[pytables-users] Table vs CArray

[Ken Walker]

I'm looking for insights regarding when to use a *Table* and when to use an
*CArray*.
As I understand:
-Tables are always 2D with unstructured data types.
-All array types can have any dimension (1D, 2D, 3D, etc).
-Arrays can be NumPy arrays and scalars
-Carrays can have mixed types (defined by Atom class), similar to
NumPy unstructured arrays.
If I have that right, what's the difference between a Table and 2D CArray?
(especially if I don't need a enlargeable array)
Are there advantages of one over the other?

For background, I'm working with HDF5 data provided to me (I don't control
the format).
All datasets are saved in Table format, and PyTables table methods have
been sufficient to find and manipulate the data I need.
The data values are floats with ints that represent location and time IDs.
Here are 2 examples of exported .coldtypes and .coldtypes.shape:
LOC_ID := int64, ()
UX := float64, ()
UY := float64, ()
UZ := float64, ()
TIME_ID := int64, ()

LOC_ID := int64, ()
VALUE := ('<f8', (6,)), (6,)
TIME_ID := int64, ()

A typical dataset has 10e5 LOC_IDs for each TIME_ID.
The number of TIME_IDs is highly variable. Some datasets will have 1, other
sets could have 1000s.
The largest data set might have 500,000 LOC_IDs and 20,000 TIME_IDs (10e10
rows).
I usually want to access all LOC_ID data for one TIME_ID (but there are
times I slice in other ways).
table.where and table.read have been very useful to slice the data as
needed.

Are there any benefits to reorganizing the Table into a 3D CArray with the
same columns/rows except using the LOC_ID as the third dimension?

Thanks,
-Ken