Question

I want to apply a transformation to certain columns of a record array and reassign these values. What is the canonical way to do this?

List = [['a',.3,.3],
        ['b',-.5,.4]]

Arr = np.rec.fromrecords(List,names=['id','var1','var2'])

And I want to apply some scaling to some variables. I’ll let this be an example but in general the scaling may be more complex than subtracting means.

scale = lambda x: x - x.mean(0)

This does not work

Arr[['var1','var2']].mean(0)
TypeError: cannot perform reduce with flexible type

So I have to convert to an unstructured array first, but then how to reassign back to the record array? Do I loop?

Unstr = Arr[['var1','var2']].view('float').reshape(len(Arr),-1)

Arr[['var1','var2']] = scale(Unstr)

IndexError: unsupported iterator index

Answer 1

Why don’t you scale variables one by one?

In [13]: %paste
import numpy as np

List = [['a',.3,.3],
        ['b',-.5,.4]]

scale = lambda x: x - x.mean()
Arr = np.rec.fromrecords(List,names=['id','var1','var2'])

vars_to_scale = ['var1', 'var2']
for var in vars_to_scale:
    Arr[var] = scale(Arr[var])
print Arr
## -- End pasted text --
[('a', 0.4, -0.04999999999999999) ('b', -0.4, 0.050000000000000044)]