python numpy matrix row wise operations: columns in each row

Question

How can I loop over a matrix and get the number of columns in each row? If I have a matrix, and some elements are NaN (empty) in the matrix, e.g.: [[4,2,9,4],[3,4,8,6],[5,NaN,7,7],[Nan,8,Nan,Nan]], how can I compute the row-wise length?

I have tried:

len(matrix) # number of rows
=len(matrix[0]) # number of columns

But that gives me the total number.

So I want to get a vector saying the number of columns in each row: [4,4,3,1] e.g.

My idea is to make a loop like this:

for i in matrix:

And then a loop where it searches. But I'm not sure how to do this

EDIT: I tried @wavy's method and it worked. Can I do like this:

# empty list
Final=[]

for i in range(matrix):
    columns=np.isnan(matrix).sum(axis=1)
    result=-columns+matrix.shape[1]
    if result==1:
        Final.append(matrix[i])
        
        
    print(Final)

I also need to put other conditions, when result==2, and when result>2

In a numpy array all columns have the same length. np.nan doesn't count as "empty". Sounds instead like you want to count the number of non-nan values. Which is fine, but the description should be clearer. — hpaulj
– hpaulj, Commented Jun 20, 2020 at 15:39

Wavy · Accepted Answer · 2020-06-20 10:58:12Z

1

This might be faster than David Wierichs suggestion:

import numpy as np 

x = np.array([[4, 2, 9, 4], [3, 4, 8, 6], [5, np.nan, 7, 7], [np.nan, 8, np.nan, np.nan]])
y = np.isnan(x).sum(axis=1)
result = -y + x.shape[1]

answered Jun 20, 2020 at 10:58

Wavy

1887 bronze badges

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

6 Comments

user11658272 Over a year ago

Thank you! Now that I have this, can I then somehow get it to say "if length of that row is 1 (so only where there is 1 number) then the number in the new vector should be that number"?

user11658272 Over a year ago

Something like this:

for i in range(matrix):         columns=np.isnan(matrix).sum(axis=1)         result=-columns+matrix.shape[1]         if result==1: # then here it should keep that number in the vector where result is 1                      print(result)

Wavy Over a year ago

After having done what I posted there you can try this: x[np.isnan(x)] = 0 for i in range(len(result)): if result[i] == 1: result[i] = np.max(x[i, :]). There's likely a faster way than looping over the result array, but I cannot immediately think of it.

user11658272 Over a year ago

Is it possible somehow I can ask you some more questions in any way?

Wavy Over a year ago

I'm online for a bit longer, we can chat here I think: chat.stackoverflow.com/rooms/216320/discuss-numpy-question

|

David Wierichs · Accepted Answer · 2020-06-20 10:56:01Z

0

You could loop over the rows and for each row use the (negated) numpy.isnan method:

lengths = [np.sum(~np.isnan(row)) for row in matrix]

As this builds up a boolean array in the np.isnan, there might be faster approaches.

answered Jun 20, 2020 at 10:56

David Wierichs

5454 silver badges11 bronze badges

Collectives™ on Stack Overflow

python numpy matrix row wise operations: columns in each row

2 Answers 2

6 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

6 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related