Question

我想为分类问题建立一个神经网络。训练集需要394个输入（共48个维度），并具有100个验证集。代码为：

def build_model():
    model = Sequential()
    model.add(Dense(2, activation = "relu"))
    model.add(Dropout(0.5))
    model.add(Dense(2, activation = "relu"))
    model.add(Dropout(0.2))
    model.add(Dense(2, activation = "sigmoid"))
    model.add(Dropout(0.1))
    model.add(Dense(2, activation = "softmax"))

    model.compile(
        optimizer=tf.keras.optimizers.Adam(learning_rate=0.000001),
        loss=['mean_squared_error'],
        metrics=['accuracy']
    )
    return model


model = build_model()

history = model.fit(
    x_train,
    y_train,
    epochs=5,
    batch_size=32,
    validation_data=(
        x_val,
        y_val
    ),
    callbacks=[ProgbarLogger(count_mode='steps',stateful_metrics=None)
    ]
)

但是我得到了输出奇怪的日志，例如：

训练394个样本，验证100个样本

Epoch 1/5
Epoch 1/5
13/13 [==============================] - 1s 113ms/step - loss: nan - accuracy: 0.9975 - val_loss: nan - val_accuracy: 1.0000
394/394 [==============================] - 1s 4ms/sample - loss: nan - accuracy: 0.9975 - val_loss: nan - val_accuracy: 1.0000
Epoch 2/5
Epoch 2/5
13/13 [==============================] - 0s 5ms/step - loss: nan - accuracy: 1.0000 - val_loss: nan - val_accuracy: 1.0000
394/394 [==============================] - 0s 167us/sample - loss: nan - accuracy: 1.0000 - val_loss: nan - val_accuracy: 1.0000
Epoch 3/5
Epoch 3/5
13/13 [==============================] - 0s 7ms/step - loss: nan - accuracy: 1.0000 - val_loss: nan - val_accuracy: 1.0000
394/394 [==============================] - 0s 217us/sample - loss: nan - accuracy: 1.0000 - val_loss: nan - val_accuracy: 1.0000
Epoch 4/5
Epoch 4/5
13/13 [==============================] - 0s 5ms/step - loss: nan - accuracy: 1.0000 - val_loss: nan - val_accuracy: 1.0000
394/394 [==============================] - 0s 162us/sample - loss: nan - accuracy: 1.0000 - val_loss: nan - val_accuracy: 1.0000
Epoch 5/5
Epoch 5/5
13/13 [==============================] - 0s 6ms/step - loss: nan - accuracy: 1.0000 - val_loss: nan - val_accuracy: 1.0000
394/394 [==============================] - 0s 217us/sample - loss: nan - accuracy: 1.0000 - val_loss: nan - val_accuracy: 1.0000

您能帮我了解这种行为的损失和准确性吗？不应该更低吗？

输出损耗= nan和精度= 1

0 个答案: