感知机绘制决策边界报错求助：特征维度不匹配与值解包不足

Hey there! Let's break down why you're hitting those errors and get your decision boundary plotted properly.

What's Causing the Errors? #

The core issue is simple: your trained perceptron is built to work with 13 features, but when you tried to plot the boundary, you only passed 2 features (the X1 and X2 from your meshgrid) to perceptron.predict(). The model throws a fit because it expects 13 inputs per sample, not 2. Your attempt to unpack 12 variables was off-base too—meshgrid only generates two 2D arrays, so that's why you got the second "not enough values" error.

Decision boundaries can only be visualized in a 2D plane, so we need to narrow our focus to two features. Below are two valid solutions depending on what you want to inspect.

Solution 1: Train a Perceptron on a Pair of Features #

If you want to see if a specific pair of features can linearly separate your data, retrain the perceptron using only those two features. Here's the adjusted code:

from matplotlib.colors import ListedColormap
import matplotlib.pyplot as plt
import numpy as np
from sklearn.linear_model import Perceptron

# Choose two features to visualize (e.g., feature 0 and feature 1)
feature_indices = [0, 1]
X_two_features = X[:, feature_indices]

# Retrain the perceptron on just these two features
perceptron_two_feat = Perceptron(random_state=42)
perceptron_two_feat.fit(X_two_features, Y)

# Plot the decision boundary
plt.clf()
X_set, y_set = X_two_features, Y
X1, X2 = np.meshgrid(
    np.arange(start=X_set[:, 0].min() - 1, stop=X_set[:, 0].max() + 1, step=0.01),
    np.arange(start=X_set[:, 1].min() - 1, stop=X_set[:, 1].max() + 1, step=0.01)
)

# Predict using the two-feature perceptron
plt.contourf(
    X1, X2,
    perceptron_two_feat.predict(np.array([X1.ravel(), X2.ravel()]).T).reshape(X1.shape),
    alpha=0.75, cmap=ListedColormap(('navajowhite', 'darkkhaki'))
)

# Plot data points
plt.xlim(X1.min(), X1.max())
plt.ylim(X2.min(), X2.max())
for i, j in enumerate(np.unique(y_set)):
    plt.scatter(
        X_set[y_set == j, 0], X_set[y_set == j, 1],
        c=ListedColormap(('red', 'green'))(i), label=j
    )

plt.title(f'Perceptron Decision Boundary (Feature {feature_indices[0]} vs Feature {feature_indices[1]})')
plt.xlabel(f'Feature {feature_indices[0]}')
plt.ylabel(f'Feature {feature_indices[1]}')
plt.legend()
plt.show()

Solution 2: Fix Other Features to Their Mean (Use Full Model) #

If you want to visualize how two features interact within your original 13-feature model, fix the other 11 features to their mean values and only vary the two you want to plot. This keeps your original trained perceptron intact:

from matplotlib.colors import ListedColormap
import matplotlib.pyplot as plt
import numpy as np

# Choose two features to visualize (e.g., feature 2 and feature 5)
feat1_idx, feat2_idx = 2, 5

# Calculate mean of all features to fix the other 11 features
X_feature_means = X.mean(axis=0)

plt.clf()
X_set, y_set = X, Y
X1, X2 = np.meshgrid(
    np.arange(start=X_set[:, feat1_idx].min() - 1, stop=X_set[:, feat1_idx].max() + 1, step=0.01),
    np.arange(start=X_set[:, feat2_idx].min() - 1, stop=X_set[:, feat2_idx].max() + 1, step=0.01)
)

# Create grid points with fixed values for non-visualized features
grid_points = np.zeros((X1.ravel().shape[0], X.shape[1]))
grid_points[:, feat1_idx] = X1.ravel()
grid_points[:, feat2_idx] = X2.ravel()
# Fill other features with their mean values
for idx in range(X.shape[1]):
    if idx != feat1_idx and idx != feat2_idx:
        grid_points[:, idx] = X_feature_means[idx]

# Predict using your original 13-feature perceptron
plt.contourf(
    X1, X2,
    perceptron.predict(grid_points).reshape(X1.shape),
    alpha=0.75, cmap=ListedColormap(('navajowhite', 'darkkhaki'))
)

# Plot data points for the two features
plt.xlim(X1.min(), X1.max())
plt.ylim(X2.min(), X2.max())
for i, j in enumerate(np.unique(y_set)):
    plt.scatter(
        X_set[y_set == j, feat1_idx], X_set[y_set == j, feat2_idx],
        c=ListedColormap(('red', 'green'))(i), label=j
    )

plt.title(f'Perceptron Decision Boundary (Feature {feat1_idx} vs Feature {feat2_idx})')
plt.xlabel(f'Feature {feat1_idx}')
plt.ylabel(f'Feature {feat2_idx}')
plt.legend()
plt.show()

Key Notes #

• Swap out the feature_indices or feat1_idx/feat2_idx values to explore different pairs of features.
• Solution 1 tells you if a pair of features alone can linearly separate your data; Solution 2 shows how two features behave within your full model.

内容的提问来源于stack exchange，提问作者gendry