时间戳

首页 > 解决方案 > 收到的标签值 17 超出 [0, 12) 的有效范围 - Keras Python

python - 收到的标签值 17 超出 [0, 12) 的有效范围 - Keras Python

问题描述

Keras 对我来说很陌生。我正在尝试将一些编程付诸实践。

数据形状如下：

Train shape X: (249951, 5, 52)  y  (249951,)
Test shape X: (263343, 5, 52)  y  (263343,)  # Do not confuse with the distribution, it is juts toy example

我的日期包含十二个标签。keras CNN架构如下：

def get_compiled_model():
    # Make a simple 2-layer densely-connected neural network.
    inputs = keras.Input(shape=(260,))
    x = keras.layers.Dense(256, activation="relu")(inputs)
    x = keras.layers.Dense(256, activation="relu")(x)
    outputs = keras.layers.Dense(12)(x)    # 12  classes
    model = keras.Model(inputs, outputs)
    model.compile(
        optimizer=keras.optimizers.Adam(),
        loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
        metrics=[keras.metrics.SparseCategoricalAccuracy()],
    )
    return model

现在，我将 12 个神经元馈送到输出层，因为我的数据包含 12 个类。但是，会显示以下错误消息：

    Use `tf.data.Iterator.get_next_as_optional()` instead.
2255/7811 [=======>......................] - ETA: 18s - loss: 0.1109 - sparse_categorical_accuracy: 0.99452021-04-19 16:32:33.591493: W tensorflow/core/framework/op_kernel.cc:1767] OP_REQUIRES failed at sparse_xent_op.cc:90 : Invalid argument: Received a label value of 17 which is outside the valid range of [0, 12).  Label values: 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 17 17 17 17 17
Traceback (most recent call last):
  File "C:\Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\contextlib.py", line 131, in __exit__
    self.gen.throw(type, value, traceback)
  File "C:\Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\ops\variable_scope.py", line 2804, in variable_creator_scope
    yield
  File "C:\Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\keras\engine\training.py", line 1098, in fit
    tmp_logs = train_function(iterator)
  File "C:\Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\def_function.py", line 780, in __call__
    result = self._call(*args, **kwds)
  File "C:\Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\def_function.py", line 807, in _call
    return self._stateless_fn(*args, **kwds)  # pylint: disable=not-callable
  File "C:\Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\function.py", line 2829, in __call__
    return graph_function._filtered_call(args, kwargs)  # pylint: disable=protected-access
  File "C:\Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\function.py", line 1843, in _filtered_call
    return self._call_flat(
  File "C:\Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\function.py", line 1923, in _call_flat
    return self._build_call_outputs(self._inference_function.call(
  File "C:\Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\function.py", line 545, in call
    outputs = execute.execute(
  File "C:\Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\site-packages\tensorflow\python\eager\execute.py", line 59, in quick_execute
    tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
tensorflow.python.framework.errors_impl.InvalidArgumentError:  Received a label value of 17 which is outside the valid range of [0, 12).  Label values: 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 17 17 17 17 17
     [[node sparse_categorical_crossentropy/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits (defined at Users\Nafees Ahmed\AppData\Local\Programs\Python\Python38\lib\threading.py:932) ]] [Op:__inference_train_function_846]
Function call stack:
train_function

主要错误： tensorflow.python.framework.errors_impl.InvalidArgumentError：收到的标签值 17 超出了 [0, 12) 的有效范围。标签值： 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 17 17 17 17 17

标签： pythontensorflowkeras

解决方案

如果您的标签（y）是数字，它将不起作用。您需要将其转换为二进制数据，因为它是多类二进制问题。也许你可以使用下面的来做到这一点。

y_train = np_utils.to_categorical(y_train, num_classes)
y_test = np_utils.to_categorical(y_test, num_classes)

num_classes是 12，在你的情况下。但是，对我来说，12 节课听起来太多了。这真的是多类而不是多标签吗？

推荐阅读