多分类结果映射0/1及CIFAR数据集预测标签类别判定

Hey there! Let's break down your questions step by step since you're working on CIFAR image classification with those 10-class probability outputs. I'll cover both identifying the predicted class and converting the probabilities to a 0-1 one-hot mapping.

Your output is an array where each element represents the model's predicted probability for one of the 10 CIFAR classes (indexed 0 to 9). To find which class the model is most confident in, you just need to find the index of the maximum value in the array.

Example with NumPy #

Since you're working with NumPy arrays (as shown in your example), here's how to do it:

import numpy as np

# Your sample prediction probability array
pred_probs = np.array([3.3655483e-04, 9.4402254e-01, 1.1646092e-03, 2.8560971e-04, 
                       1.4086446e-04, 7.1564602e-05, 2.4985364e-03, 6.5030693e-04, 
                       3.4783698e-05, 5.0794542e-02], dtype=np.float32)

# Get the index of the highest probability (this is your predicted class)
pred_class_idx = np.argmax(pred_probs)
print(f"Predicted class index: {pred_class_idx}")

# Optional: Map the index to CIFAR-10's human-readable class names
cifar10_classes = ["airplane", "automobile", "bird", "cat", "deer", 
                   "dog", "frog", "horse", "ship", "truck"]
print(f"Predicted class name: {cifar10_classes[pred_class_idx]}")

This will output Predicted class index: 1 and Predicted class name: automobile for your sample array, since the 2nd element (index 1) has the highest probability (~94.4%).

The method you mentioned is exactly right: set the position of the maximum probability to 1, and all others to 0. Here's how to implement this cleanly:

Using NumPy #

# Initialize a zero array with the same shape as your prediction probabilities
one_hot = np.zeros_like(pred_probs)

# Set the index of the highest probability to 1
one_hot[pred_class_idx] = 1

print("0-1 mapping result:")
print(one_hot)

For your sample input, this will output:

array([0., 1., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)

Bonus: Using PyTorch (if you're using a PyTorch model) #

If you're working with PyTorch tensors instead of NumPy arrays, the process is similar:

import torch

# Convert your NumPy array to a PyTorch tensor
pred_probs_tensor = torch.tensor(pred_probs)

# Get the predicted class index
pred_class_idx_tensor = torch.argmax(pred_probs_tensor)

# Create the one-hot tensor
one_hot_tensor = torch.zeros(10)
one_hot_tensor[pred_class_idx_tensor] = 1

# Convert back to NumPy if needed
print(one_hot_tensor.numpy())

内容的提问来源于stack exchange，提问作者amaresh hiremani