Display column names for as used in pytable

[uvemas]

User lkjlkj9asjpd9jp said:

Hi,
i am using python pandas to generate hdf tables (pytables is working behind the scenes). When I open the generated hdf5 file I can open the table and display its contents, however it seems that multiple columns are joined together apparently based on their data type (all float columns go together, all int columns as well, etc.). The column names are generic like 'values_block_1'. The information of the original column names is stored in the user attributes of the table where name='values_block_1' and value=[<orig_col_name0>, <o_c_n1>, ...]

I would like to have the possibility to display the columns and the data in the original layout: original column names, original number and order of columns. This would make it easier working with the data (e.g. comparing with different table)

[ankostis]

Maybe somenone shoudlimplement another MVC based on pandas dataFrames,
similar to vitables.vttables package.
Does this make sense?

[ankostis]

@uvemas (and @lkjlkj9asjpd9jp) I finally did the sacrilegious cludge and used pandas when reading DataFrame-created tables.
It just works, ie assumes well-behaved dataframes, no vertical-index, etc.
Yoy can find it in my pandas branch (efa60b2).

I had to move some methods from the model to the buffer, where all the "is df" logic currently resides.
Having done that, I believe is not right - I should have crafted 2 separate DFModel + DFBuffer classes, because curring all the extra weight for arrays just complicated the whole thing; DataFrames have almost all things required readily available (i.e. chunking. headers/indices).

Anyhow, a small refactor of the Lea-model along the lines i started is definitely needed.