我收到Tensorflow错误“ValueError: None values not supported”。尝试填充tf数据集时。有人能帮帮我吗？

Question

我已经将我的数据存储在秩2 (N X 3)张量中，我正在尝试收集该张量的特定行，以生成单个示例的数据，并将其提供给Keras训练循环。基本上，这个想法很简单如下:索引Ib (比方说100)和Ie (比方说207)对于每个示例都是指定我的输入数据是来自行76，77，78，...，100,105,110,115，... 205,207,208,209，...的300x3张量。(直到每个示例总共300个索引)的数据张量。注意Ib和Ie之间的步幅为5。虽然附加的代码在手动获取数据时运行良好(如代码示例所示)，但当Tensorflow稍后尝试执行相同的操作时会发生一些情况。此外，我知道如果我在Ib之前和之后选择固定数量的行数，下面的代码就可以正常工作。当使用符号张量Ib (?)调用时，行updates = tf.range(start=Ib, limit=Ie, delta=5, dtype=tf.int32)似乎不能产生任何合理的输出，从而导致代码在下一行中断。我使用的是Python 3.7 / Tensorflow 2.3.0，附带的代码会产生输出

2020-08-18 10:25:51.384252: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN)to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2020-08-18 10:25:51.394210: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x7ff732d46110 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-08-18 10:25:51.394226: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version

trainINDX =   0, X.shape = [ 300, 3 ]
trainINDX =   1, X.shape = [ 300, 3 ]
trainINDX =   2, X.shape = [ 300, 3 ]
trainINDX =   3, X.shape = [ 300, 3 ]
trainINDX =   4, X.shape = [ 300, 3 ]
trainINDX =   5, X.shape = [ 300, 3 ]
trainINDX =   6, X.shape = [ 300, 3 ]
trainINDX =   7, X.shape = [ 300, 3 ]
trainINDX =   8, X.shape = [ 300, 3 ]
trainINDX =   9, X.shape = [ 300, 3 ]

Traceback (most recent call last):
  File "testData.py", line 93, in <module>
    main( )
  File "testData.py", line 80, in main
    trainData = trainData.map(fetchData, num_parallel_calls=tf.data.experimental.AUTOTUNE)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/data/ops/dataset_ops.py", line 1702, in map
    preserve_cardinality=True)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/data/ops/dataset_ops.py", line 4084, in __init__
    use_legacy_function=use_legacy_function)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/data/ops/dataset_ops.py", line 3371, in __init__
    self._function = wrapper_fn.get_concrete_function()
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/function.py", line 2939, in get_concrete_function
    *args, **kwargs)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/function.py", line 2906, in _get_concrete_function_garbage_collected
    graph_function, args, kwargs = self._maybe_define_function(args, kwargs)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/function.py", line 3213, in _maybe_define_function
    graph_function = self._create_graph_function(args, kwargs)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/function.py", line 3075, in _create_graph_function
    capture_by_value=self._capture_by_value),
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/framework/func_graph.py", line 986, in func_graph_from_py_func
    func_outputs = python_func(*func_args, **func_kwargs)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/data/ops/dataset_ops.py", line 3364, in wrapper_fn
    ret = _wrapper_helper(*args)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/data/ops/dataset_ops.py", line 3299, in _wrapper_helper
    ret = autograph.tf_convert(func, ag_ctx)(*nested_args)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/autograph/impl/api.py", line 258, in wrapper
    raise e.ag_error_metadata.to_exception(e)
ValueError: in user code:

    testData.py:57 fetchData  *
        indices    = tf.range(start=sampleCount, limit=sampleCount+updates.shape[0], dtype=tf.int32)
    /Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/ops/variables.py:1074 _run_op
        return tensor_oper(a.value(), *args, **kwargs)
    /Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/ops/math_ops.py:1125 binary_op_wrapper
        return func(x, y, name=name)
    /Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/util/dispatch.py:201 wrapper
        return target(*args, **kwargs)
    /Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/ops/math_ops.py:1443 _add_dispatch
        y = ops.convert_to_tensor(y, dtype_hint=x.dtype.base_dtype, name="y")
    /Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/framework/ops.py:1499 convert_to_tensor
        ret = conversion_func(value, dtype=dtype, name=name, as_ref=as_ref)
    /Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/framework/constant_op.py:338 _constant_tensor_conversion_function
        return constant(v, dtype=dtype, name=name)
    /Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/framework/constant_op.py:264 constant
        allow_broadcast=True)
    /Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/framework/constant_op.py:282 _constant_impl
        allow_broadcast=allow_broadcast))
    /Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/framework/tensor_util.py:444 make_tensor_proto
        raise ValueError("None values not supported.")

    ValueError: None values not supported.

import numpy as np
import tensorflow as tf

def main( ):

    inputData      = np.zeros((1000, 3), dtype=np.float32)
    inputData[:,0] = np.sin(np.arange(1000)/360)
    inputData[:,1] = np.cos(np.arange(1000)/360)
    inputData[:,2] = np.sin(np.arange(1000)/360) * np.cos(np.arange(1000)/360) # Generate some input data
    inputData      = tf.convert_to_tensor(inputData, dtype=tf.float32)

    outputData     = np.random.randint(low=0, high=3, size=1000, dtype=np.int32) # Generate random output data

    eventData      = np.zeros((1000,2), dtype=np.int32)
    eventData[:,0] = np.arange(1000)                                                                  # Begin index of sparse sampling
    eventData[:,1] = np.arange(1000) + np.random.randint(low=80, high=121, size=1000, dtype=np.int32) # End   index of sparse sampling
    eventData      = tf.convert_to_tensor(eventData, dtype=tf.int32)

    totalSampleCount = int(1000)
    eventCount       = int(1000)
    inputDim         = int(300)
    outputDim        = int(3)
    batchSize        = int(256)
    epochCount       = int(5)
    stepsPerEpoch    = np.floor(666/batchSize)

    trainINDX      = np.arange(666)
    validationINDX = np.arange(666,1000)

    model = tf.keras.Sequential([
        tf.keras.layers.Conv1D(filters=15, kernel_size=15, strides=1, padding='same', dilation_rate=1, activation='relu', use_bias=True, kernel_initializer='glorot_uniform', bias_initializer='zeros', input_shape=(inputDim,3)),
        tf.keras.layers.MaxPooling1D(pool_size=5, strides=None, padding='valid'),
        tf.keras.layers.Flatten(),
        tf.keras.layers.Dense(256, activation='relu', use_bias=True),
        tf.keras.layers.Dropout(0.50),
        tf.keras.layers.Dense(outputDim, activation='softmax')
    ])

    model.compile( optimizer=tf.keras.optimizers.Adam(learning_rate=1e-3, beta_1=0.9, beta_2=0.999, amsgrad=True),
                   loss=tf.keras.losses.SparseCategoricalCrossentropy(),
                   metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])

    def fetchData(eventINDX, y): # This function picks up the requested elements from inputData

        sampleINDX = tf.zeros([inputDim], dtype=tf.int32) # Initialize sampleINDX to zero

        Ib = tf.gather_nd(eventData, [tf.cast(eventINDX, dtype=tf.int32), 0]) # Begin of sparse sampling
        Ie = tf.gather_nd(eventData, [tf.cast(eventINDX, dtype=tf.int32), 1]) # End   of sparse sampling

        indices    = tf.range(start=0, limit=24, dtype=tf.int32)
        updates    = tf.range(start=Ib-24, limit=Ib, dtype=tf.int32)
        sampleINDX = tf.tensor_scatter_nd_update(sampleINDX, tf.expand_dims(indices, axis=1), updates)

        sampleCount = tf.Variable(24, dtype=tf.int32)

        updates    = tf.range(start=Ib, limit=Ie, delta=5, dtype=tf.int32)
        indices    = tf.range(start=sampleCount, limit=sampleCount+updates.shape[0], dtype=tf.int32)
        sampleINDX = tf.tensor_scatter_nd_update(sampleINDX, tf.expand_dims(indices, axis=1), updates)

        sampleCount.assign_add(updates.shape[0])
           
        remainingSampleCount = tf.math.subtract(tf.constant(inputDim, dtype=tf.int32), sampleCount)

        indices    = tf.range(start=sampleCount, limit=inputDim, dtype=tf.int32)
        updates    = tf.range(start=Ie, limit=Ie+remainingSampleCount, dtype=tf.int32)
        sampleINDX = tf.tensor_scatter_nd_update(sampleINDX, tf.expand_dims(indices, axis=1), updates)

        X = tf.gather(inputData, tf.math.floormod(sampleINDX, totalSampleCount), axis=0)

        return X, y

    print('')
    for i in range(10):
        X, y = fetchData(i,outputData[i])
        print('trainINDX = %3d, X.shape = [ %3d, %d ]' % (i, X.shape[0], X.shape[1]))
    print('')

    trainData = tf.data.Dataset.from_tensor_slices((trainINDX, outputData[trainINDX]))
    trainData = trainData.shuffle(buffer_size=trainINDX.size, reshuffle_each_iteration=True) 
    trainData = trainData.map(fetchData, num_parallel_calls=tf.data.experimental.AUTOTUNE)
    trainData = trainData.repeat()
    trainData = trainData.batch(batchSize, drop_remainder=True)
    trainData = trainData.prefetch(buffer_size=tf.data.experimental.AUTOTUNE)

    validationData = tf.data.Dataset.from_tensor_slices((validationINDX, outputData[validationINDX]))
    validationData = validationData.map(fetchData, num_parallel_calls=tf.data.experimental.AUTOTUNE)
    validationData = validationData.batch(batchSize, drop_remainder=False)
    validationData = validationData.prefetch(buffer_size=tf.data.experimental.AUTOTUNE)

    history = model.fit(x=trainData, steps_per_epoch=stepsPerEpoch, validation_data=validationData, verbose=1, epochs=epochCount)

if __name__== "__main__":
  main( )

任何帮助解决上述问题的人都将不胜感激。提前谢谢你！

Answer 1

看起来用“tf.shape(更新)”替换"updates.shape“可以解决这个问题。然而，这导致了下面描述的另一个问题。

import numpy as np
import tensorflow as tf

def main( ):

    inputData      = np.zeros((1000, 3), dtype=np.float32)
    inputData[:,0] = np.sin(np.arange(1000)/360)
    inputData[:,1] = np.cos(np.arange(1000)/360)
    inputData[:,2] = np.sin(np.arange(1000)/360) * np.cos(np.arange(1000)/360) # Generate some input data
    inputData      = tf.convert_to_tensor(inputData, dtype=tf.float32)

    outputData     = np.random.randint(low=0, high=3, size=1000, dtype=np.int32) # Generate random output data

    eventData      = np.zeros((1000,2), dtype=np.int32)
    eventData[:,0] = np.arange(1000)                                                                  # Begin index of sparse sampling
    eventData[:,1] = np.arange(1000) + np.random.randint(low=80, high=121, size=1000, dtype=np.int32) # End   index of sparse sampling
    eventData      = tf.convert_to_tensor(eventData, dtype=tf.int32)

    totalSampleCount = int(1000)
    eventCount       = int(1000)
    inputDim         = int(300)
    outputDim        = int(3)
    batchSize        = int(256)
    epochCount       = int(5)
    stepsPerEpoch    = int(np.floor(666/batchSize))

    trainINDX      = np.arange(666)
    validationINDX = np.arange(666,1000)

    model = tf.keras.Sequential([
        tf.keras.layers.Conv1D(filters=15, kernel_size=15, strides=1, padding='same', dilation_rate=1, activation='relu', use_bias=True, kernel_initializer='glorot_uniform', bias_initializer='zeros', input_shape=(inputDim,3)),
        tf.keras.layers.MaxPooling1D(pool_size=5, strides=None, padding='valid'),
        tf.keras.layers.Flatten(),
        tf.keras.layers.Dense(256, activation='relu', use_bias=True),
        tf.keras.layers.Dropout(0.50),
        tf.keras.layers.Dense(outputDim, activation='softmax')
    ])

    model.compile( optimizer=tf.keras.optimizers.Adam(learning_rate=1e-3, beta_1=0.9, beta_2=0.999, amsgrad=True),
                   loss=tf.keras.losses.SparseCategoricalCrossentropy(),
                   metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])

    def fetchData(eventINDX, y): # This function picks up the requested elements from inputData

        sampleINDX = tf.zeros([inputDim], dtype=tf.int32) # Initialize sampleINDX to zero

        Ib = tf.gather_nd(eventData, [tf.cast(eventINDX, dtype=tf.int32), 0]) # Begin of sparse sampling
        Ie = tf.gather_nd(eventData, [tf.cast(eventINDX, dtype=tf.int32), 1]) # End   of sparse sampling

        indices    = tf.range(start=0, limit=24, dtype=tf.int32)
        updates    = tf.range(start=Ib-24, limit=Ib, dtype=tf.int32)
        sampleINDX = tf.tensor_scatter_nd_update(sampleINDX, tf.expand_dims(indices, axis=1), updates)

        sampleCount = tf.Variable(24, dtype=tf.int32)

        updates    = tf.range(start=Ib, limit=Ie, delta=5, dtype=tf.int32)
        indices    = tf.range(start=sampleCount, limit=sampleCount+tf.shape(updates)[0], dtype=tf.int32)
        sampleINDX = tf.tensor_scatter_nd_update(sampleINDX, tf.expand_dims(indices, axis=1), updates)

        sampleCount.assign_add(tf.shape(updates)[0])
           
        remainingSampleCount = tf.math.subtract(tf.constant(inputDim, dtype=tf.int32), sampleCount)

        indices    = tf.range(start=sampleCount, limit=inputDim, dtype=tf.int32)
        updates    = tf.range(start=Ie, limit=Ie+remainingSampleCount, dtype=tf.int32)
        sampleINDX = tf.tensor_scatter_nd_update(sampleINDX, tf.expand_dims(indices, axis=1), updates)

        X = tf.gather(inputData, tf.math.floormod(sampleINDX, totalSampleCount), axis=0)

        return X, y

    print('')
    for i in range(10):
        X, y = fetchData(i,outputData[i])
        print('trainINDX = %3d, X.shape = [ %3d, %d ]' % (i, X.shape[0], X.shape[1]))
    print('')

    trainData = tf.data.Dataset.from_tensor_slices((trainINDX, outputData[trainINDX]))
    trainData = trainData.shuffle(buffer_size=trainINDX.size, reshuffle_each_iteration=True) 
    trainData = trainData.map(fetchData, num_parallel_calls=tf.data.experimental.AUTOTUNE)
    trainData = trainData.repeat()
    trainData = trainData.batch(batchSize, drop_remainder=True)
    trainData = trainData.prefetch(buffer_size=tf.data.experimental.AUTOTUNE)

    validationData = tf.data.Dataset.from_tensor_slices((validationINDX, outputData[validationINDX]))
    validationData = validationData.map(fetchData, num_parallel_calls=tf.data.experimental.AUTOTUNE)
    validationData = validationData.batch(batchSize, drop_remainder=False)
    validationData = validationData.prefetch(buffer_size=tf.data.experimental.AUTOTUNE)

    history = model.fit(x=trainData, steps_per_epoch=stepsPerEpoch, validation_data=validationData, verbose=1, epochs=epochCount)

if __name__== "__main__":
  main( )

现在附加文件的输出是

trainINDX =   0, X.shape = [ 300, 3 ]
trainINDX =   1, X.shape = [ 300, 3 ]
trainINDX =   2, X.shape = [ 300, 3 ]
trainINDX =   3, X.shape = [ 300, 3 ]
trainINDX =   4, X.shape = [ 300, 3 ]
trainINDX =   5, X.shape = [ 300, 3 ]
trainINDX =   6, X.shape = [ 300, 3 ]
trainINDX =   7, X.shape = [ 300, 3 ]
trainINDX =   8, X.shape = [ 300, 3 ]
trainINDX =   9, X.shape = [ 300, 3 ]

Epoch 1/5
Traceback (most recent call last):
  File "testData.py", line 93, in <module>
    main( )
  File "testData.py", line 90, in main
    history = model.fit(x=trainData, steps_per_epoch=stepsPerEpoch, validation_data=validationData, verbose=1, epochs=epochCount)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/keras/engine/training.py", line 108, in _method_wrapper
    return method(self, *args, **kwargs)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/keras/engine/training.py", line 1098, in fit
    tmp_logs = train_function(iterator)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/def_function.py", line 780, in __call__
    result = self._call(*args, **kwds)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/def_function.py", line 840, in _call
    return self._stateless_fn(*args, **kwds)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/function.py", line 2829, in __call__
    return graph_function._filtered_call(args, kwargs)  # pylint: disable=protected-access
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/function.py", line 1848, in _filtered_call
    cancellation_manager=cancellation_manager)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/function.py", line 1924, in _call_flat
    ctx, args, cancellation_manager=cancellation_manager))
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/function.py", line 550, in call
    ctx=ctx)
  File "/Users/relaxation82/Library/Python/3.7/lib/python/site-packages/tensorflow/python/eager/execute.py", line 60, in quick_execute
    inputs, attrs, num_outputs)
tensorflow.python.framework.errors_impl.FailedPreconditionError:  Error while reading resource variable _AnonymousVar24 from Container: localhost. This could mean that the variable was uninitialized. Not found: Resource localhost/_AnonymousVar24/N10tensorflow3VarE does not exist.
     [[{{node range_3/ReadVariableOp}}]]
     [[IteratorGetNext]] [Op:__inference_train_function_1569]

Function call stack:
train_function