If you

datetimes

key included a list of columns I would map a dictionary of column name to date type pairs and pass those to

df.astype