Numpy record array - field names for all dimensions

Robert Kern robert.kern at
Wed Dec 3 22:22:04 CET 2008

ShanMayne wrote:
> Greetings All

Greetings! If you have more numpy questions, you will find numpy-discussion to 
be a better forum:

> I am seeking to represent datasets where each data element is the
> calculated result from several (4 for now) other data types. A matrix-
> like (in the general mathematical sense) seems logical, where the
> intersection of each of the 4 values (from different data sets) holds
> the value derived from those 4 values here serving as indexes.
> So, each matrix/array element is associated with 4 fields.
> eg:
> matrix element/output value =  24.235  -->
>      'Formula' = 'C12H24O2N2'
>      'Solvent' =  'Acetonitrile'
>      'fragmentation_method' = 'CID'
>      'resolution' = 'unit'
> ideally I would like to call the output value by indexing the matrix
> with the input information. eg:
> matrix['C12H24O2N2']['Acetonitrile']['CID']['unit'] = 24.235
> Numpy's record arrays seemingly don't allow all dimensions to carry
> field names. ie. each column/row carrying a label. Instead fieldname
> usage appears to create a "new dimension" as denoted by square
> brackets.

Pretty much. You can make nested dtypes, but that's not really the data 
structure that you want. You probably want a simple dictionary.

d = {
   ('C12H24O2N2','Acetonitrile','CID','unit'): 24.235,

assert d['C12H24O2N2','Acetonitrile','CID','unit'] == 24.235

If you want to make partial queries (e.g. Formula='C12H23O2N2' and 
resolution='unit'), this becomes more like a typical relational database, but 
you can probably get along with a few simple functions to loop over the 
dictionary and pull out the relevant keys pretty quickly.

Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
  that is made terrible by our own mad attempt to interpret it as though it had
  an underlying truth."
   -- Umberto Eco

More information about the Python-list mailing list