numpy searchsorted to return nan if there is no suitable index instead of "0 or N"

Question

Context

Currently, numpy.searchsorted() will:

"if there is no suitable index, return either 0 or N"

I will be performing multiple indexing and subsetting operations on the original unsorted a using the indices generated like so:

import numpy as np

a = np.array([1,2,3,4,5])
start = np.array([0.0,1.0,3.0,3.0,5.0,6.0])
stop = np.array([1.0,3.0,3.0,5.0,6.0,7.0])

starts = np.searchsorted(a, start, side="left")
stops = np.searchsorted(a, stop, side="right")
length = range(len(starts))
indices = [np.arange(starts[i], stops[i] + 1) for i in length]

NOTE: The actual code uses 3 different numpy.ndarray (single above represented as a), one for each column of a 2D array (from a pandas.DataFrame).

Task

I would like to perform the binary search to ultimately return the rows of the original unsorted array (one for each element of (start, stop).

My approach is to use the return of np.argsort(), intersect the indices that satisfy the conditions for each search on the 3 columns (1D arrays), and then subset to retrieve the original values, not only from the input arrays of searchsorted() but other columns of the original unsorted 2D array (hence indices are required to get those rows).

As searchsorted() returns 0 or a.shape[0], it will result in creating a subset that contains all values with np.arange(). However, given my current approach, this is not suitable.

How can I return nan from saerchsorted() so that subsequent subsetting will return empty, or is there an alternative method to achieve the task?

I appreciate your assistance.