raise NotImplemented for date parsing args in read_excel #11544

[grahamjeffries]

Fixes #11544

The parse_dates and date_parser args are passed to TextReader and then to TextFileReader where they don't seem to have an effect. It was decided to raise the exception at the _parse_excel level however, following suit with the handling of chunksize args.

[jreback]

use double backticks are NotImplementedError and say: keywords parse_dates and date_parser

[jreback]

lgtm. minor comments. ping when updated & green.

[grahamjeffries]

Good to go. I also updated the NotImplemented message for the chunksize keyword.

[jreback]

thanks!

[jorisvandenbossche]

AFAIK, the parse_dates keyword for read_excel does work.
It only works a bit in unexpected ways sometimes, as it is a keyword to specify whether to parse strings to a datetime. And most of the time, if you have dates in an excel file, the column is formatted as dates, the engine knows this and the column will be of datetime dtype regardless of what you specify with parse_dates. Hence the confusion.
But if you have a column that is formatted as text in excel, parse_dates works perfectly to parse these strings into a datetime64 column.

So I think the more correct fix is to clarify the documentation. And maybe also warn on parse_dates=True when you didn't specify index_col

cc @TomAugspurger I just saw you giving that comment on SO: http://stackoverflow.com/questions/34403682/why-does-pandas-parse-dates-when-parse-dates-false

[jorisvandenbossche]

As a small example it actually does something:

In [43]: dti = pd.date_range('2014-1-1', periods=10)

In [44]: df = pd.DataFrame({'dates':dti, 'strings':pd.Series(dti).astype(str)})

In [46]: df.to_excel('test.xlsx', index=False)

In [47]: pd.read_excel('test.xlsx').dtypes
Out[47]:
dates      datetime64[ns]
strings            object
dtype: object

In [48]: pd.read_excel('test.xlsx', parse_dates=['strings']).dtypes
Out[48]:
dates      datetime64[ns]
strings    datetime64[ns]
dtype: object

[TomAugspurger]

Thanks, I misunderstood how read_excel handled parse_dates.

On Dec 22, 2015, at 5:14 AM, Joris Van den Bossche notifications@github.com wrote:

AFAIK, the parse_dates keyword for read_excel does work.
It only works a bit in unexpected ways sometimes, as it is a keyword to specify whether to parse strings to a datetime. And most of the time, if you have dates in an excel file, the column is formatted as dates, the engine knows this and the column will be of datetime dtype regardless of what you specify with parse_dates. Hence the confusion.
But if you have a column that is formatted as text in excel, parse_dates works perfectly to parse these strings into a datetime64 column.

So I think the more correct fix is to clarify the documentation. And maybe also warn on parse_dates=True when you didn't specify index_col

cc @TomAugspurger I just saw you giving that comment on SO: http://stackoverflow.com/questions/34403682/why-does-pandas-parse-dates-when-parse-dates-false

—
Reply to this email directly or view it on GitHub.

[grahamjeffries]

Thanks for the clarification. I'll submit another PR with doc fixes and a warning on parse_dates=True and no index_col.

[grahamjeffries]

See PR #12051