当我 运行 model.predict(X) 时,我得到 ndim 的 ValueError

I get ValueError with the ndim when I run model.predict(X)

我使用此代码在我的数据上训练我的模型

tf.keras.backend.clear_session()
tf.random.set_seed(50)
np.random.seed(50)
train_set = windowed_dataset(x_train, window_size=30, batch_size=15, shuffle_buffer=shuffle_buffer_size)
model = tf.keras.models.Sequential([
  tf.keras.layers.Conv1D(filters=100, kernel_size=5,
                      strides=1, padding="causal",
                      activation="relu",
                      input_shape=[None, 1]),
  tf.keras.layers.LSTM(100, return_sequences=True),
  tf.keras.layers.LSTM(100, return_sequences=True),
  #tf.keras.layers.Dense(30, activation="relu"),
  #tf.keras.layers.Dense(30, activation="relu"),
  tf.keras.layers.Dense(1),
  tf.keras.layers.Lambda(lambda x: x * 400)
])


optimizer = tf.keras.optimizers.Adam(
    learning_rate=0.00001, beta_1=0.9, beta_2=0.999, epsilon=1e-07, amsgrad=True,
    name='Adam'
)
model.compile(loss=tf.keras.losses.Huber(),
              optimizer=optimizer,
              metrics=["mae"])
history = model.fit(train_set,epochs=100)

这里是 model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv1d (Conv1D)              (None, 30, 100)           600       
_________________________________________________________________
lstm (LSTM)                  (None, 30, 100)           80400     
_________________________________________________________________
lstm_1 (LSTM)                (None, 30, 100)           80400     
_________________________________________________________________
dense (Dense)                (None, 30, 1)             101       
_________________________________________________________________
lambda (Lambda)              (None, 30, 1)             0         
=================================================================
Total params: 161,501
Trainable params: 161,501
Non-trainable params: 0
_________________________________________________________________
None

我正在尝试运行这个代码

model.predict(
    x_valid, batch_size=None, verbose=0, steps=None, callbacks=None, max_queue_size=10,
    workers=1, use_multiprocessing=False
)

并返回此错误消息:

ValueError: Input 0 of layer sequential is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: [None, 1]

我尝试使用此函数 np.array(x_valid).reshape(300,1) 重塑 x_valid,但没有成功。

我把ndim扩大三倍解决了这个问题

    test_input = x_valid[425]
    test_input = np.expand_dims(test_input,axis=0)
    test_input = np.expand_dims(test_input,axis=0)
    test_input = np.expand_dims(test_input,axis=0)

    print(model.predict(test_input))
    # OUTPUT [[[71.46894]]]

您的问题来自于您需要添加 batch_dimension 才能预测一个数据点。

这在处理TensorFlow和Keras时是必要的,即使你预测一个单一的样本,你也需要添加batch_size of 1.

您需要做的是:

  1. 从您的测试集中获取一项(例如,test_input = x_valid[0]
  2. 构造1的batch_size,即test_input = np.expand_dims(test_input,axis=0)
  3. 现在用模型预测,即prediction = model.predict(test_input)

问题来自不正确的测试数据维度。 x_input 的形状为 (15,30,1),因此由此得出测试数据也必须具有 3 维形状(例如 [1,1,1])。在您的代码中,测试数据是一个 1-dim 数组,因此您应该使用 'test_input = np.expand_dims(test_input,axis=0)'

扩展 dims TWICE 以达到 3-dim 数组