InvalidArgumentError: indices[1] = [0,1247] is out of order. Many sparse ops require sorted indices.

问题

用稀疏矩阵作为model的输入，提示错误：

InvalidArgumentError: indices[1] = [0,1247] is out of order. Many sparse ops require sorted indices.
    Use `tf.sparse.reorder` to create a correctly ordered copy.

原因

稀疏矩阵的索引无序，如下所示：

SparseTensor(indices=tf.Tensor(
[[   0 1274]
 [   0 1247]
 [   0 1194]
 ...
 [2707  329]
 [2707  186]
 [2707   19]], shape=(49216, 2), dtype=int64), values=tf.Tensor([0.11111111 0.11111111 0.11111111 ... 0.07692308 0.07692308 0.07692308], shape=(49216,), dtype=float32), dense_shape=tf.Tensor([2708 1433], shape=(2,), dtype=int64))

解决

将稀疏矩阵的索引变有序：

features_s = tf.sparse.reorder(features_s)
print(features_s)

SparseTensor(indices=tf.Tensor(
[[   0   19]
 [   0   81]
 [   0  146]
 ...
 [2707 1328]
 [2707 1412]
 [2707 1414]], shape=(49216, 2), dtype=int64), values=tf.Tensor([0.11111111 0.11111111 0.11111111 ... 0.07692308 0.07692308 0.07692308], shape=(49216,), dtype=float32), dense_shape=tf.Tensor([2708 1433], shape=(2,), dtype=int64))