Hello,

I wanted to point out that the definition of a "division" is not consistent between the different input formats.

divisions appear to be a tuple which is either

• Style A
  (first index in first division, last index is first division, last index in second division, ... , last index in n-1 division, last index in n division)
• Style B
  (first index in first division, first index in second division, ... , first index in n division, last index in n division)
• Style C
  All divisions are labeled as None.

I think this is how this is currently set-up...

• read_csv uses Style C
• from_array uses Style B (Except that it skips the last division when the index is already the last record. Based on the logic, I'm not sure this can every happen, actually. See here.
• from_pandas uses Style B
• from_bcolz uses Style B
• from_dask_array uses Style B
• from_castra uses Style A
• read_hdf uses Style C

So which one is the most "correct"?

Essentially, I'm trying to use a series.where statement where if a condition is met, you use one column, otherwise, you use another (the others is actually a sequence in a from_pandas array.) I think the fact the divisions are slightly different is breaking this.

Happy to program a fix. It looks straightforward enough for all but the from_castra and read_hdf.

Thanks.

Update: I started watching march madness too early and messed up some of these