使用 seaborn 可视化缺失数据的出现频率

Question

我想创建一个 24x20 矩阵（8 个部分，每个部分有 60 个单元格或 6x10），以便通过循环（=每个 480 个值) 通过 panda 数据框在数据集中并为每一列绘制它'A','B','C'。

到目前为止，我可以映射创建的 csv 文件并在矩阵中以正确的方式映射值，并在更改缺失数据 (nan & inf) 到0 或类似0.01234 的东西对数据的影响最小，另一方面可以绘制。 以下是我目前的脚本：

import numpy as np
import pandas as pd
import os
import seaborn as sns
import matplotlib.pyplot as plt

def mkdf(ListOf480Numbers):
    normalMatrix = np.array_split(ListOf480Numbers,8)
    fixMatrix = []
    for i in range(8):
        lines = np.array_split(normalMatrix[i],6)
        newMatrix = [0,0,0,0,0,0]
        for j in (1,3,5):
            newMatrix[j] = lines[j]
        for j in (0,2,4):
            newMatrix[j] = lines[j][::-1]
        fixMatrix.append(newMatrix) 
    return fixMatrix

def print_df(fixMatrix):
    values = []
    for i in range(6):
        values.append([*fixMatrix[6][i], *fixMatrix[7][i]])
    for i in range(6):
        values.append([*fixMatrix[4][i], *fixMatrix[5][i]])
    for i in range(6):
        values.append([*fixMatrix[2][i], *fixMatrix[3][i]])
    for i in range(6):
        values.append([*fixMatrix[0][i], *fixMatrix[1][i]])
    df = pd.DataFrame(values)
    return (df)




dft = pd.read_csv('D:\Feryan.TXT', header=None)
id_set = dft[dft.index % 4 == 0].astype('int').values
A = dft[dft.index % 4 == 1].values
B = dft[dft.index % 4 == 2].values
C = dft[dft.index % 4 == 3].values
data = {'A': A[:,0], 'B': B[:,0], 'C': C[:,0]}

df = pd.DataFrame(data, columns=['A','B','C'], index = id_set[:,0])  

nan = np.array(df.isnull())
inf = np.array(df.isnull())
df = df.replace([np.inf, -np.inf], np.nan)
df[np.isinf(df)] = np.nan    # convert inf to nan
#dff = df[df.isnull().any(axis=1)]   # extract sub data frame

#df = df.fillna(0)
#df = df.replace(0,np.nan)



#next iteration create all plots, change the number of cycles
cycles = int(len(df)/480)
print(cycles)
for cycle in range(3):
    count =  '{:04}'.format(cycle)
    j = cycle * 480
    new_value1 = df['A'].iloc[j:j+480]
    new_value2 = df['B'].iloc[j:j+480]
    new_value3 = df['C'].iloc[j:j+480]
    df1 = print_df(mkdf(new_value1))
    df2 = print_df(mkdf(new_value2))
    df3 = print_df(mkdf(new_value3))              
    for i in df:
        try:
            os.mkdir(i)
        except:
            pass
            df1.to_csv(f'{i}/norm{i}{count}.csv', header=None, index=None) 
            df2.to_csv(f'{i}/norm{i}{count}.csv', header=None, index=None)
            df3.to_csv(f'{i}/norm{i}{count}.csv', header=None, index=None)

    #plotting all columns ['A','B','C'] in-one-window side by side


    fig, ax = plt.subplots(nrows=1, ncols=3 , figsize=(20,10))
    plt.subplot(131)

    ax = sns.heatmap(df1.isnull(), cbar=False)
    ax.axhline(y=6, color='w',linewidth=1.5)
    ax.axhline(y=12, color='w',linewidth=1.5)
    ax.axhline(y=18, color='w',linewidth=1.5)
    ax.axvline(x=10, color='w',linewidth=1.5)

    plt.title('Missing-data frequency in A', fontsize=20 , fontweight='bold', color='black', loc='center', style='italic')
    plt.axis('off')

    plt.subplot(132)
    ax = sns.heatmap(df2.isnull(), cbar=False)
    ax.axhline(y=6, color='w',linewidth=1.5)
    ax.axhline(y=12, color='w',linewidth=1.5)
    ax.axhline(y=18, color='w',linewidth=1.5)
    ax.axvline(x=10, color='w',linewidth=1.5)
    plt.title('Missing-data frequency in B', fontsize=20 , fontweight='bold', color='black', loc='center', style='italic')
    plt.axis('off')

    plt.subplot(133)
    ax = sns.heatmap(df3.isnull(), cbar=False)
    ax.axhline(y=6, color='w',linewidth=1.5)
    ax.axhline(y=12, color='w',linewidth=1.5)
    ax.axhline(y=18, color='w',linewidth=1.5)
    ax.axvline(x=10, color='w',linewidth=1.5) 
    plt.title('Missing-data frequency in C', fontsize=20 , fontweight='bold', color='black', loc='center', style='italic')
    plt.axis('off')

    plt.suptitle(f'Missing-data visualization', color='yellow', backgroundcolor='black', fontsize=15, fontweight='bold')
    plt.subplots_adjust(top=0.92, bottom=0.02, left=0.05, right=0.96, hspace=0.2, wspace=0.2)
    fig.text(0.035, 0.93, 'dataset1' , fontsize=19, fontweight='bold', rotation=42., ha='center', va='center',bbox=dict(boxstyle="round",ec=(1., 0.5, 0.5),fc=(1., 0.8, 0.8)))
    #fig.tight_layout()
    plt.savefig(f'{i}/result{count}.png') 
    #plt.show()      

问题是我不知道如何正确绘制缺失数据出现的频率，以了解它经常发生在哪些部分和单元格中。

Note1 缺失值越多颜色越亮，100% 缺失的数据通过循环呈现为白色颜色和纯黑色颜色表示非缺失值。可能会有一个从黑色 0% 到 100% 白色的条形图。

Note2 我还提供了 3 个周期的数据集示例文本文件，其中包含少量缺失数据，但可以手动修改和增加：dataset

预期结果应如下所示：

Answer 1

您可以将您的 nan/inf 数据存储在一个单独的数组中，您可以将每个 nan/inf 的周期累加起来。

您的数组似乎总是具有相同的大小，因此我将它们定义为固定大小。您可以更改它以匹配您的数据：

df1MissingDataFrequency = np.zeros((24,20))

然后您可以将它们相加，得到 nan 值（您已经在代码中将 inf 替换为 nan）：

df1MissingDataFrequency = df1MissingDataFrequency + np.isnan(df1).astype(int)

你所有的周期。

您的缩进似乎有些问题。我不知道这是否仅适用于您在此处发布的代码，或者在您的实际代码中是否相同，但目前您每个周期都制作一个新图并且您重新定义@ 987654331@ 对应每个i。

由于缺少频率数据，您的代码应如下所示：

import numpy as np
import pandas as pd
import os
import seaborn as sns
import matplotlib.pyplot as plt

def mkdf(ListOf480Numbers):
    normalMatrix = np.array_split(ListOf480Numbers,8)
    fixMatrix = []
    for i in range(8):
        lines = np.array_split(normalMatrix[i],6)
        newMatrix = [0,0,0,0,0,0]
        for j in (1,3,5):
            newMatrix[j] = lines[j]
        for j in (0,2,4):
            newMatrix[j] = lines[j][::-1]
        fixMatrix.append(newMatrix) 
    return fixMatrix

def print_df(fixMatrix):
    values = []
    for i in range(6):
        values.append([*fixMatrix[6][i], *fixMatrix[7][i]])
    for i in range(6):
        values.append([*fixMatrix[4][i], *fixMatrix[5][i]])
    for i in range(6):
        values.append([*fixMatrix[2][i], *fixMatrix[3][i]])
    for i in range(6):
        values.append([*fixMatrix[0][i], *fixMatrix[1][i]])
    df = pd.DataFrame(values)
    return (df)


dft = pd.read_csv('D:/Feryan2.txt', header=None)
id_set = dft[dft.index % 4 == 0].astype('int').values
A = dft[dft.index % 4 == 1].values
B = dft[dft.index % 4 == 2].values
C = dft[dft.index % 4 == 3].values
data = {'A': A[:,0], 'B': B[:,0], 'C': C[:,0]}

df = pd.DataFrame(data, columns=['A','B','C'], index = id_set[:,0])  

nan = np.array(df.isnull())
inf = np.array(df.isnull())
df = df.replace([np.inf, -np.inf], np.nan)
df[np.isinf(df)] = np.nan    # convert inf to nan


df1MissingDataFrequency = np.zeros((24,20))
df2MissingDataFrequency = np.zeros((24,20))
df3MissingDataFrequency = np.zeros((24,20))


#next iteration create all plots, change the number of cycles
cycles = int(len(df)/480)
print(cycles)
for cycle in range(3):
    count =  '{:04}'.format(cycle)
    j = cycle * 480
    new_value1 = df['A'].iloc[j:j+480]
    new_value2 = df['B'].iloc[j:j+480]
    new_value3 = df['C'].iloc[j:j+480]
    df1 = print_df(mkdf(new_value1))
    df2 = print_df(mkdf(new_value2))
    df3 = print_df(mkdf(new_value3))              
    for i in df:
        try:
            os.mkdir(i)
        except:
            pass
    df1.to_csv(f'{i}/norm{i}{count}.csv', header=None, index=None) 
    df2.to_csv(f'{i}/norm{i}{count}.csv', header=None, index=None)
    df3.to_csv(f'{i}/norm{i}{count}.csv', header=None, index=None)

    df1MissingDataFrequency = df1MissingDataFrequency + np.isnan(df1).astype(int)
    df2MissingDataFrequency = df2MissingDataFrequency + np.isnan(df2).astype(int)
    df3MissingDataFrequency = df3MissingDataFrequency + np.isnan(df3).astype(int)

#plotting all columns ['A','B','C'] in-one-window side by side
fig, ax = plt.subplots(nrows=1, ncols=3 , figsize=(10,7))
plt.subplot(131)

ax = sns.heatmap(df1MissingDataFrequency, cbar=False, cmap="gray")
ax.axhline(y=6, color='w',linewidth=1.5)
ax.axhline(y=12, color='w',linewidth=1.5)
ax.axhline(y=18, color='w',linewidth=1.5)
ax.axvline(x=10, color='w',linewidth=1.5)

plt.title('Missing-data frequency in A', fontsize=20 , fontweight='bold', color='black', loc='center', style='italic')
plt.axis('off')

plt.subplot(132)
ax = sns.heatmap(df2MissingDataFrequency, cbar=False, cmap="gray")
ax.axhline(y=6, color='w',linewidth=1.5)
ax.axhline(y=12, color='w',linewidth=1.5)
ax.axhline(y=18, color='w',linewidth=1.5)
ax.axvline(x=10, color='w',linewidth=1.5)
plt.title('Missing-data frequency in B', fontsize=20 , fontweight='bold', color='black', loc='center', style='italic')
plt.axis('off')

plt.subplot(133)
ax = sns.heatmap(df3MissingDataFrequency, cbar=False, cmap="gray")
ax.axhline(y=6, color='w',linewidth=1.5)
ax.axhline(y=12, color='w',linewidth=1.5)
ax.axhline(y=18, color='w',linewidth=1.5)
ax.axvline(x=10, color='w',linewidth=1.5) 
plt.title('Missing-data frequency in C', fontsize=20 , fontweight='bold', color='black', loc='center', style='italic')
plt.axis('off')

plt.suptitle(f'Missing-data visualization', color='yellow', backgroundcolor='black', fontsize=15, fontweight='bold')
plt.subplots_adjust(top=0.92, bottom=0.02, left=0.05, right=0.96, hspace=0.2, wspace=0.2)
fig.text(0.035, 0.93, 'dataset1' , fontsize=19, fontweight='bold', rotation=42., ha='center', va='center',bbox=dict(boxstyle="round",ec=(1., 0.5, 0.5),fc=(1., 0.8, 0.8)))
#fig.tight_layout()
plt.savefig(f'{i}/result{count}.png') 
#plt.show()      

这会给你想要的输出：

编辑

本着DRY 的精神，我编辑了您的代码，这样您就没有 df1、df2、df3、new_values1 ……而且您复制和粘贴的都是相同的东西。您已经循环了i，因此您应该使用它来实际处理数据框中的三个不同列：

dft = pd.read_csv('C:/Users/frefra/Downloads/Feryan2.txt', header=None).replace([np.inf, -np.inf], np.nan)
id_set = dft[dft.index % 4 == 0].astype('int').values
A = dft[dft.index % 4 == 1].values
B = dft[dft.index % 4 == 2].values
C = dft[dft.index % 4 == 3].values
data = {'A': A[:,0], 'B': B[:,0], 'C': C[:,0]}
df = pd.DataFrame(data, columns=['A','B','C'], index = id_set[:,0])


new_values = []
dfs = []
nan_frequencies = np.zeros((3,24,20))

#next iteration create all plots, change the number of cycles
cycles = int(len(df)/480)
print(cycles)
for cycle in range(cycles):
    count =  '{:04}'.format(cycle)
    j = cycle * 480
    for idx,i in enumerate(df):
        try:
            os.mkdir(i)
        except:
            pass
        new_value = df[i].iloc[j:j+480]        
        new_values.append(new_value)
        dfi = print_df(mkdf(new_value))
        dfs.append(dfi)
        dfi.to_csv(f'{i}/norm{i}{count}.csv', header=None, index=None) 
        nan_frequencies[idx] = nan_frequencies[idx] + np.isnan(dfi).astype(int)


#plotting all columns ['A','B','C'] in-one-window side by side
fig, ax = plt.subplots(nrows=1, ncols=3 , figsize=(10,7))

for idx,i in enumerate(df):

    plt.subplot(1,3,idx+1)

    ax = sns.heatmap(nan_frequencies[idx], cbar=False, cmap="gray")
    ax.axhline(y=6, color='w',linewidth=1.5)
    ax.axhline(y=12, color='w',linewidth=1.5)
    ax.axhline(y=18, color='w',linewidth=1.5)
    ax.axvline(x=10, color='w',linewidth=1.5)

    plt.title('Missing-data frequency in ' + i, fontsize=20 , fontweight='bold', color='black', loc='center', style='italic')
    plt.axis('off')