Choose per-row maxima
Selecting the winning class or option for each row of a score matrix.
NumPy · Function
np.argmax
Returns the integer indices of maximum values in an array, either globally or along a chosen axis. Ties select the first occurrence, and keepdims can preserve reduced dimensions.
How it works
Pass an array-like object to select the position of its largest value. With no axis, NumPy conceptually flattens the input; with an axis, it returns one index per slice, and optional keepdims preserves reduced axes as length-one dimensions.
Syntax
numpy.argmax(a, axis=None, out=None, *, keepdims=<no value>)Parameters
| Name | Type | Description |
|---|---|---|
a | array_like | Input array. |
axis | int, optional | Axis along which to search; when omitted, indexing is relative to the flattened array. |
out | array, optional | Destination array for the result; it must have an appropriate shape and dtype. |
keepdims | bool, optional | Whether reduced axes remain as dimensions of size one. |
Returns
index_array: ndarray of ints. It contains indices into the input, with the selected axis removed, or retained at size one when keepdims=True.
Key points
- argmax locates the positions of maximum values rather than returning those maximum values.
- Leaving axis as None searches the flattened input; setting an axis performs the search along that dimension.
- The out parameter lets callers place computed indices into a supplied destination array.
- keepdims=True retains reduced dimensions as singleton axes.
- The result contains integer indices, not the corresponding maximum values.
Common mistakes
- When the maximum occurs more than once, the result identifies the first occurrence rather than all matching positions.
Example: Find maximum positions
import numpy as np
x = np.array([[1, 9, 3], [8, 2, 7]])
print(np.argmax(x))
print(np.argmax(x, axis=0))
print(np.argmax(x, axis=1, keepdims=True))Computes the flattened index, column-wise indices, and row-wise indices while retaining the reduced dimension.
When to use it
Locate a global maximum
Finding the location of the largest value in an n-dimensional array.
Preserve dimensions
Producing shape-compatible index arrays for later broadcasting.
Interview questions
QWhat does np.argmax return when axis is omitted?
With the default axis=None, argmax returns an integer index into the flattened array. Supplying an axis computes indices independently along that dimension.
Follow-up: How can the multidimensional coordinates of the global maximum be recovered? Use np.unravel_index on the flattened argmax result and the input shape.
QHow do axis and keepdims affect the result shape?
The output shape removes the selected axis, so each remaining position receives the index of its maximum along that axis. With keepdims=True, the reduced axis remains with size one.
Follow-up: Why might keepdims=True be useful? It preserves dimensional alignment, allowing the result to broadcast against the original array.
QWhat are the return type and constraints on out?
The function returns an ndarray of integer indices. An optional out array can receive the result, but it must have the appropriate shape and dtype.
Follow-up: What happens with tied maxima? NumPy returns the index corresponding to the first occurrence.
Related
numpy.ndarray.argmaxnumpy.argminnumpy.amaxnumpy.unravel_index