Skip to content

Untruncation Actions

LegUntruncation

Bases: RmAction

Leg-level demand untruncation action.

This action will untruncate demand on legs using the specified algorithm. It is called only once at the beginning of each sample.

produces class-attribute instance-attribute 

produces: set[str] = {'leg_demand'}

frequency class-attribute instance-attribute 

frequency = 'begin_sample'

which_data instance-attribute 

which_data = which_data

Which data to use for untruncation.

algorithm instance-attribute 

algorithm: Literal[
    "em",
    "em_py",
    "none",
    "naive1",
    "naive2",
    "naive3",
    "pd",
    "mfem",
] = algorithm

Untruncation algorithm.

There are several available algorithms:

none applies no untruncation, and assumes that demand was the same as sales. Applying this algorithm is still important even if no detruncation is desired, as PassengerSim tracks historical demand separately from sales and without "none" the historical demand used in forecasting would be zero.

em is an expectation-maximization model.

em_py is an expectation-maximization model implemented in Python. It is slow but useful for educational purposes.

naive1 is not recommended for use.

naive2 is not recommended for use.

naive3 is not recommended for use.

pd is a projection-detruncation model, based on the method developed by Hopperstad and described by Weatherford & Polt.

mfem Multi-Flight Expectation Maximization. Takes into account spill and recapture across multiple departures

maxiter instance-attribute 

maxiter = maxiter

Maximum number of iterations for the EM and PD algorithms.

tolerance instance-attribute 

tolerance = tolerance

Tolerance for the EM and PD algorithms.

initialization_method instance-attribute 

initialization_method: Literal["default", "pods"] = (
    initialization_method
)

Method for initializing the EM algorithm.

The default method is to use all available data on the first EM iteration. The pods method is to use only the data from unclosed observations on the first EM iteration.

minimum_mu instance-attribute 

minimum_mu = minimum_mu

Minimum value for the mean of the demand distribution.

If the computed mean is less than this value, untruncation will result in zero demand. Setting this to a very small but non-zero value can help avoid thin-path problems, where untruncation results in some non-zero demand on every path-class, even though many path-classes have zero historical sales and probably will stay that way.

minimum_sigma instance-attribute 

minimum_sigma = minimum_sigma

Minimum value for the standard deviation of the demand distribution.

If the computed sigma is less than this value, this value is used instead.

__init__ 

__init__(
    *,
    carrier: str = "",
    minimum_sample: int = 10,
    dcps: Collection[int] = (),
    algorithm: Literal[
        "em",
        "em_py",
        "none",
        "naive1",
        "naive2",
        "naive3",
        "pd",
        "mfem",
    ] = "em",
    which_data: Literal[
        "total", "yieldable", "priceable"
    ] = "total",
    maxiter: int = 20,
    tolerance: float = 0.01,
    initialization_method: Literal[
        "default", "pods"
    ] = "default",
    minimum_mu: float = 0.01,
    minimum_sigma: float = 0.1,
)

run 

run(sim: Simulation, days_prior: int)

PathUntruncation

Bases: RmAction

Path-level demand untruncation tool.

produces class-attribute instance-attribute 

produces: set[str] = {'path_demand'}

frequency class-attribute instance-attribute 

frequency = 'begin_sample'

which_data instance-attribute 

which_data = which_data

Which data to use for untruncation.

algorithm instance-attribute 

algorithm = algorithm

Untruncation algorithm.

There are several available algorithms:

none applies no untruncation, and assumes that demand was the same as sales. Applying this algorithm is still important even if no detruncation is desired, as PassengerSim tracks historical demand separately from sales and without "none" the historical demand used in forecasting would be zero.

em is an expectation-maximization model.

em_py is an expectation-maximization model implemented in Python. It is slow but useful for educational purposes.

naive1 is not recommended for use.

naive2 is not recommended for use.

naive3 is not recommended for use.

pd is a projection-detruncation model, based on the method developed by Hopperstad and described by Weatherford & Polt.

mfem Multi-Flight Expectation Maximization. Takes into account spill and recapture across multiple departures

maxiter instance-attribute 

maxiter = maxiter

Maximum number of iterations for the EM and PD algorithms.

tolerance instance-attribute 

tolerance = tolerance

Tolerance for the EM and PD algorithms.

initialization_method instance-attribute 

initialization_method = initialization_method

Method for initializing the EM algorithm.

The default method is to use all available data on the first EM iteration. The pods method is to use only the data from unclosed observations on the first EM iteration.

minimum_mu instance-attribute 

minimum_mu = minimum_mu

Minimum value for the mean of the demand distribution.

If the computed mean is less than this value, untruncation will result in zero demand. Setting this to a very small but non-zero value can help avoid thin-path problems, where untruncation results in some non-zero demand on every path-class, even though many path-classes have zero historical sales and probably will stay that way.

minimum_sigma instance-attribute 

minimum_sigma = minimum_sigma

Minimum value for the standard deviation of the demand distribution.

If the computed sigma is less than this value, this value is used instead.

__init__ 

__init__(
    *,
    carrier: str = "",
    minimum_sample: int = 10,
    dcps: Collection[int] = (),
    algorithm: Literal[
        "em",
        "em_py",
        "none",
        "naive1",
        "naive2",
        "naive3",
        "pd",
        "mfem",
    ] = "em",
    which_data: Literal[
        "total", "yieldable", "priceable"
    ] = "total",
    maxiter: int = 20,
    tolerance: float = 0.01,
    initialization_method: Literal[
        "default", "pods"
    ] = "default",
    minimum_mu: float = 0.01,
    minimum_sigma: float = 0.1,
)

run 

run(sim: Simulation, days_prior: int)