ValueError: can only convert an array of size 1 to a Python scalar for Numpy Select

Question

newX and breaks are numpy ndarray of shape (24000L,) and (7L,) respectively. newX is within range of (0,9). yvals is a lookup table based on breaks segmented into 10000 (nsegs) samples called xvals (not used here).

deltaX=(breaks[-1]-breaks[0])/nsegs
xvals=[x+i*deltaX for i in range(nsegs+1)]

Both breaks and yvals are monotonically increasing. Basically, I want yvals[0] to be returned when newX value is less than the breaks[0], and yvals[-1] when newX value is higher than breaks[-1]. For other values I want it to generate an index for yvals corresponding to a point closest to xval (an example is shown below) When I do the following:

condlist=[newX<=breaks[0] , newX < breaks[-1] , newX >= breaks[-1]]
choicelist=[yvals[0] , yvals[((newX-breaks[0]))/deltax.astype(int).item()] , yvals[-1]]
ans = np.select(condlist,choicelist)

I get the error in the choicelist line -

ValueError: can only convert an array of size 1 to a Python scalar

How can I fix this?

Answer 1

No need for numpy.select here, your goal can be simply stated

choicelist = yVals[ix]

where ix is an array of indices to be computed. What I see is newX being transformed linearly, rounded to integers, and clipped so that indices do not go beyond the range 0...len(yVals)-1. All this is expressed by

ix = np.clip(np.around((newX-breaks[0])/deltax).astype(int), 0, len(yVals)-1)

where

• np.around((newX-breaks[0])/deltax) rounds the result of computation to nearest integer. Without the rounding .astype(int) would floor it, which is less desirable.
• np.clip(..., 0, len(yVals)-1) clips the results so that all indices are valid.