ENH: astype() can now take col label -> dtype mapping as arg; GH7271

[StephenKappel]

New PR for what was started in #12086.

closes #7271

By passing a dict of {column name/column index: dtype}, multiple columns can be cast to different data types in a single command. Now users can do:

df = df.astype({'my_bool', 'bool', 'my_int': 'int'})

or:

df = df.astype({0, 'bool', 1: 'int'})

instead of:

df['my_bool'] = df.my_bool.astype('bool')
df['my_int'] = df.my_int.astype('int')

[codecov-io]

Current coverage is 84.38%

Merging #13375 into master will increase coverage by <.01%

@@             master     #13375   diff @@
==========================================
  Files           142        142          
  Lines         51223      51241    +18   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
+ Hits          43224      43242    +18   
  Misses         7999       7999          
  Partials          0          0          

Powered by Codecov. Last updated by ada6bf3...5fe82e3

[jreback]

add pandas extension type

[StephenKappel]

Updated

[jreback]

elif

give a nice error message with the exception

[jorisvandenbossche]

Can you make this something like "data type, or dict of column -> type".
And then you can sum the different options (numpy.dtype, Python type, pandas extension type) in the explanation below.

[jorisvandenbossche]

@StephenKappel Can you rebase this?

This will raise an error when you have duplicate columns (also if it is not the column you are astyping). But maybe we are OK with that?

[jreback]

this is going to raise if the columns are duplicates (not the dtype dict). Put them in order first (the original order). before passing to the concat. Further doing self[col] also will not work with duplicates. something like:

results = []

# this is safe for duplicates
for c, col in self:
    if c in dtype:
      results.append( # cast col here)
    else:
     results.append(col)

new_df = concat(results, axis=1, copy=False)
...

[jreback]

can you rebase / update

[StephenKappel]

rebased and updated to work with duplicate column names

[jreback]

you cannot contruct a duplicate this way it only has non duplicate keys

[jreback]

construct via unique named series and then set the columns
or use a list

[StephenKappel]

Jeez. I swear I know better than that... not sure what I was thinking. Fixed now.

[jreback]

thanks! nice PR