Python Pandas 数据帧拆分

Question

I have this kind of dataFrame which I would like to split into seperate dataframes:

A B C Mark
3 5 6 T
4 5 2 T
3 4 5 B
5 6 7 B
3 4 5 T
2 5 2 T

For instance the table above should be split into three pandas dataframes. First dataframe the two rows with Mark "T" as one dataframe, the second dataframe the next two rows with Mark "B" and the third dataframe the last two rows with Mark "T".

df1

A B C Mark
3 5 6 T
4 5 2 T

df2

 A B C Mark
   3 4 5 B
   5 6 7 B

df3

A B C Mark
3 4 5 T
2 5 2 T

Answer 1

Create dictionary of DataFrames with consecutive counter by shift and cumsum with convert groupby object to tuples and then to dictionary:

dfs = dict(tuple(df.groupby(df['Mark'].ne(df['Mark'].shift()).cumsum())))
print (dfs)
{1:    A  B  C Mark
0  3  5  6    T
1  4  5  2    T, 2:    A  B  C Mark
2  3  4  5    B
3  5  6  7    B, 3:    A  B  C Mark
4  3  4  5    T
5  2  5  2    T}

Select each DataFrame:

print (dfs[1])
print (dfs[2])
print (dfs[3])

Answer 2

Create a dictionary as below:

frames = {}
for i, grp in df.groupby(df.Mark.ne(df.Mark.shift()).cumsum()):
    frames.update([('df_'+str(i),grp)])

{'df_1':    A  B  C Mark
 0  3  5  6    T
 1  4  5  2    T, 'df_2':    A  B  C Mark
 2  3  4  5    B
 3  5  6  7    B, 'df_3':    A  B  C Mark
 4  3  4  5    T
 5  2  5  2    T}

You can then test by printing all the dfs as :

print(frames['df_1'])

   A  B  C Mark
0  3  5  6    T
1  4  5  2    T

Answer 3

Another way around wrapping this into np.array_split for the given Post: However, np.array_split returns a list of DataFrames hence you can list then down and even loop through the list.

Outcome:

>>> np.array_split(df, 3)
[   A  B  C Mark
0  3  5  6    T
1  4  5  2    T,    A  B  C Mark
2  3  4  5    B
3  5  6  7    B,    A  B  C Mark
4  3  4  5    T
5  2  5  2    T]

Listing them as an individual dfs:

>>> df[0]
   A  B  C Mark
0  3  5  6    T
1  4  5  2    T

>>> df[1]
   A  B  C Mark
2  3  4  5    B
3  5  6  7    B

>>> df[2]
   A  B  C Mark
4  3  4  5    T
5  2  5  2    T

Or you can assign them names:

df1 = df[0]
df2 = df[1]
df2 = df[2]