6

I am grouping values by type as follows:

groups = frame.columns.to_series().groupby(frame.dtypes).groups

by I get error:

TypeError: data type not understood

What would be the right way to go about grouping columns by datatype to prevent such errors?

EDIT: Sample input

0             0       0  1985     ATL        NL  barkele01  870000         428.0   428.0      1955.0    ...           Leonard Harold   225.0   77.0    R      R  1976-09-14  1987-09-26  barkl001  barkele01       both
1             1       1  1985     ATL        NL  bedrost01  550000         559.0   559.0      1957.0    ...            Stephen Wayne   200.0   75.0    R      R  1981-08-14  1995-08-09  bedrs001  bedrost01       both
2             2       2  1985     ATL        NL  benedbr01  545000         614.0   614.0      1955.0    ...              Bruce Edwin   175.0   73.0    R      R  1978-08-18  1989-09-11  beneb001  benedbr01       both
3             3       3  1985     ATL        NL   campri01  633333           NaN     NaN         NaN    ...                      NaN     NaN    NaN  NaN    NaN         NaN         NaN       NaN        NaN  left_only
4             4       4  1985     ATL        NL  ceronri01  625000        1466.0  1466.0      1954.0    ...             Richard Aldo   192.0   71.0    R      R  1975-08-17  1992-07-10  ceror001  ceronri01       both
5             5       5  1985     ATL        NL  chambch01  800000        1481.0  1481.0      1948.0    ...      Carroll Christopher   195.0   73.0    L      R  1971-05-28  1988-05-08  chamc001  chambch01       both

Sample output would be smth like 

{float: [columns], int:[columns], string:[columns]}
jpp
  • 159,742
  • 34
  • 281
  • 339
YohanRoth
  • 3,153
  • 4
  • 31
  • 58

1 Answers1

4

You can use groupby along axis=1:

type_dct = {str(k): list(v) for k, v in df.groupby(df.dtypes, axis=1)}

For your sample dataframe, this gives:

{'int64': [0, 1, 2, 3, 7],
 'float64': [8, 9, 10, 14, 15],
 'object': [4, 5, 6, 11, 12, 13, 16, 17, 18, 19, 20, 21, 22]}

Note there is no such thing as a string series dtype with Pandas. object dtype represents pointers to arbitrary Python objects, including strings. See Strings in a DataFrame, but dtype is object for more details.

jpp
  • 159,742
  • 34
  • 281
  • 339