About the Runway Functions category

Runway Functions
43

Hello! Thank you so much for all the information on top? I just wanted to double check two points.

  1. Can I assume that the exact same airport “classes”/“bins” (including other) will be on the partial_submission_format.csv if an airport appears? If so, can I also assume that they will appear in the exact same order as they do in the open_screen dataset?
    (For example, for katl can we assume that the possible configs and the order that they appar in the submission format is exactly:
    0 katl:D_10_8L_A_10_8L
    1 katl:D_10_8R_9L_A_10_8L_9R
    2 katl:D_10_8R_A_10_8R
    3 katl:D_26L_27L_A_26R_27L_28
    4 katl:D_26L_27R_28_A_26R_27L_28
    5 katl:D_26L_27R_A_26L_27L_28
    6 katl:D_26L_27R_A_26L_27R
    7 katl:D_26L_27R_A_26L_27R_28
    8 katl:D_26L_27R_A_26R_27L
    9 katl:D_26L_27R_A_26R_27L_28
    10 katl:D_26L_27R_A_26R_27R_28
    11 katl:D_26L_27R_A_26R_28
    12 katl:D_26L_27R_A_27L_28
    13 katl:D_26L_28_A_26L_28
    14 katl:D_26L_28_A_26R_27L_28
    15 katl:D_26L_28_A_26R_28
    16 katl:D_26R_27R_A_26R_27L_28
    17 katl:D_26R_28_A_26R_28
    18 katl:D_8L_9L_A_10_8L_9R
    19 katl:D_8R_9L_A_10_8L_9R
    20 katl:D_8R_9L_A_10_8R_9R
    21 katl:D_8R_9L_A_10_9R
    22 katl:D_8R_9L_A_8L_9R
    23 katl:D_8R_9L_A_8R_9L
    24 katl:D_8R_9R_A_10_8L_9R
    25 katl:D_9L_A_9R
    26 katl:other
    )

  2. Is there any margin for floating point error for adding the probability to one? I am doing panda/numpy “norm” but as expected there are small amounts of floating point errors.

Thank you so much!

44

Yes, the final evaluation set will use the same configurations in the same order as the open and prescreened datasets.

Yep, you can actually see the exact check we run in the runtime repo. https://github.com/drivendataorg/nasa-airport-config-runtime/blob/7393e3c48d091146c241566714b75b914eb845c7/runtime/scripts/check_prediction.py#L48

It uses numpy.allclose, which should account for floating point error.