tf.data：函数尝试转换为 numpy 数组失败？

Question

我正在尝试构建一个tf.data 管道，最终计算skipgrams，但出现错误

NotImplementedError: Cannot convert a symbolic Tensor (cond/Identity:0) to a numpy array.
This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported

我的管道：

text_vector_ds = (
    text_ds
    .batch(1024)
    .map(vectorize_layer)
    .map(my_func)
)

在哪里

text_ds = tf.data.TextLineDataset(file)

vectorize_layer = tensorflow.keras.layers.experimental.preprocessing.TextVectorization(
    standardize='lower_and_strip_punctuation',
    max_tokens=4096,
    output_mode='int',
    output_sequence_length=5)

class MyFunc():
    def _make_fat_diagonal(self, size: int) -> tf.Tensor:

        fat_ones = tf.linalg.band_part(
            tf.ones([size,size], dtype=tf.int64),
            num_lower=self.window,
            num_upper=self.window
        )
    
        return tf.linalg.set_diag(fat_ones, tf.zeros(size, dtype=tf.int64))

    def __call__(self, input):
   
        # Ensure the input is rank 2
        if tf.rank(input) == 1:
            input = tf.expand_dims(input, axis=0)
        input_shape = tf.shape(input)
        num_input_cols = input_shape[1]
     
        return = self._make_fat_diagonal(num_input_cols)

my_func = MyFunc()

部分堆栈跟踪是

../testw2v/skipgram/skipgram.py:333 _make_fat_diagonal  *
    fat_ones = tf.linalg.band_part(
/opt/conda/envs/emb2/lib/python3.7/site-packages/tensorflow/python/util/dispatch.py:201 wrapper  **
    return target(*args, **kwargs)
/opt/conda/envs/emb2/lib/python3.7/site-packages/tensorflow/python/ops/array_ops.py:3120 ones
    output = _constant_if_small(one, shape, dtype, name)
/opt/conda/envs/emb2/lib/python3.7/site-packages/tensorflow/python/ops/array_ops.py:2804 _constant_if_small
    if np.prod(shape) < 1000:
<__array_function__ internals>:6 prod
    
/opt/conda/envs/emb2/lib/python3.7/site-packages/numpy/core/fromnumeric.py:3031 prod
    keepdims=keepdims, initial=initial, where=where)
/opt/conda/envs/emb2/lib/python3.7/site-packages/numpy/core/fromnumeric.py:87 _wrapreduction
    return ufunc.reduce(obj, axis, dtype, out, **passkwargs)
/opt/conda/envs/emb2/lib/python3.7/site-packages/tensorflow/python/framework/ops.py:855 __array__
    " a NumPy call, which is not supported".format(self.name))

NotImplementedError: Cannot convert a symbolic Tensor (cond/Identity:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported

我想它不喜欢提取维度来运行_make_fat_diagonal()，尽管我不确定我会如何表达这一点。在管道之外，该函数在text_ds 数据集的各个元素上工作得很好。如您所见，我很小心只使用 Tensorflow 方法。

正确的做法是什么？

Answer 1

截至今天，这似乎是由错误 https://github.com/tensorflow/models/issues/9706 造成的。恢复到 python 3.6 使其按预期工作。