M¶

class passengersim.rm.standard_systems.M[source]¶

Bases: RmSys

A standard RM system of type “M”.

This RM system uses path-level bid price controls with Pro-BP optimization, along with EM untruncation of yieldable demand and hybrid conditional path forecasting.

Parameters:

forecast_algorithm : {'additive_pickup', 'exp_smoothing', 'multiplicative_pickup'}, default 'additive_pickup'

Specifies which leg-level forecasting algorithm to use for generating leg demand forecasts. Options are ‘additive_pickup’, ‘exp_smoothing’, or ‘multiplicative_pickup’. The default is ‘additive_pickup’.

bid_price_vector : bool, default True

If True, enables bid price vector optimization in ProBP. If False, uses scalar bid prices.

fare_adjustment : {'mr', 'ki', None}, default 'mr'

Specifies the type of fare adjustment to apply in the conditional path forecasting algorithm. The first option is ‘mr’ for marginal revenue adjustment, inspired by the work of Fiig, Isler, Hopperstad, and Belobaba (2010), which assumes a continuous pricing curve with a negative exponential customer sellup function. The second option is ‘ki’ for Karl Isler’s method, which is a discrete pricing adjustment method that assumes no particular functional form of the customer sellup function. It is also possible to select None for no fare adjustment, although this is not usually recommended except for restricted product sets.

fare_adjustment_scale : float, default 0.5

Specifies the scale factor to use for fare adjustment in the conditional path forecasting algorithm. This factor determines the magnitude of the fare adjustment applied to the forecasts. A value of 1.0 applies the full adjustment, while values less than 1.0 scale down the adjustment. In practice, smaller values are used when the product sets are more restricted, and higher values are used when the product sets are less restricted. Theoretically, the full fare adjustment scale of 1.0 would be appropriate for completely unrestricted product sets, while no fare adjustment (0.0) should be used for heavily restricted product sets. The default setting of 0.5 is a moderate adjustment that works reasonably well in cases with semi-restricted product sets.

regression_weight : {None, 'sellup', 'sellup^2', 'fare', 'none'}, default None

Specifies the type of regression weight to use in the conditional path forecasting algorithm. Options are ‘sellup’ to weight by sellup factor, ‘sellup^2’ to weight by the square of the sellup factor, ‘fare’ to weight by fare amount, ‘none’ for no weighting.

variance_rollup_algorithm : {'tf', 'dep'}, default 'tf'

Specifies how to roll up variance when combining priceable and yieldable forecasts. Options are ‘tf’ to roll up variance by time frame, or ‘dep’ to roll up variance to departure. This setting has no effect if the variance_is_ratio_of_mean parameter is set to a value greater than zero.

variance_is_ratio_of_mean : float, default 2.0

Assume that the variance is this ratio of the mean. When this is set to a value greater than zero, the variance of the forecast is set to this fixed ratio of the mean. Note that many algorithms for optimization use the forecast standard deviation, which is the square root of the variance, but it is the variance that is set to this ratio times the mean. When set to zero (the default), the variance is computed from mean squared error of the linear regression model used to compute the mean, and is not a fixed ratio of the mean. The default value of 2.0 is reflective of methods used in practice.

max_cap : float, default 0.0

Maximum weighting factor for the conditional forecast. If set to a value greater than zero, the weighting factor used in the conditional forecast is capped at this maximum value to prevent excessively high weights from extreme sellup factors. The max cap will be more important when the regression weight is set to ‘sellup^2’, as this weighting can grow very large for extreme sellup factors. If this is applied, it is recommended to set the cap to 10.0, which is consistent with prior work on hybrid forecasting methods.

q_allocation_algorithm : {'tf', 'dep'}, default 'tf'

Specifies how to allocate variance from aggregate Q forecasts to class-level forecasts. Options are ‘tf’ to allocate variance by time frame, or ‘dep’ to allocate variance to departure. This setting has no effect if the variance_is_ratio_of_mean parameter is set to a value greater than zero.

Notes

This RM system consists of the following actions executed in order:

1. EM Untruncation of Leg Demands

   This step applies the EM algorithm to detruncate observed path sales into inferred true demand levels. It runs only once at the beginning of each sample day, and detruncates demand for all timeframes.

2. Hybrid-Conditional Path Forecasting

   This step generates path-level demand forecasts using a hybrid conditional class-based forecasting algorithm (additive pickup by default). It runs full computations to produce path forecasts for the entire booking horizon in one pass at the beginning of each sample day, and on later DCPs it simply moves a pointer forward through that array of forecasts to provide the correct forecast values at that time.

3. Path Forecast Daily Decay Adjustment

   This step applies a daily decay adjustment to the path-level forecasts, to account for the changes in expected demand to come in between DCPs. It runs every day that isn’t a DCP, to adjust the path forecasts accordingly.

4. ProBP Optimization

   Optimizes path-level bid prices using the Probabilistic Bid Price (ProBP) algorithm. This step runs every day, to update the bid price controls based on the current path forecasts, current sales, and the ProBP optimization logic.

References

Fiig, T., Isler, K., Hopperstad, C., & Belobaba, P. (2010). “Optimization of mixed fare structures: Theory and applications.” Journal of Revenue and Pricing Management, 9(1/2), 152-170.

Initialize the RM system.

Parameters:

carrier : str

The carrier code for which this RM system is configured.

cfg : Config, optional

The top level configuration object for the simulation, which is also used to help initialize this system. For example, the collection of data collection points (DCPs) given as days prior to departure, is used for steps that have a frequency of “DCP” to identify which days to run.

**kwargs

Additional keyword arguments to configure the RM actions in this system.

Raises:

ValueError – If availability_control is not defined, or if no actions are defined, or if any provided keyword argument does not match any action’s configuration options.

Attributes

availability_control	The type of availability control used in this RM system.
actions	List of RM actions (or action factories) to execute in this RM system.

availability_control : Literal['leg', 'cabin', 'bp', 'bp_loose', 'classless', None] = 'bp'¶

The type of availability control used in this RM system.

The selected availability control is injected into the Carrier object at the time it is created in the simulation, as this setting is used during the passenger arrival simulation loop, and instead of during the RM system steps run from the action queue each day.

Subclasses of RmSys must set this class variable to one of the allowed values (other than None) to indicate the type of availability control used by that RM system.

actions : list[RmActionFactory | type[RmAction]] = [<passengersim.rm._common.RmActionFactory object>, <passengersim.rm._common.RmActionFactory object>, <class 'passengersim.rm.standard_forecasting.PathForecastDailyDecay'>, <passengersim.rm._common.RmActionFactory object>]¶

List of RM actions (or action factories) to execute in this RM system.

Each item in this list should be either an RmActionFactory instance, or an RmAction subclass. If an RmAction subclass is provided, its configure() class method will be called to create an RmActionFactory with no configurable parameters.

Subclasses of RmSys must set this class variable to a non-empty list of actions to be executed in order when this RM system runs.