Question

我想基于现有列向DataFrame添加新列。新列只是三列三列值的元组：

df0.shape
# (5410185, 17)
new_col = pd.Series(list(zip(df0['a'], df0['b'], df0['c'])))
new_col.shape
# (5410185,)
new_col.isnull().sum()
# 0
df0['abc'] = new_col
df0['abc'].isnull().sum()
# 14334

我在样本df上尝试了相同的方法，它按预期工作：

test = pd.DataFrame(np.random.randint(0,1000,100000000).reshape(1000000,100))
test['new'] = pd.Series(list(zip(test[1], test[2], test[3])))
test['new'].isnull().sum()
# 0

＆＃39;分配＆＃39;也会产生相同的结果：

df0 = df0.assign(new_col2 = pd.Series(list(zip(df0['a'], df0['b'], df0['c']))))
df0['new_col2'].isnull().sum()
# 14334

我发现了两个类似的问题，this和this。我怀疑我的问题也与索引有关。似乎有89个不相同的值：

np.sum(df0.index == new_col.index)
# 89

指定与df0索引相同的系列：

df0.index = new_col
df0['abc'] = df0.index
df0['abc'].isnull().sum()
# 0

更新以下是@ jezreal解决方案的一些基准测试：

%time df0['abc'] = pd.Series(list(zip(df0['a'], df0['b'], df0['c'])), index=df0.index)
Wall time: 2.32 s

% time df0['abc'] = df0[['a','b','c']].apply(tuple, axis=1)
Wall time: 1min 42s

%time df0['abc'] = df0.set_index(['a','b','c']).index.values
Wall time: 8.68 s

% time df0['abc'] = pd.Series([tuple(x) for x in df0[['a','b','c']].values.tolist()], index=df0.index)
Wall time: 9.83 s

Answer 1

我认为需要与新df0的{{1}}相同的索引来对齐数据：

Series

或使用df0['abc'] = pd.Series(list(zip(df0['a'], df0['b'], df0['c'])), index=df0.index)：

apply

样品：

df0['abc'] = df0[['a','b','c']].apply(tuple, axis=1)

df0 = pd.DataFrame({'a':list('abcdef'),
                   'b':[4,5,4,5,5,4],
                   'c':[7,8,9,4,2,3],
                   'D':[1,3,5,7,1,0],
                   'E':[5,3,6,9,2,4],
                   'F':list('aaabbb')}, index=[1,1,2,2,9,10])

print (df0)
    D  E  F  a  b  c
1   1  5  a  a  4  7
1   3  3  a  b  5  8
2   5  6  a  c  4  9
2   7  9  b  d  5  4
9   1  2  b  e  5  2
10  0  4  b  f  4  3

df0['abc'] = pd.Series(list(zip(df0['a'], df0['b'], df0['c'])))

print (df0)
    D  E  F  a  b  c        abc
1   1  5  a  a  4  7  (b, 5, 8)
1   3  3  a  b  5  8  (b, 5, 8)
2   5  6  a  c  4  9  (c, 4, 9)
2   7  9  b  d  5  4  (c, 4, 9)
9   1  2  b  e  5  2        NaN
10  0  4  b  f  4  3        NaN

df0['abc'] = pd.Series(list(zip(df0['a'], df0['b'], df0['c'])), index=df0.index)

使用pd.Series（）从现有列添加新列会创建NaN值

1 个答案: