如何解决Keras中训练准确率起始Epoch为0、绘图Epoch轴从1开始显示的问题

Let's work through resolving both of your issues clearly:

Why This Happens #

Keras uses 0-based indexing for storing training history entries. When you train for 20 epochs, the hist.history lists (like accuracy and val_accuracy) will have 20 elements indexed from 0 to 19—these correspond directly to the 1st through 20th training epochs. This is why your plot shows an x-range of 0-19 instead of the intuitive 1-20.

Step-by-Step Fix #

1. Adjust the Plot to Use 1-Based Epoch Numbers #

Instead of relying on matplotlib's default 0-based x-values, explicitly create a list of epoch numbers starting at 1. Here's your modified code:

hist = model.fit(train_generator, validation_data=val_generator, epochs=20, batch_size=batchsize)

plt.figure(figsize=(8,8))
plt.subplot(2,1,1)

# Generate 1-based epoch sequence (1 to 20)
epochs = range(1, len(hist.history['accuracy']) + 1)

# Plot using the 1-based epoch values as the x-axis
plt.plot(epochs, hist.history['accuracy'], marker='o')
plt.plot(epochs, hist.history['val_accuracy'], marker='p')

plt.axis(ymin=0.0, ymax=1)
plt.grid()
plt.title('VGG16 Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epochs')
# Optional: Force x-ticks to show every epoch number for clarity
plt.xticks(epochs)
plt.legend(['Training Accuracy', 'Validation Accuracy'])
plt.show()

2. Display 1-Based Epochs in Accuracy Records (If Needed) #

If you want to print or log accuracy values with 1-based epoch labels (instead of 0-based), loop through the history with an offset:

for epoch_num, acc, val_acc in zip(epochs, hist.history['accuracy'], hist.history['val_accuracy']):
    print(f"Epoch {epoch_num}: Training Accuracy = {acc:.4f}, Validation Accuracy = {val_acc:.4f}")

Key Changes Explained #

• epochs = range(1, len(hist.history['accuracy']) + 1): Creates a sequence from 1 to your total number of epochs (20), aligning with how we naturally count training cycles.
• Plotting with epochs as the x-axis: Replaces the default 0-based indices with human-readable 1-based epoch numbers, fixing your plot's x-range issue.
• plt.xticks(epochs): Optional but ensures every epoch number is visible on the x-axis, making the plot easier to interpret.

This solution will resolve both problems: your plot will now show an x-range of 1 to 20, and any printed accuracy records can be labeled with intuitive 1-based epoch numbers.

内容的提问来源于stack exchange，提问作者Monika Roopak