Pandas Dataframe:如何将一列拆分为多个一键编码的列

时间:2019-06-24 09:10:30

标签: python python-3.x pandas dataframe one-hot-encoding

我有一个像这样的文本文件:

444537110                         3 11112111022002200022022111121222002...

输入文件中的最后一个字段的长度为50k个字符,并且只能是0或1。2。我想要这个最终字段的一个热编码版本。所以我的预期结果是这样的数据框:

id          chip   g1_0 g1_1 g1_2 g2_0 g2_1 g2_2 g3_0 g3_1 g3_2 g4_0 ... 
444537110   3      0    1    0    0    1    0    0    1    0    0

我通过读取输入文件创建了一个初始数据框:

df = pd.read_csv('test.txt', index_col=0, sep='\s+', header=None, names = ['chip', 'genos'])

这将创建一个包含3列的数据框:

id        chip  genos
444537110    3  1111211102200220000022022111121222000200022002...

我认为我可以使用如下所示的方法创建初始的单个列,然后使用pandas get_dummies函数进行一次热编码,但是我无法创建单个列。我尝试过

[c for c in df['genos'].str]

但这不会分隔字符

我在这里查看过类似的问题和答案:How can I one hot encode in Python?

但这仅处理一种热编码,而不处理拆分非常大的列所带来的复杂性

3 个答案:

答案 0 :(得分:1)

首先创建DataFrame,并将字符串转换为列表,然后调用get_dummies

df1 = pd.DataFrame([list(x) for x in df['genos']], index=df.index).add_prefix('g')
df2 = pd.get_dummies(df1)

如果需要将新的columnt添加到原始列中(如果可能,则缺少某些组合),将DataFrame.reindex用于带有_的拆分列以及MultiIndex.from_product创建的所有组合:

df1 = pd.DataFrame([list(x) for x in df.pop('genos')], index=df.index).add_prefix('g')
df2 = pd.get_dummies(df1)

splitted = df2.columns.str.split('_')
df2.columns = [splitted.str[0].astype(int) + 1, splitted.str[1].astype(int)]
#
mux = pd.MultiIndex.from_product([df2.columns.get_level_values(0), [0,1,2]])
df2 = df2.reindex(mux, axis=1, fill_value=0)
df2.columns = [f'g{a}_{b}' for a, b in df2.columns]
print (df2)
   g1_0  g1_1  g1_2  g2_0  g2_1  g2_2  g3_0  g3_1  g3_2  g4_0  ...  g32_2  \
0     0     1     0     0     1     0     0     1     0     0  ...      1   

   g33_0  g33_1  g33_2  g34_0  g34_1  g34_2  g35_0  g35_1  g35_2  
0      1      0      0      1      0      0      0      0      1  

[1 rows x 105 columns]

答案 1 :(得分:1)

考虑到@Dan对您将要以50k * 3列结尾这一事实的评论,您可以这样做来获得所需的输出:

string ="444537110 3 11112111022002200022022111121222002"
df = pd.DataFrame([string.split(" ")],columns=['id','chip','genos'])
max_number_of_genes = int(df.genos.apply(lambda x : len([y for y in x])).max())

#Create columns 
for gene in range(1,max_number_of_genes+1):
    for y in range(4):
        df['g{}_{}'.format(gene, y)] = 0

#Iterating over genos values 
for row_number, row in df.iterrows():
    genos = [int(x) for x in df.iloc[row_number, 2]]
    for gene_number, gene in enumerate(genos):     
        df.loc[row_number, 'g{}_{}'.format(gene_number+1, gene)] = 1 

print(df)

输出

+----+------------+-------+--------------------------------------+-------+-------+-------+-------+-------+-------+-------+------+--------+--------+--------+--------+--------+--------+--------+--------+--------+-------+
|    |    id      | chip  |                genos                 | g1_0  | g1_1  | g1_2  | g1_3  | g2_0  | g2_1  | g2_2  | ...  | g33_2  | g33_3  | g34_0  | g34_1  | g34_2  | g34_3  | g35_0  | g35_1  | g35_2  | g35_3 |
+----+------------+-------+--------------------------------------+-------+-------+-------+-------+-------+-------+-------+------+--------+--------+--------+--------+--------+--------+--------+--------+--------+-------+
| 0  | 444537110  |    3  | 11112111022002200022022111121222002  |    0  |    1  |    0  |    0  |    0  |    1  |    0  | ...  |     0  |     0  |     1  |     0  |     0  |     0  |     0  |     0  |     1  |     0 |
+----+------------+-------+--------------------------------------+-------+-------+-------+-------+-------+-------+-------+------+--------+--------+--------+--------+--------+--------+--------+--------+--------+-------+

答案 2 :(得分:0)

如果仅拆分5万个字符,则可以使用原始Python(出于可读性考虑):

>>> a,b,c = zip(*[{0:(1,0,0),1:(0,1,0),2:(0,0,1)}[int(c)] for c in df['genos'][0]])
>>> a,b,c
((0, 0, 0, 0, 0, 0, ...), (1, 1, 1, 1, 0, 1, ...), (0, 0, 0, 0, 1, 0, ...))