Question

这是我第一次创建LSTM RNN。我对生成的“训练损失和验证损失”图的含义感到困惑。这是图形的图像：

以下是训练数据的示例：


                        lat         long    trip_id mode_cat
datetime            id              
2011-08-27 06:13:01 20  39.979973   116.305745  1   1
2011-08-27 06:13:02 20  39.979957   116.305688  1   1
2011-08-27 06:13:03 20  39.979960   116.305693  1   1
2011-08-27 06:13:04 20  39.979970   116.305717  1   1
2011-08-27 06:13:05 20  39.979985   116.305732  1   1

日期时间和ID（user_id）设置为索引。

以下是用于创建移动窗口和LSTM的代码：

def moving_window(dataset_x, dataset_y, past_history):
    data, labels = [], []
    for i in range(past_history, len(dataset_x)):
        indices = range(i-past_history, i)
        data.append(dataset_x[indices])
        labels.append(dataset_y[i])
    return np.array(data), np.array(labels)

past_history = 60

x_train_single, y_train_single = moving_window(dataset_train_x, dataset_train_y, past_history)
x_test_single, y_test_single = moving_window(dataset_test_x, dataset_test_y, past_history)

buffer_size = len(x_train_single)//10
batch_size = 256

train_data_single = tf.data.Dataset.from_tensor_slices((x_train_single, y_train_single))
train_data_single = train_data_single.cache().shuffle(buffer_size).batch(batch_size).repeat()

test_data_single = tf.data.Dataset.from_tensor_slices((x_test_single, y_test_single))
test_data_single = test_data_single.batch(batch_size).repeat()

single_step_model = tf.keras.models.Sequential()
single_step_model.add(tf.keras.layers.LSTM(32,
                                           input_shape=x_train_single.shape[-2:]))
single_step_model.add(tf.keras.layers.Dense(1))

single_step_model.compile(optimizer=tf.keras.optimizers.RMSprop(), loss='mae')

evaluation_interval = len(x_train_single)//batch_size
epochs = 10

single_step_history = single_step_model.fit(train_data_single, epochs=epochs,
                                            steps_per_epoch=evaluation_interval,
                                            validation_data=test_data_single,
                                            validation_steps=50)

Answer 1

如果训练损失与验证损失之间的差距较大，则表明您的模型过拟合；如果训练损失较大，则表明您的模型不拟合。如果您的训练损失和验证损失重叠或彼此接近，则表明您的模型现在适合预测。

模型过拟合

训练和验证损失图的含义

1 个答案: