tf.raw_ops.UpperBound

Applies upper_bound(sorted_search_values, values) along each row.

tf.raw_ops.UpperBound(
    sorted_inputs, values, out_type=tf.dtypes.int32, name=None
)

Each set of rows with the same index in (sorted_inputs, values) is treated independently. The resulting row is the equivalent of calling np.searchsorted(sorted_inputs, values, side='right').

The result is not a global index to the entire Tensor, but rather just the index in the last dimension.

A 2-D example: sorted_sequence = [[0, 3, 9, 9, 10], [1, 2, 3, 4, 5]] values = [[2, 4, 9], [0, 2, 6]]

result = UpperBound(sorted_sequence, values)

result == [[1, 2, 4], [0, 2, 5]]