Searching & Counting (argmax, argmin, nonzero, count_nonzero)0%

Set Operations & Distinct Elements (np.unique, intersect1d, isin)

Searching & Counting (argmax, argmin, nonzero, count_nonzero)

Beginner12 min read•Updated: Jul 11, 2026

Locating Extrema: argmax(), argmin(), & nonzero()

Finding where extreme values occur—such as locating the maximum probability class in neural network classification, pinpointing the peak signal in a spectroscopy scan, or identifying active sensor triggers—is a standard task across numerical science.

NumPy provides fast vector routines for searching arrays:

np.argmax() / np.argmin(): Indices of global or directional extrema.
np.unravel_index(): Converting flat 1D indices into multi-dimensional coordinates.
np.nonzero(): Finding coordinates of non-zero elements.

1. Locating Extrema in 1D Arrays

1import numpy as np
2 
3sensor_voltages = np.array([0.45, 1.20, 4.88, 2.15, 0.95, 4.88])
4 
5# Index of the first maximum element
6max_idx = np.argmax(sensor_voltages)
7print("Max index:", max_idx)               # Index 2
8print("Max value:", sensor_voltages[max_idx]) # 4.88
9 
10# Index of the minimum element
11min_idx = np.argmin(sensor_voltages)
12print("Min index:", min_idx)               # Index 0
13print("Min value:", sensor_voltages[min_idx]) # 0.45

Pro Tip

Tie-Breaking Rule: If the maximum or minimum value occurs multiple times in the array, both argmax() and argmin() return the index of the first occurrence.

2. Directional Extrema with the `axis` Parameter

In classification machine learning (such as MNIST digit recognition or image tagging), the model outputs a 2D matrix of shape (batch_size, num_classes) containing predicted probability scores. To extract the predicted class for each sample, compute argmax across class columns (axis=1):

1# 4 samples, 3 classes (e.g. [Cat, Dog, Bird])
2predicted_probabilities = np.array([
3    [0.10, 0.85, 0.05],  # Sample 0 -> Dog (index 1)
4    [0.72, 0.18, 0.10],  # Sample 1 -> Cat (index 0)
5    [0.05, 0.25, 0.70],  # Sample 2 -> Bird (index 2)
6    [0.30, 0.60, 0.10]   # Sample 3 -> Dog (index 1)
7])
8 
9# Extract predicted class label for every sample
10predicted_classes = np.argmax(predicted_probabilities, axis=1)
11print("Predicted classes:", predicted_classes)  # [1, 0, 2, 1]

3. Multi-Dimensional Coordinates with `np.unravel_index()`

When argmax() is called on a multi-dimensional array without specifying an axis, it flattens the array conceptually and returns a single integer representing the flat index.

To convert that flat integer into true $(row, col)$ or $(depth, row, col)$ coordinates, use np.unravel_index():

1heatmap = np.array([
2    [12, 45, 23],
3    [88, 15, 99],   # 99 is maximum at row 1, col 2
4    [34, 76, 50]
5])
6 
7# 1. Flat argmax
8flat_idx = np.argmax(heatmap)
9print("Flat max index:", flat_idx)  # Index 5 (in flattened [12, 45, 23, 88, 15, 99...])
10 
11# 2. Unravel into true 2D matrix coordinates:
12coords = np.unravel_index(flat_idx, heatmap.shape)
13print("2D Matrix coordinates (row, col):", coords)  # (1, 2)
14print("Value at coordinates:", heatmap[coords])      # 99

4. Finding Non-Zero Elements with `np.nonzero()`

np.nonzero() returns a tuple of coordinate arrays, one for each dimension, containing the indices where elements are non-zero:

1sparse_grid = np.array([
2    [0, 5, 0],
3    [3, 0, 0],
4    [0, 0, 9]
5])
6 
7rows, cols = np.nonzero(sparse_grid)
8print("Non-zero row indices:", rows)  # [0, 1, 2]
9print("Non-zero col indices:", cols)  # [1, 0, 2]
10 
11# Count non-zero elements fast:
12print("Total non-zero count:", np.count_nonzero(sparse_grid))  # 3

Interactive Quiz• Knowledge Check

Multiple Choice Questions

Question 1

In the case of multiple identical maximum values in an array, which index does np.argmax() return?

Question 2

In a neural network probability matrix probs of shape (100, 10) (100 samples, 10 classes), which command extracts the predicted class for each sample?

Question 3

Which NumPy function maps a flat 1D argmax index back to a tuple of multi-dimensional matrix coordinates?

Question 4

What does np.count_nonzero(arr) do?

Question 5

If arr = np.array([[0, 8], [3, 0]]), what is returned by np.nonzero(arr)?

Next Lesson

Set Operations & Distinct Elements (np.unique, intersect1d, isin)

Continue learning with hands-on practice, examples, and exercises in the upcoming topic.

Related Lessons

Previous Lesson	Next Lesson
Sorting Arrays & Extracting Indices (np.sort vs np.argsort)	Set Operations & Distinct Elements (np.unique, intersect1d, isin)

Practice Quiz

Test your understanding of this lesson with 5 questions. Each question has one correct answer.

Previous Next

PreviousPrevNext

MSK Institute • Shikohabad

Loading Page Content...

Preparing courses, live labs, and learning resources.सामग्री लोड हो रही है, कृपया प्रतीक्षा करें...

Connecting serverPlease wait...

💻 Practical Labs🚀 Free Demo📜 Verified Certificates

Set Operations & Distinct Elements (np.unique, intersect1d, isin)

Searching & Counting (argmax, argmin, nonzero, count_nonzero)

Beginner12 min read•Updated: Jul 11, 2026

Study Materials Cheatsheet

Locating Extrema: argmax(), argmin(), & nonzero()

NumPy provides fast vector routines for searching arrays:

np.argmax() / np.argmin(): Indices of global or directional extrema.
np.unravel_index(): Converting flat 1D indices into multi-dimensional coordinates.
np.nonzero(): Finding coordinates of non-zero elements.

1. Locating Extrema in 1D Arrays

1import numpy as np
2 
3sensor_voltages = np.array([0.45, 1.20, 4.88, 2.15, 0.95, 4.88])
4 
5# Index of the first maximum element
6max_idx = np.argmax(sensor_voltages)
7print("Max index:", max_idx)               # Index 2
8print("Max value:", sensor_voltages[max_idx]) # 4.88
9 
10# Index of the minimum element
11min_idx = np.argmin(sensor_voltages)
12print("Min index:", min_idx)               # Index 0
13print("Min value:", sensor_voltages[min_idx]) # 0.45

Pro Tip

Tie-Breaking Rule: If the maximum or minimum value occurs multiple times in the array, both argmax() and argmin() return the index of the first occurrence.

2. Directional Extrema with the `axis` Parameter

1# 4 samples, 3 classes (e.g. [Cat, Dog, Bird])
2predicted_probabilities = np.array([
3    [0.10, 0.85, 0.05],  # Sample 0 -> Dog (index 1)
4    [0.72, 0.18, 0.10],  # Sample 1 -> Cat (index 0)
5    [0.05, 0.25, 0.70],  # Sample 2 -> Bird (index 2)
6    [0.30, 0.60, 0.10]   # Sample 3 -> Dog (index 1)
7])
8 
9# Extract predicted class label for every sample
10predicted_classes = np.argmax(predicted_probabilities, axis=1)
11print("Predicted classes:", predicted_classes)  # [1, 0, 2, 1]

3. Multi-Dimensional Coordinates with `np.unravel_index()`

When argmax() is called on a multi-dimensional array without specifying an axis, it flattens the array conceptually and returns a single integer representing the flat index.

To convert that flat integer into true $(row, col)$ or $(depth, row, col)$ coordinates, use np.unravel_index():

1heatmap = np.array([
2    [12, 45, 23],
3    [88, 15, 99],   # 99 is maximum at row 1, col 2
4    [34, 76, 50]
5])
6 
7# 1. Flat argmax
8flat_idx = np.argmax(heatmap)
9print("Flat max index:", flat_idx)  # Index 5 (in flattened [12, 45, 23, 88, 15, 99...])
10 
11# 2. Unravel into true 2D matrix coordinates:
12coords = np.unravel_index(flat_idx, heatmap.shape)
13print("2D Matrix coordinates (row, col):", coords)  # (1, 2)
14print("Value at coordinates:", heatmap[coords])      # 99

4. Finding Non-Zero Elements with `np.nonzero()`

np.nonzero() returns a tuple of coordinate arrays, one for each dimension, containing the indices where elements are non-zero:

1sparse_grid = np.array([
2    [0, 5, 0],
3    [3, 0, 0],
4    [0, 0, 9]
5])
6 
7rows, cols = np.nonzero(sparse_grid)
8print("Non-zero row indices:", rows)  # [0, 1, 2]
9print("Non-zero col indices:", cols)  # [1, 0, 2]
10 
11# Count non-zero elements fast:
12print("Total non-zero count:", np.count_nonzero(sparse_grid))  # 3

Interactive Quiz• Knowledge Check

Multiple Choice Questions

Question 1

In the case of multiple identical maximum values in an array, which index does np.argmax() return?

Question 2

In a neural network probability matrix probs of shape (100, 10) (100 samples, 10 classes), which command extracts the predicted class for each sample?

Question 3

Which NumPy function maps a flat 1D argmax index back to a tuple of multi-dimensional matrix coordinates?

Question 4

What does np.count_nonzero(arr) do?

Question 5

If arr = np.array([[0, 8], [3, 0]]), what is returned by np.nonzero(arr)?

Next Lesson

Set Operations & Distinct Elements (np.unique, intersect1d, isin)

Continue learning with hands-on practice, examples, and exercises in the upcoming topic.

Related Lessons

Previous Lesson	Next Lesson
Sorting Arrays & Extracting Indices (np.sort vs np.argsort)	Set Operations & Distinct Elements (np.unique, intersect1d, isin)

Practice Quiz

Test your understanding of this lesson with 5 questions. Each question has one correct answer.

Previous Next

Locating Extrema: argmax(), argmin(), & nonzero()

1. Locating Extrema in 1D Arrays

2. Directional Extrema with the axis Parameter

3. Multi-Dimensional Coordinates with np.unravel_index()

4. Finding Non-Zero Elements with np.nonzero()

Multiple Choice Questions

Set Operations & Distinct Elements (np.unique, intersect1d, isin)

Related Lessons

Practice Quiz

Loading Page Content...

Locating Extrema: argmax(), argmin(), & nonzero()

1. Locating Extrema in 1D Arrays

2. Directional Extrema with the axis Parameter

3. Multi-Dimensional Coordinates with np.unravel_index()

4. Finding Non-Zero Elements with np.nonzero()

Multiple Choice Questions

Set Operations & Distinct Elements (np.unique, intersect1d, isin)

Related Lessons

Practice Quiz

2. Directional Extrema with the `axis` Parameter

3. Multi-Dimensional Coordinates with `np.unravel_index()`

4. Finding Non-Zero Elements with `np.nonzero()`

2. Directional Extrema with the `axis` Parameter

3. Multi-Dimensional Coordinates with `np.unravel_index()`

4. Finding Non-Zero Elements with `np.nonzero()`