Locate populated entries
Sparse or masked data
NumPy · Function
np.nonzero
Returns the indices of non-zero elements as one index array per input dimension. It also works with boolean conditions to locate entries where a predicate is true.
How it works
np.nonzero evaluates an array-like input and returns a tuple containing one index array for each dimension. Each array holds coordinates of non-zero elements, ordered according to row-major, C-style traversal; boolean conditions can therefore be passed to locate elements satisfying a predicate.
Syntax
numpy.nonzero(a)Parameters
| Name | Type | Description |
|---|---|---|
a | array_like | Array-like input whose values are tested for being non-zero. |
Returns
A tuple of arrays containing the indices of non-zero elements, with one array for each dimension of the input.
Key points
- Use np.nonzero to locate entries whose values are not zero.
- The output is a tuple, with a separate index array corresponding to every input dimension.
- Each output array stores coordinates along its associated dimension.
- Results are generated in row-major, C-style order, giving predictable coordinate traversal.
- Pass a boolean comparison to identify locations meeting a condition rather than only literal non-zero values.
Common mistakes
- The result is organized by dimension, not as one coordinate row per element; use np.argwhere when row-wise coordinates are needed.
- For extracting values, prefer boolean indexing over indexing with np.nonzero, especially when handling zero-dimensional arrays.
Example: Find non-zero elements and matching positions
import numpy as np
x = np.array([[0, 2, 0], [3, 0, 4]])
rows, columns = np.nonzero(x)
print(rows) # [0 1 1]
print(columns) # [1 0 2]
mask = x > 2
print(np.nonzero(mask))The function returns one index array per dimension; applying it to a boolean condition finds positions where the condition is true.
When to use it
Find values satisfying a comparison
Thresholding
Obtain dimension-separated coordinates
Multidimensional coordinate extraction
Interview questions
QWhat does np.nonzero return for a two-dimensional array?
It returns a tuple with one array for each input dimension. The arrays contain the coordinates of non-zero elements, and the coordinates follow row-major C-style order.
Follow-up: How would you obtain one coordinate row per matching element? Use np.argwhere, which groups the indices by element.
QHow can np.nonzero find elements satisfying a condition?
A comparison such as a > 3 produces a boolean array. Because False is treated as zero, np.nonzero(a > 3) returns the positions where the comparison is true.
Follow-up: How can you retrieve the matching values instead of their coordinates? Use boolean indexing, for example a[a > 3].
QWhat input does np.nonzero accept, and what does it identify?
Its input parameter a is array-like. The function tests the values and reports the indices of entries that are not zero.
Follow-up: What ordering should callers expect? The returned indices follow row-major, C-style order.
Related
numpy.argwherenumpy.nonzeronumpy.atleast_1d