # "past" values at a given index
    leading_window  = col_window[1:]  # "current values at a given index
    decreasing_inds = np.where(leading_window < trailing_window)[0]
    quotient = leading_window[decreasing_inds]/trailing_window[decreasing_inds]
    quotient_times = col_time[decreasing_inds]

    protocols[protname] = {
        "col_time": col_time,
        "col_window": col_window,
        "quotient_times": quotient_times,
        "quotient": quotient,
    }

    fig, (ax1, ax2) = plt.subplots(1,2, sharey=False, sharex=False)
    g = sns.distplot(quotient, hist=True, label=protname, ax=ax1, rug=True)
    ax1.set_title('basic distplot (kde=True)')
    # get distplot line points
    line = g.get_lines()[0]
    xd = line.get_xdata()
    yd = line.get_ydata()
    # 
    def normalize(x):
        return (x - x.min(0)) / x.ptp(0)
    #normalize points
    yd2 = normalize(yd)
    # plot them in another graph
    ax2.plot(xd, yd2)
    ax2.set_title('basic distplot (kde=True)\nwith normalized y plot values')

    plt.show()

选项 2

下面，我尝试执行kde并对获得的估计进行归一化。我不是统计专家，所以 kde 的使用在某些方面可能是错误的（它与 seaborn 的不同，正如您在屏幕截图中看到的那样，这是因为 seaborn 比我做得更好。它只是试图模仿kde 适合 scipy。我猜结果还不错)

截图：

代码：

import numpy as np
from scipy import stats
import matplotlib
import matplotlib.pyplot as plt
import seaborn as sns
import sys

print('System versions          : {}'.format(sys.version))
print('System versions          : {}'.format(sys.version_info))
print('Numpy versqion           : {}'.format(np.__version__))
print('matplotlib.pyplot version: {}'.format(matplotlib.__version__))
print('seaborn version          : {}'.format(sns.__version__))

protocols = {}

types = {"data_v": "data_v.csv"}

for protname, fname in types.items():
    col_time,col_window = np.loadtxt(fname,delimiter=',').T
    trailing_window = col_window[:-1] # "past" values at a given index
    leading_window  = col_window[1:]  # "current values at a given index
    decreasing_inds = np.where(leading_window < trailing_window)[0]
    quotient = leading_window[decreasing_inds]/trailing_window[decreasing_inds]
    quotient_times = col_time[decreasing_inds]

    protocols[protname] = {
        "col_time": col_time,
        "col_window": col_window,
        "quotient_times": quotient_times,
        "quotient": quotient,
    }

    fig, (ax1, ax2, ax3, ax4) = plt.subplots(1,4, sharey=False, sharex=False)
    diff=quotient_times
    ax1.plot(diff, quotient, ".", label=protname, color="blue")
    ax1.set_ylim(0, 1.0001)
    ax1.set_title(protname)
    ax1.set_xlabel("quotient_times")
    ax1.set_ylabel("quotient")
    ax1.legend()

    sns.distplot(quotient, hist=True, label=protname, ax=ax2, rug=True)
    ax2.set_title('basic distplot (kde=True)')

    # taken from seaborn's source code (utils.py and distributions.py)
    def seaborn_kde_support(data, bw, gridsize, cut, clip):
        if clip is None:
            clip = (-np.inf, np.inf)
        support_min = max(data.min() - bw * cut, clip[0])
        support_max = min(data.max() + bw * cut, clip[1])
        return np.linspace(support_min, support_max, gridsize)

    kde_estim = stats.gaussian_kde(quotient, bw_method='scott')

    # manual linearization of data
    #linearized = np.linspace(quotient.min(), quotient.max(), num=500)

    # or better: mimic seaborn's internal stuff
    bw = kde_estim.scotts_factor() * np.std(quotient)
    linearized = seaborn_kde_support(quotient, bw, 100, 3, None)

    # computes values of the estimated function on the estimated linearized inputs
    Z = kde_estim.evaluate(linearized)

    # 
    def normalize(x):
        return (x - x.min(0)) / x.ptp(0)

    # normalize so it is between 0;1
    Z2 = normalize(Z)
    for name, func in {'min': np.min, 'max': np.max}.items():
        print('{}: source={}, normalized={}'.format(name, func(Z), func(Z2)))

    # plot is different from seaborns because not exact same method applied
    ax3.plot(linearized, Z, ".", label=protname, color="orange")
    ax3.set_title('Non linearized gaussian kde values')

    # manual kde result with Y axis avalues normalized (between 0;1)
    ax4.plot(linearized, Z2, ".", label=protname, color="green")
    ax4.set_title('Normalized gaussian kde values')

    plt.show()

输出：

System versions          : 3.7.2 (default, Feb 21 2019, 17:35:59) [MSC v.1915 64 bit (AMD64)]
System versions          : sys.version_info(major=3, minor=7, micro=2, releaselevel='final', serial=0)
Numpy versqion           : 1.16.2
matplotlib.pyplot version: 3.0.2
seaborn version          : 0.9.0
min: source=0.0021601491646143518, normalized=0.0
max: source=9.67319154426489, normalized=1.0

与评论相反，密谋：

[(x-min(quotient))/(max(quotient)-min(quotient)) for x in quotient]

不改变行为！它仅更改核密度估计的源数据。曲线形状将保持不变。

Quoting seaborn's distplot doc:

This function combines the matplotlib hist function (with automatic calculation of a good default bin size) with the seaborn kdeplot() and rugplot() functions. It can also fit scipy.stats distributions and plot the estimated PDF over the data.

默认：

kde : bool, optional set to True Whether to plot a gaussian kernel density estimate.

默认使用kde。引用 seaborn 的 kde 文档：

Fit and plot a univariate or bivariate kernel density estimate.

引用 SCiPy gaussian kde method doc:

Representation of a kernel-density estimate using Gaussian kernels.

Kernel density estimation is a way to estimate the probability density function (PDF) of a random variable in a non-parametric way. gaussian_kde works for both uni-variate and multi-variate data. It includes automatic bandwidth determination. The estimation works best for a unimodal distribution; bimodal or multi-modal distributions tend to be oversmoothed.

请注意，正如您自己提到的，我相信您的数据是双峰的。它们看起来也很离散。据我所知，离散分布函数可能无法像连续分布函数那样分析，拟合可能会很棘手。

这里有一个包含各种定律的例子：

import numpy as np
from scipy.stats import uniform, powerlaw, logistic
import matplotlib
import matplotlib.pyplot as plt
import seaborn as sns
import sys

print('System versions          : {}'.format(sys.version))
print('System versions          : {}'.format(sys.version_info))
print('Numpy versqion           : {}'.format(np.__version__))
print('matplotlib.pyplot version: {}'.format(matplotlib.__version__))
print('seaborn version          : {}'.format(sns.__version__))

protocols = {}

types = {"data_v": "data_v.csv"}

for protname, fname in types.items():
    col_time,col_window = np.loadtxt(fname,delimiter=',').T
    trailing_window = col_window[:-1] # "past" values at a given index
    leading_window  = col_window[1:]  # "current values at a given index
    decreasing_inds = np.where(leading_window < trailing_window)[0]
    quotient = leading_window[decreasing_inds]/trailing_window[decreasing_inds]
    quotient_times = col_time[decreasing_inds]

    protocols[protname] = {
        "col_time": col_time,
        "col_window": col_window,
        "quotient_times": quotient_times,
        "quotient": quotient,
    }
    fig, [(ax1, ax2, ax3), (ax4, ax5, ax6)] = plt.subplots(2,3, sharey=False, sharex=False)
    diff=quotient_times
    ax1.plot(diff, quotient, ".", label=protname, color="blue")
    ax1.set_ylim(0, 1.0001)
    ax1.set_title(protname)
    ax1.set_xlabel("quotient_times")
    ax1.set_ylabel("quotient")
    ax1.legend()
    quotient2 = [(x-min(quotient))/(max(quotient)-min(quotient)) for x in quotient]
    print(quotient2)
    sns.distplot(quotient, hist=True, label=protname, ax=ax2, rug=True)
    ax2.set_title('basic distplot (kde=True)')
    sns.distplot(quotient2, hist=True, label=protname, ax=ax3, rug=True)
    ax3.set_title('logistic distplot')

    sns.distplot(quotient, hist=True, label=protname, ax=ax4, rug=True, kde=False, fit=uniform)
    ax4.set_title('uniform distplot')
    sns.distplot(quotient, hist=True, label=protname, ax=ax5, rug=True, kde=False, fit=powerlaw)
    ax5.set_title('powerlaw distplot')
    sns.distplot(quotient, hist=True, label=protname, ax=ax6, rug=True, kde=False, fit=logistic)
    ax6.set_title('logistic distplot')
    plt.show()

输出：

System versions          : 3.7.2 (default, Feb 21 2019, 17:35:59) [MSC v.1915 64 bit (AMD64)]
System versions          : sys.version_info(major=3, minor=7, micro=2, releaselevel='final', serial=0)
Numpy versqion           : 1.16.2
matplotlib.pyplot version: 3.0.2
seaborn version          : 0.9.0
[1.0, 0.05230125523012544, 0.0433775382360589, 0.024590765616971128, 0.05230125523012544, 0.05230125523012544, 0.05230125523012544, 0.02836946874603772, 0.05230125523012544, 0.05230125523012544, 0.05230125523012544, 0.05230125523012544, 0.03393500048652319, 0.05230125523012544, 0.05230125523012544, 0.05230125523012544, 0.0037013196009011043, 0.0, 0.05230125523012544]

截图：

Answer 2

在最近的更新中，必须使用 sns.distplot has been deprecated and sns.histplot 来代替。因此，要获得规范化的 histogram/density 必须使用以下语法：

sns.histplot(x, kind='hist', stat='density');

或

sns.plot(x, stat='density');

而不是

sns.distplot(x, kde=False, norm_hist=True);

PS：要获取密度而不是直方图，必须将 kind 值更改为 'kde'。

如何规范化seaborn distplot？

How to normalize seaborn distplot?

python

statistics

distribution

python-3.x

seaborn

前言

选项 1

选项 2