numpy.unique throwing error

Question

I'm trying to use the following function:

def randomChose(bp, xsteps, ysteps, bs):
    # Number of points to be chosen
    s = int((bp * xsteps * ysteps) / (bs * bs))

    # Generating an array representing the input indexes
    indices = numpy.arange(xsteps * ysteps)

    # Resampling without replacement
    cs = npr.choice(indices, size=s, replace=False)

    f = []
    for idx in cs:
        nb = indices[max(idx-(bs*bs/2), 0):min(idx+(bs*bs/2)+1, xsteps*ysteps)]
        f.append(nb)
    f = numpy.array(f).flatten()
    fix = numpy.unique(numpy.array(f))

    return fix

Which takes as parameter a number bp, the data dimension xsteps * ysteps and a % bs.

What I want to do is to choose a number of valid indexes considering some neighborhood in this image.

However, I keep receiving error when calling numpy.unique , not always, though:

ValueError                                Traceback (most recent call last)
<ipython-input-35-1b5914c3cbc7> in <module>()
      9     svf_y = []
     10     for s in range(samples):
---> 11         fix = randomChose(bp, xsteps, ysteps, bs)
     12         rs_z0, rs_z1, rs_z2 = interpolate(len(fix), xsteps, ysteps, mean_rs)
     13         ds_z0, ds_z1, ds_z2 = interpolate(len(fix), xsteps, ysteps, mean_ds)

<ipython-input-6-def08adce84b> in randomChose(bp, xsteps, ysteps, bs)
     14         f.append(nb)
     15     f = numpy.array(f).flatten()
---> 16     fix = numpy.unique(numpy.array(f))
     17 
     18     return f

/usr/local/lib/python2.7/dist-packages/numpy/lib/arraysetops.pyc in unique(ar, return_index, return_inverse, return_counts)
    198         ar.sort()
    199         aux = ar
--> 200     flag = np.concatenate(([True], aux[1:] != aux[:-1]))
    201 
    202     if not optional_returns:

ValueError: all the input arrays must have same number of dimensions

This is how I call it:

nx = 57.2
ny = 24.0
xsteps = 144
ysteps = 106
bs = 5     # Block size
bp = 0.1   # Percentage of blocks
fix = randomChose(bp, xsteps, ysteps, bs)

I'm trying to understand what is wrong. As far as I understood, such method expect a ndarray as input, which is being given.

Thank you for any help.

Answer 1

First:

f.append(nb)

should become:

f.append(list(nb))

That makes fa list of lists, that Numpy will have a chance to convert to a Numpy array of integers, BUT ONLY if all the lists have the same length. If not, you will only have a one dimension Numpy array of lists, and flatten() will have no effect.

You may add a

print(type(f[0]))

after the flattening.

Answer 2

The problem is with the edges. Eg, if idx=0 ,

nb = indices[max(idx-(bs*bs/2), 0):min(idx+(bs*bs/2)+1, xsteps*ysteps)]

is going to be [0] - ie, with only one value instead of an xy coordinate. Then you will not be able to flatten your array properly.