reduceat should handle outlier indices gracefully (Trac #237)

[numpy-gitbot]

Original ticket http://projects.scipy.org/numpy/ticket/237 on 2006-08-07 by trac user martin_wiechert, assigned to unknown.

to the very least .reduceat (a, I) should accept len (a) in I

usage example:

def weighted_histo (binborders, data):
    sd = sort (data)
    I = sd.searchsorted (binborders)
    # Now elements of I may range between 0 and len (sd).
    return add.reduceat (sd, I)

[numpy-gitbot]

@teoliphant wrote on 2006-08-08

Be more specific about the desired behavior. Instead of raising an error, should it instead return the identity of the operator (assuming one is defined).

[numpy-gitbot]

trac user martin_wiechert wrote on 2006-08-08

At least for nondecreasing I the "natural" (read: expected by naive user like me) behaviour would be

<operator>.reduceat (a, I) ==
array ([<operator>.reduce (a [i:j]) for i,j in zip (I, r_[I [1:], len (a)])])

Cf. also ticket #834.

[numpy-gitbot]

Milestone changed to 1.1 by @alberts on 2007-05-12

[numpy-gitbot]

Milestone changed to Unscheduled by @cournape on 2009-03-02