“不能将 float Nan 转换为 int”但没有 Nan？

Question

我有一个数据框并尝试进行以下操作：

data['SD_rates']=np.array([int((data['actual value'][i]-data['means'][i])/data['std'][i]) for i in range (len(data['means']))])  

它与以下消息中断： “无法将 float Nan 转换为 int”

我理解这是一个错误，但使用 data.isnull() 测试了 df，并且没有涉及的列包含 NaN（我通过发送 data.to_csv 手动控制它）。

我什至用 fillna(-1, inplace=True) 填充了 data['std'] ，但它仍然中断了。我不明白为什么，因为没有除以 0（我还控制了该列中没有零，所以没有原始 0 和 Null/Nan 填充 -1），实际值和均值是 fillna(0 ) 用于缺失值，并且无论如何减法不能产生 nan（数据范围在 [0-10] 中）。

可能出了什么问题？ （正如我所说，触发操作之前的数据是正确的......）。谢谢

---

这是一个代码 sn-p:
我的一个假设是，在某种程度上，groupby 可能会生成 NaN，在计算我的平均值时我无法摆脱（但我相信它被 pandas 自动忽略了......）并且没有填充 0 或-1（我故意选择-1作为标准差以避免除以0）。

def stats_setting(data):

    print('Stats settings')
    print(data.columns)
    print(data.dtypes) 
    #sys.exit()

    data['marks']=np.log1p(data['marks'].astype(float))
    data['students']=np.log1p(data['students'].astype(float))#Rossman9 think this has to be tested
    #were filled with fillna before)

#First Part: by studentType and Assortment
    types_DoM_select=['Type','Type2','Category']

#First Block:types_DoM students grouped by categories
#wonder if can do a groupby of groupb
    print("types_DoM_marks_means")
    types_DoM_marks_means = data.groupby(types_DoM_select)['marks'].mean()
    types_DoM_marks_means.name = 'types_DoM_marks_means'
    types_DoM_marks_means = types_DoM_marks_means.reset_index()
    data = pd.merge(data, types_DoM_marks_means, on = types_DoM_select, how='left')

    print("types_DoM_students_means")
    types_DoM_students_means = data.groupby(types_DoM_select)['students'].mean() #.students won't work. Why?
    types_DoM_students_means.name = 'types_DoM_students_means'
    types_DoM_students_means=types_DoM_students_means.reset_index()    
    data = pd.merge(data, types_DoM_students_means, on = types_DoM_select, how='left')

    print("types_DoM_marks_medians")
    types_DoM_marks_medians = data.groupby(types_DoM_select)['marks'].median()
    types_DoM_marks_medians.name = 'types_DoM_marks_medians'
    types_DoM_marks_medians = types_DoM_marks_medians.reset_index()
    data = pd.merge(data, types_DoM_marks_medians, on = types_DoM_select, how='left')

    print("types_DoM_students_medians")
    types_DoM_students_medians = data.groupby(types_DoM_select)['students'].median() #.students won't work. Why?
    types_DoM_students_medians.name = 'types_DoM_students_medians'
    types_DoM_students_medians=types_DoM_students_medians.reset_index()    
    data = pd.merge(data, types_DoM_students_medians, on = types_DoM_select, how='left')
    print("types_DoM_marks_std")
    types_DoM_marks_std = data.groupby(types_DoM_select)['marks'].std()
    types_DoM_marks_std.name = 'types_DoM_marks_std'
    types_DoM_marks_std = types_DoM_marks_std.reset_index()
    data = pd.merge(data, types_DoM_marks_std, on = types_DoM_select, how='left')


    print("types_DoM_students_std")
    types_DoM_students_std = data.groupby(types_DoM_select)['students'].std()
    types_DoM_students_std.name = 'types_DoM_students_std'
    types_DoM_students_std = types_DoM_students_std.reset_index()
    data = pd.merge(data, types_DoM_students_std, on = types_DoM_select, how='left')

    data['types_DoM_marks_means'].fillna(-1, inplace=True)
    data['types_DoM_students_means'].fillna(-1, inplace=True)
    data['types_DoM_marks_medians'].fillna(-1, inplace=True)
    data['types_DoM_students_medians'].fillna(-1, inplace=True)
    data['types_DoM_marks_std'].fillna(-1, inplace=True)
    data['types_DoM_students_std'].fillna(-1, inplace=True)

#Second Part: by specific student
    student_DoM_select=['Type','Type2','Category']

#First Block:student_DoM
#wonder if can do a groupby of groupb
    print("student_DoM_marks_means")
    student_DoM_marks_means = data.groupby(student_DoM_select)['marks'].mean()
    student_DoM_marks_means.name = 'student_DoM_marks_means'
    student_DoM_marks_means = student_DoM_marks_means.reset_index()
    data = pd.merge(data, student_DoM_marks_means, on = student_DoM_select, how='left')

    print("student_DoM_students_means")
    student_DoM_students_means = data.groupby(student_DoM_select)['students'].mean() #.students won't work. Why?
    student_DoM_students_means.name = 'student_DoM_students_means'
    student_DoM_students_means=student_DoM_students_means.reset_index()    
    data = pd.merge(data, student_DoM_students_means, on = student_DoM_select, how='left')

    print("student_DoM_marks_medians")
    student_DoM_marks_medians = data.groupby(student_DoM_select)['marks'].median()
    student_DoM_marks_medians.name = 'student_DoM_marks_medians'
    student_DoM_marks_medians = student_DoM_marks_medians.reset_index()
    data = pd.merge(data, student_DoM_marks_medians, on = student_DoM_select, how='left')

    print("student_DoM_students_medians")
    student_DoM_students_medians = data.groupby(student_DoM_select)['students'].median() #.students won't work. Why?
    student_DoM_students_medians.name = 'student_DoM_students_medians'
    student_DoM_students_medians=student_DoM_students_medians.reset_index()    
    data = pd.merge(data, student_DoM_students_medians, on = student_DoM_select, how='left')

    # May I use data['marks','students','marksMean','studentsMean','marksMedian','studentsMedian']=data['marks','students','marksMean','studentsMean','marksMedian','studentsMedian'].astype(int) to spare memory?

    print("student_DoM_marks_std")
    student_DoM_marks_std = data.groupby(student_DoM_select)['marks'].std()
    student_DoM_marks_std.name = 'student_DoM_marks_std'
    student_DoM_marks_std = student_DoM_marks_std.reset_index()
    data = pd.merge(data, student_DoM_marks_std, on = student_DoM_select, how='left')

    print("student_DoM_students_std")
    student_DoM_students_std = data.groupby(student_DoM_select)['students'].std()
    student_DoM_students_std.name = 'student_DoM_students_std'
    student_DoM_students_std = student_DoM_students_std.reset_index()
    data = pd.merge(data, student_DoM_students_std, on = student_DoM_select, how='left')

    data['student_DoM_marks_means'].fillna(0, inplace=True)
    data['student_DoM_students_means'].fillna(0, inplace=True)
    data['student_DoM_marks_medians'].fillna(0, inplace=True)
    data['student_DoM_students_medians'].fillna(0, inplace=True)
    data['student_DoM_marks_std'].fillna(0, inplace=True)
    data['student_DoM_students_std'].fillna(0, inplace=True)

#Third Part: Exceptional students 

    #I think int is better here as it helps defining categories but can't use it.#    
    #print(data.isnull().sum())
    #print(data['types_DoM_marks_std'][data['types_DoM_marks_std']==0].sum())
    #data.to_csv('ex')
    #print(data.columns)

#Original version:#int raises the "can't convert Nan float to int. While there were no Nan as I verified in the data just before sending it to the    
    data['Except_student_IP2_DoM_marks_means']=np.array([int((data['student_IP2_DoM_marks_means'][i]-data['types_IP2_DoM_marks_means'][i])/data['types_IP2_DoM_students_std'][i]) for i in range (len(data['year']))])  
    data['Except_student_IP2_DoM_marks_medians']=np.array([int((data['student_IP2_DoM_marks_medians'][i]-data['types_IP2_DoM_marks_means'][i])/data['types_IP2_DoM_students_std'][i]) for i in range (len(data['year']))])  
#Second version: raises no error but final data (returned) is filled with these stupid NaN
    data['Except_student_P2M_DoM_marks_means']=np.array([np.round((data['student_DoM_marks_means'][i]-data['types_DoM_marks_means'][i])/data['types_DoM_marks_std'][i],0) for i in range (len(data['year']))])  
    data['Except_student_P2M_DoM_marks_medians']=np.array([np.round((data['student_DoM_marks_medians'][i]-data['types_DoM_marks_medians'][i])/data['types_DoM_marks_std'][i],0) for i in range (len(data['year']))])  

#End
    return data

Answer 1

您的数据框中没有 Nan 很可能是正确的，但您是在计算中创建它们。请参阅以下内容：

In [15]: import pandas as pd
In [16]: df = pd.DataFrame([[1, 2], [0, 0]], columns=['actual value', 'col2'])
         df['means'] = df.mean(axis=1)
         df['std'] = df.std(axis=1)

In [17]: df
Out[17]:
   actual value  col2  means  std
0             1     2    1.5  0.5
1             0     0    0.0  0.0

所以数据框没有任何Nans，但是计算呢？

In [21]: [(df['actual value'][i]-df['means'][i])/df['std'][i] for i in range (len(df['means']))]
Out[21]: [-1.0, nan]

现在，当您调用 int 时，您会在结果列表中收到错误消息。 最后，我建议（如果可能的话）直接在底层数组中执行操作，而不是使用 for 循环，因为它会更快。

In [25]: (df['actual value']-df['means'])/df['std']
Out[25]:
0    -1
1   NaN
dtype: float64

这可能无法实现，具体取决于所需的 0 除法的返回值。