seaborn：选择的 KDE 带宽为 0。无法估计密度

Question

import pandas as pd
import seaborn as sns

ser_test = pd.Series([1,0,1,4,6,0,6,5,1,3,2,5,1])
sns.kdeplot(ser_test, cumulative=True)

以上代码生成如下CDF图：

但是当系列的元素被修改为：

ser_test = pd.Series([1,0,1,1,6,0,6,1,1,0,2,1,1])
sns.kdeplot(ser_test, cumulative=True)

我收到以下错误：

ValueError: 无法将字符串转换为浮点数：'scott'

RuntimeError: Selected KDE bandwidth is 0. 无法估计密度。

此错误是什么意思，如何解决它以生成 CDF（即使它非常倾斜）。

编辑：我使用的是 seaborn 版本 0.9.0

完整的跟踪如下：

ValueError: could not convert string to float: 'scott'

During handling of the above exception, another exception occurred:

RuntimeError                              Traceback (most recent call last)
<ipython-input-93-7cee594b4526> in <module>
      1 ser_test = pd.Series([1,0,1,1,6,0,6,1,1,0,2,1,1])
----> 2 sns.kdeplot(ser_test, cumulative=True)

~/.local/lib/python3.5/site-packages/seaborn/distributions.py in kdeplot(data, data2, shade, vertical, kernel, bw, gridsize, cut, clip, legend, cumulative, shade_lowest, cbar, cbar_ax, cbar_kws, ax, **kwargs)
    689         ax = _univariate_kdeplot(data, shade, vertical, kernel, bw,
    690                                  gridsize, cut, clip, legend, ax,
--> 691                                  cumulative=cumulative, **kwargs)
    692 
    693     return ax

~/.local/lib/python3.5/site-packages/seaborn/distributions.py in _univariate_kdeplot(data, shade, vertical, kernel, bw, gridsize, cut, clip, legend, ax, cumulative, **kwargs)
    281         x, y = _statsmodels_univariate_kde(data, kernel, bw,
    282                                            gridsize, cut, clip,
--> 283                                            cumulative=cumulative)
    284     else:
    285         # Fall back to scipy if missing statsmodels

~/.local/lib/python3.5/site-packages/seaborn/distributions.py in _statsmodels_univariate_kde(data, kernel, bw, gridsize, cut, clip, cumulative)
    353     fft = kernel == "gau"
    354     kde = smnp.KDEUnivariate(data)
--> 355     kde.fit(kernel, bw, fft, gridsize=gridsize, cut=cut, clip=clip)
    356     if cumulative:
    357         grid, y = kde.support, kde.cdf

~/.local/lib/python3.5/site-packages/statsmodels/nonparametric/kde.py in fit(self, kernel, bw, fft, weights, gridsize, adjust, cut, clip)
    138             density, grid, bw = kdensityfft(endog, kernel=kernel, bw=bw,
    139                     adjust=adjust, weights=weights, gridsize=gridsize,
--> 140                     clip=clip, cut=cut)
    141         else:
    142             density, grid, bw = kdensity(endog, kernel=kernel, bw=bw,

~/.local/lib/python3.5/site-packages/statsmodels/nonparametric/kde.py in kdensityfft(X, kernel, bw, weights, gridsize, adjust, clip, cut, retgrid)
    451         bw = float(bw)
    452     except:
--> 453         bw = bandwidths.select_bandwidth(X, bw, kern) # will cross-val fit this pattern?
    454     bw *= adjust
    455 

~/.local/lib/python3.5/site-packages/statsmodels/nonparametric/bandwidths.py in select_bandwidth(x, bw, kernel)
    172         # eventually this can fall back on another selection criterion.
    173         err = "Selected KDE bandwidth is 0. Cannot estimate density."
--> 174         raise RuntimeError(err)
    175     else:
    176         return bandwidth

RuntimeError: Selected KDE bandwidth is 0. Cannot estimate density.

Answer 1

由于 statsmodels 出现问题。

无论如何，要解决从 0.10.0 开始的 seaborn 版本的问题，只需将 diag_kws={'bw': 1} 设置为 arg。

尝试找出带宽的最佳值。

Answer 2

您可以尝试三个选项

首先：使用默认设置显示 KDE 块

sns.distplot(ser_test, hist = False, rug = True, rug_kws = {'color' : 'r'})

第二个：KDE 用窄带宽显示单个概率块

sns.distplot(ser_test, hist = False, rug = True, rug_kws = {'color' : 'r'}, kde_kws = {'bw' : 1})

第三：选择不同的三角核函数（块状）

sns.distplot(ser_test, hist = False, rug = True, rug_kws = {'color' : 'r'}, kde_kws = {'bw' : 1.5, 'kernel' : 'tri'})

Answer 3

如果您不想等待 seaborn git 更新以稳定版本发布，您可以尝试the issue page 中的解决方案之一。特别是 henrymartin1 建议尝试在 try/catch 块内手动传递小带宽（由 ahartikainen 建议），该块会抓取此特定错误的文本（因此仍会引发其他错误）：

try:
    sns.distplot(df)
except RuntimeError as re:
    if str(re).startswith("Selected KDE bandwidth is 0. Cannot estimate density."):
        sns.distplot(df, kde_kws={'bw': 0.1})
    else:
        raise re

这对我有用。

Answer 4

pip uninstall statsmodels 用同样的错误解决了类似的问题。

Answer 5

这里发生的事情是 Seaborn（或者更确切地说，它依赖于计算 KDE 的库 - scipy 或 statsmodels）无法计算出“带宽”，scaling parameter used in the calculation。您可以手动传递它。我玩了几个值，发现 1.5 给出的图表与您之前的比例相同：

sns.kdeplot(ser_test, cumulative=True, bw=1.5)

另见here。如果没有，值得安装statsmodels。