Shapes.reduceDims is difficult to understand

I find the documentation opaque:

  /**
   * Reshapes the operand by reducing the shape to the specified axis.
   *
   * @param <T> the type of Operand
   * @param scope current scope
   * @param operand the operand
   * @param axis the axis
   * @return the reshaped operand
   */
  @Endpoint(name = "reduceDims")
  public static <T extends TType> Operand<T> reduceDims(
      Scope scope, Operand<T> operand, Operand<TInt32> axis) {

I had no success searching for a similar function in Python TensorFlow. So I read the code, but it is fairly complex to follow.

Start with the Shapes.reduceDims method and read its current Javadoc and implementation. Compare the operation with the available TensorFlow behavior, then update the documentation so the reshape, axis, operand, and resulting shape are understandable; done when a reader can tell how reduceDims behaves without tracing complex code.