[SciPy-user] numpy.asarray Numeric.asarray, fmin_* return type

[Steve Schmerler]

Steve Schmerler wrote:
> Hi
> 
> Using optimize.fmin_powell from a recent svn build with the initial 
> guess x0 being a scalar I get this error:
> 
> Traceback (most recent call last):
>    File "trans.py", line 151, in ?
>      tau = fmin_powell(func, tau_start)
>    File "/usr/lib/python2.3/site-packages/scipy/optimize/optimize.py", 
> line 1453, in fmin_powell
>      N = len(x)
> TypeError: len() of unsized object
> 
> where x = asarray(x0). Comparing numpy's and Numeric's asarray I found
> 
> ===================================================================================
> In [8]: import numpy as nu
> 
> In [9]: import Numeric as NU
> 
> In [10]: x = nu.asarray([1,2]); len(x)
> Out[10]: 2
> 
> In [11]: y = NU.asarray([1,2]); len(y)
> Out[11]: 2
> 
> In [12]: x = nu.asarray(3); len(x)
> ---------------------------------------------------------------------------
> exceptions.TypeError                                 Traceback (most 
> recent call last)
> 
> /home/elcorto/<console>
> 
> TypeError: len() of unsized object
> 
> In [13]: y = NU.asarray(3); len(y)
> Out[13]: 1
> 
> ===================================================================================
> 

OK an easy workarround is to provide "scalar" start values as x = 
array([x0]) instead of just x0, then len(x) = 1.

(providing x0 as it is (scalar) worked for me with scipy 0.3.2)

But another thing:

fmin, fmin_cg and fmin_bfgs (I haven't tried the others) return
<type 'numpy.ndarray'> when the initial guess is of the same type (i.e. 
array([x0]) which seems reasonable. Unfortunately fmin_powell returns 
<type 'float64_arrtype'> in this case.

cheers,
steve