numpy.searchsorted with 2D array

Question

I have this numpy array where the values in each row will always be sorted and monotonically increasing:

a = np.array([[1, 2, 3, 4, 8],
       [2, 5, 6, 7, 8],
       [5, 7, 11, 12, 13]])

and I want to search for the following values (which are NOT sorted or monotonic) for each row:

b = np.array([4.5, 2.3, 11.6])

so that I get an answer of:

[4, 1, 3]

However, searchsorted does not support this (it feels like it needs an axis keyword).

Is there an EFFICIENT way I can do this for a very large array? Obviously with a for loop I can index the array a and b like this:

for i in np.arange(np.alen(a)):
     print a[i].searchsorted(b[i])

but this is slow when a is large.

Is there a way to do this in numpy that is more efficient?

why the negative score on this question?

ccbunney
– ccbunney

2014-03-26 08:52:25 +00:00
Commented Mar 26, 2014 at 8:52 — ccbunney
– ccbunney, Commented Mar 26, 2014 at 8:52

Andy Hayden · Accepted Answer · 2014-03-21 17:13:30Z

2

You can searchsorted on the ravel/flattened array:

In [11]: np.searchsorted(a.ravel(), b)
Out[11]: array([3, 6])

You can then use divmod on the result (which gets the row and column):

In [12]: divmod(np.searchsorted(a.ravel(), b), a.shape[1])
Out[12]: (array([0, 1]), array([3, 1]))

answered Mar 21, 2014 at 17:13

Andy Hayden

378k110 gold badges640 silver badges546 bronze badges

Sign up to request clarification or add additional context in comments.

7 Comments

Jaime Over a year ago

That requires that every row has larger values then any preceding row, which is the case of the OP's example, but seems a strong condition for general arrays.

ccbunney Over a year ago

@Andy - an elegant solution to the example in the question, but, as Jamie says, it does not work for cases where b is not monotonically ascending. I have updated the question so it is more clear. Thanks.

Andy Hayden Over a year ago

@KernowBunney ah, ok so something more efficient but similar to np.diag(np.apply_along_axis(np.searchsorted, 1, a, b)). Dropping to cython is an easy option for perf of your for loop.

ccbunney Over a year ago

@Andy Hayden - thanks for that - apply_along_axis works and is at least more succinct/Numpy-like than a Python loop. I will investigte Cython. Cheers!

Mad Physicist Over a year ago

@GoingMyWay. apply_along_axis is a glorified python level for loop under the hood, just like vectorize

|

Collectives™ on Stack Overflow

numpy.searchsorted with 2D array

1 Answer 1

7 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

7 Comments

Your Answer

Sign up or log in

Post as a guest

Related