如何从Python分类器获取各类别概率并验证准确率？

Got it, let's walk through this step by step with a practical example using scikit-learn—this approach works for any number of classes (including your 9-class use case). I’ll use a synthetic dataset to make things concrete, but you can swap it out with your own data easily.

Step 1: Setup and Train a Classifier #

First, we’ll train a classification model that supports outputting class probabilities (most scikit-learn classifiers do, like LogisticRegression, RandomForestClassifier, or GradientBoostingClassifier).

import numpy as np
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score

# Create a synthetic 9-class dataset (matches your use case)
X, y = make_classification(
    n_samples=1000, n_features=20, n_informative=15,
    n_classes=9, random_state=42
)

# Split into training and test sets
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)

# Train a classifier (Logistic Regression here, use your model of choice)
model = LogisticRegression(multi_class="multinomial", max_iter=1000, random_state=42)
model.fit(X_train, y_train)

Step 2: Get Class Probabilities for Each Sample #

Use the predict_proba() method to get the probability distribution across all classes for each test sample. The output is a 2D array where each row corresponds to a sample, and each column is the probability of that sample belonging to a specific class.

# Get class probabilities for the test set
class_probs = model.predict_proba(X_test)

# Check the shape: (number of test samples, number of classes)
print(f"Probabilities shape: {class_probs.shape}")  # Should be (200, 9) for our synthetic data

# Example: Print probabilities for the first 2 samples
print("\nFirst sample probabilities:")
print(class_probs[0])
print("\nSecond sample probabilities:")
print(class_probs[1])

Step 3: Predict Class Labels from Probabilities #

To get the predicted class label, we just take the index of the maximum probability in each row. You can do this manually with np.argmax(), or use the model’s built-in predict() method (which does this exact logic under the hood).

# Manual prediction from probabilities
predicted_labels = np.argmax(class_probs, axis=1)

# Compare with model's predict() method (should be identical)
model_predicted_labels = model.predict(X_test)
print("\nAre manual predictions same as model's predict()?", np.array_equal(predicted_labels, model_predicted_labels))

Step 4: Calculate Classification Accuracy #

Now we can compare our predicted labels with the true test labels to compute the classification accuracy.

# Calculate accuracy
accuracy = accuracy_score(y_test, predicted_labels)
print(f"\nClassification Accuracy: {accuracy:.4f}")

Quick Notes #

• Not all classifiers support predict_proba() out of the box—for example, SVC in scikit-learn requires setting probability=True during initialization to enable this feature.
• The order of classes in class_probs matches the order of model.class_ (you can run print(model.class_) to see which index maps to which class label).
• For multi-class problems, make sure your classifier is configured correctly (like using multi_class="multinomial" for Logistic Regression).

内容的提问来源于stack exchange，提问作者Vidya Marathe