Python 如何知道我的神经网络模型的准确性？

我已经训练了我的神经网络模型。我想知道这个训练时期我的模型的准确性。我必须得到平均值还是最后一个？ 这是我的输出

25/25 - 12s - loss: 1.3415 - accuracy: 0.3800 - val_loss: 1.0626 - val_accuracy: 0.5000
Epoch 2/20
25/25 - 12s - loss: 1.0254 - accuracy: 0.5000 - val_loss: 1.1129 - val_accuracy: 0.4000
Epoch 3/20
25/25 - 12s - loss: 0.9160 - accuracy: 0.6500 - val_loss: 0.8640 - val_accuracy: 0.7000
Epoch 4/20
25/25 - 12s - loss: 0.8237 - accuracy: 0.6300 - val_loss: 0.8494 - val_accuracy: 0.6000
Epoch 5/20
25/25 - 11s - loss: 0.7411 - accuracy: 0.7320 - val_loss: 0.7320 - val_accuracy: 0.8000
Epoch 6/20
25/25 - 12s - loss: 0.7625 - accuracy: 0.6600 - val_loss: 1.0259 - val_accuracy: 0.6000
Epoch 7/20
25/25 - 12s - loss: 0.8317 - accuracy: 0.6800 - val_loss: 0.5907 - val_accuracy: 0.7500
Epoch 8/20
25/25 - 12s - loss: 0.5557 - accuracy: 0.8100 - val_loss: 0.4630 - val_accuracy: 0.9000
Epoch 9/20
25/25 - 11s - loss: 0.6640 - accuracy: 0.7629 - val_loss: 0.3308 - val_accuracy: 0.9500
Epoch 10/20
25/25 - 12s - loss: 0.5674 - accuracy: 0.8200 - val_loss: 0.5039 - val_accuracy: 0.8000
Epoch 11/20
25/25 - 12s - loss: 0.5566 - accuracy: 0.8200 - val_loss: 0.2161 - val_accuracy: 0.9500
Epoch 12/20
25/25 - 16s - loss: 0.5190 - accuracy: 0.8400 - val_loss: 0.3210 - val_accuracy: 0.8500
Epoch 13/20
25/25 - 12s - loss: 0.5437 - accuracy: 0.7800 - val_loss: 0.7253 - val_accuracy: 0.6500
Epoch 14/20
25/25 - 12s - loss: 0.5035 - accuracy: 0.8300 - val_loss: 0.4291 - val_accuracy: 0.8500
Epoch 15/20
25/25 - 11s - loss: 0.4276 - accuracy: 0.8600 - val_loss: 0.2902 - val_accuracy: 0.8500
Epoch 16/20
25/25 - 11s - loss: 0.4913 - accuracy: 0.8000 - val_loss: 0.3027 - val_accuracy: 0.9000
Epoch 17/20
25/25 - 11s - loss: 0.2931 - accuracy: 0.9100 - val_loss: 0.2718 - val_accuracy: 0.9000
Epoch 18/20
25/25 - 11s - loss: 0.4554 - accuracy: 0.8500 - val_loss: 0.4412 - val_accuracy: 0.8000
Epoch 19/20
25/25 - 11s - loss: 0.3803 - accuracy: 0.8400 - val_loss: 0.2479 - val_accuracy: 1.0000
Epoch 20/20
25/25 - 12s - loss: 0.2692 - accuracy: 0.9200 - val_loss: 0.1805 - val_accuracy: 1.0000
<tensorflow.python.keras.callbacks.History at 0x7f64eec7ada0>```

---

假设您这样训练您的模型：

history = model.fit(...)

您可以通过历史记录访问准确性。历史记录['acc']。其他有用的指标：

损失-损失 val_acc-验证精度 val_损失-验证损失

---

只有当您有验证集时，才显示最后两个历元。

最后一个历元被视为模型精度。因此，根据上述模型，训练精度为92%，验证精度为100%@H_J这意味着过度拟合还是欠拟合？因为验证精度大于训练精度，所以我不会考虑过度拟合。我假设你们在培训和验证方面都有足够的人口。理论上，这将被认为是过分合适的。理论上，因为您的验证集中似乎只有20个样本，太少了，无法判断您是否拟合过度。@LukaszTracewski我使用的是validation_split=0.3，生成数据后，我有1533个训练数据和655个验证数据。人口足够吗？