Min, Max, Argmin & Argmax: Finding Extremes in Data

In NumPy, these functions help us find:

• min / max → The smallest or largest value in an array.
• argmin / argmax → The index of the smallest or largest value.

These are super useful in Deep Learning (DL) for:

• Finding the best prediction in classification models.
• Identifying extreme values in loss functions.
• Choosing optimal weights in optimization algorithms.

🔹 NumPy Examples

1️⃣ min & max

import numpy as np

arr = np.array([3, 1, 7, 0, 5])

print("Min:", np.min(arr))  # 0
print("Max:", np.max(arr))  # 7

👉 Finds the smallest and largest values.

2️⃣ argmin & argmax (Find Index)

print("Index of Min:", np.argmin(arr))  # 3 (Index of 0)
print("Index of Max:", np.argmax(arr))  # 2 (Index of 7)

👉 Returns index, not the value itself!

3️⃣ min & max on Multi-Dimensional Arrays

arr2D = np.array([[3, 7, 2], 
                  [5, 1, 8]])

print("Min (Overall):", np.min(arr2D))  # 1
print("Max (Overall):", np.max(arr2D))  # 8

# Find min/max along each axis
print("Min per Column:", np.min(arr2D, axis=0))  # [3, 1, 2]
print("Max per Row:", np.max(arr2D, axis=1))  # [7, 8]

👉 axis=0 → column-wise
👉 axis=1 → row-wise

🔹 Why are These Important in Deep Learning?

1️⃣ Finding the Best Prediction in Classification

In a multi-class classification, we get probabilities for each class. We use argmax to find the class with the highest probability.

softmax_outputs = np.array([[0.1, 0.3, 0.6], 
                            [0.7, 0.2, 0.1]])

predictions = np.argmax(softmax_outputs, axis=1)
print("Predicted Classes:", predictions)  # [2, 0]

✅ argmax helps pick the most probable class.

2️⃣ Finding the Best and Worst Loss Values

When training deep learning models, we track loss values. min and argmin help us find the best (lowest) loss.

loss_values = np.array([0.9, 0.5, 0.2, 0.6])

best_loss = np.min(loss_values)  # Smallest loss
best_epoch = np.argmin(loss_values)  # Epoch with lowest loss

print("Best Loss:", best_loss)  # 0.2
print("Best Epoch:", best_epoch)  # 2

✅ Useful for early stopping and model evaluation.

3️⃣ Finding Maximum Activation in a Neural Network

In a convolutional neural network (CNN), we apply ReLU activation and find the maximum activation in a feature map.

feature_map = np.array([[0.1, 0.5, 0.2], 
                         [0.7, 0.3, 0.9]])

max_activation = np.max(feature_map)  # 0.9
max_location = np.argmax(feature_map)  # 5 (Index in flattened array)

print("Max Activation:", max_activation)
print("Max Activation Index:", max_location)

✅ Helps find the most activated neuron.

Conclusion 🚀