Simple FCFS¶

This is a very simple network, with only 3 markets, used to demonstrate some features of PassengerSim.

In [1]:

import passengersim as pax

pax.versions()

import passengersim as pax pax.versions()

passengersim 0.57.dev4+g8e386693f
passengersim.core 0.57.dev0+g85ef573c.d20251014

This example uses network/01-base.yaml configuration file. Within a Jupyter notebook, we can directly initialize a PassengerSim Simulation instance from this file using the from_yaml class constructor:

In [2]:

sim = pax.Simulation.from_yaml("network/01-base.yaml")

sim = pax.Simulation.from_yaml("network/01-base.yaml")

/var/folders/l1/yt63lf3n60b1dc25d_y2d1q80000gn/T/ipykernel_39966/2459294051.py:1: UserWarning: Restriction 'R2' found in fares but not used in any choice model
  sim = pax.Simulation.from_yaml("network/01-base.yaml")
/var/folders/l1/yt63lf3n60b1dc25d_y2d1q80000gn/T/ipykernel_39966/2459294051.py:1: UserWarning: Restriction 'R1' found in fares but not used in any choice model
  sim = pax.Simulation.from_yaml("network/01-base.yaml")
/var/folders/l1/yt63lf3n60b1dc25d_y2d1q80000gn/T/ipykernel_39966/2459294051.py:1: UserWarning: Restriction 'R3' found in fares but not used in any choice model
  sim = pax.Simulation.from_yaml("network/01-base.yaml")

Running the simulation is as simple as calling the run command, which runs the simulation and returns a summary output object.

In [3]:

summary = sim.run()

summary = sim.run()

Task Completed after 6.10 seconds

The contents of the final summary is controlled by Config.outputs.reports, which allows the user to add reports for more detail, or drop some unneccessary reports to improve runtime (sometimes substantially).

In [4]:

sim.config.outputs.reports

sim.config.outputs.reports

Out[4]:

{'bookings_by_timeframe',
 'carrier_history',
 'demand_to_come',
 'fare_class_mix',
 'leg_forecasts',
 'load_factors',
 'total_demand'}

For this example, several default reports are included, which allows us to access a number of pre-packaged visualizations for the results.

In [5]:

summary.fig_carrier_revenues()

summary.fig_carrier_revenues()

Out[5]:

In [6]:

summary.fig_carrier_load_factors()

summary.fig_carrier_load_factors()

Out[6]:

In [7]:

summary.fig_carrier_mileage()

summary.fig_carrier_mileage()

Out[7]:

In [8]:

summary.fig_fare_class_mix()

summary.fig_fare_class_mix()

Out[8]:

All demand is in the lowest fare class, because the simulation has no restrictions against customers simply buying the least expensive fare.

In [9]:

summary.fig_bookings_by_timeframe()

summary.fig_bookings_by_timeframe()

Out[9]:

We are not limited to the pre-packaged visualizations. The various summary tables available in the summary object are all just regular pandas DataFrames, so we can use all the usual Python and Pandas tools for analysis. For example, the demand_to_come table summarizes the total demand to come at each timeframe for every simulation sample.

In [10]:

summary.demand_to_come

summary.demand_to_come

Out[10]:

					days_prior	63	56	49	42	35	31	28	24	21	17	14	10	7	5	3	1	0
iteration	trial	sample	segment	orig	dest
0	0	100	business	BOS	LAX	65	61	57	55	51	51	49	48	42	39	39	31	25	19	14	4	0
					ORD	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
				ORD	LAX	68	63	59	59	54	52	50	48	47	44	36	32	25	20	16	4	0
			leisure	BOS	LAX	71	58	56	49	43	39	35	30	28	21	16	12	7	7	4	1	0
					ORD	49	43	39	34	32	31	27	24	19	15	13	11	11	10	6	1	0
		...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
		499	business	BOS	ORD	83	77	75	72	70	66	65	62	59	56	55	45	36	27	21	2	0
				ORD	LAX	155	141	134	131	120	115	114	111	106	99	91	81	62	46	32	12	0
			leisure	BOS	LAX	215	182	165	148	126	113	109	89	84	64	51	42	35	35	20	7	0
					ORD	133	111	105	91	76	65	58	49	47	33	26	18	18	15	6	2	0
				ORD	LAX	198	165	146	132	114	102	95	74	57	51	43	32	24	21	11	4	0

2400 rows × 17 columns

With this data, we can do whatever analysis we like. Here we'll compute the correlation between total demands (from the beginning of the booking curve at DCP 63 all the way to the end) of different passenger types in different markets.

In [11]:

summary.demand_to_come[63].unstack(["segment", "orig", "dest"]).corr()

summary.demand_to_come[63].unstack(["segment", "orig", "dest"]).corr()

Out[11]:

		segment	business			leisure
		orig	BOS		ORD	BOS		ORD
		dest	LAX	ORD	LAX	LAX	ORD	LAX
segment	orig	dest
business	BOS	LAX	1.000000	0.083927	0.231989	0.337721	0.109801	0.136725
		ORD	0.083927	1.000000	0.132940	0.117638	0.371579	0.111497
	ORD	LAX	0.231989	0.132940	1.000000	0.190033	0.217089	0.352117
leisure	BOS	LAX	0.337721	0.117638	0.190033	1.000000	0.156787	0.140052
		ORD	0.109801	0.371579	0.217089	0.156787	1.000000	0.147393
	ORD	LAX	0.136725	0.111497	0.352117	0.140052	0.147393	1.000000

The summary object also has a command to dump all the summary tables to an Excel workbook, if you prefer to analyze the results there instead of in Jupyter.

In [12]:

summary.to_xlsx("outputs/3mkt-01.xlsx")

summary.to_xlsx("outputs/3mkt-01.xlsx")

Download 3mkt-01.xlsx

Comparing against Targets¶

In addition to summary reports for a single run, we can also use PassengerSim's contrast package to compare simulation runs to each other, or against exogenously defined target results.

In [13]:

import targets

target = targets.load(1, sim.config)

import targets target = targets.load(1, sim.config)

In [14]:

from passengersim import contrast

comps = contrast.Contrast(
    {
        "simulation": summary,
        "target": target,
    }
)

from passengersim import contrast comps = contrast.Contrast( { "simulation": summary, "target": target, } )

In [15]:

comps.fig_carrier_revenues()

comps.fig_carrier_revenues()

Out[15]:

In [16]:

comps.fig_bookings_by_timeframe(by_carrier="AL1")

comps.fig_bookings_by_timeframe(by_carrier="AL1")

Out[16]:

In [17]:

comps.fig_segmentation_by_timeframe("bookings", by_carrier=False, by_class=True)

comps.fig_segmentation_by_timeframe("bookings", by_carrier=False, by_class=True)

Out[17]:

We can look at carrier forecasts of demand on individual legs.

In [18]:

comps.fig_leg_forecasts(by_leg_id=111, of=["mu", "sigma"])

comps.fig_leg_forecasts(by_leg_id=111, of=["mu", "sigma"])

Out[18]:

We can compare the mean and standard deviation of demand to come.

In [19]:

comps.fig_demand_to_come("mean") | comps.fig_demand_to_come("std")

comps.fig_demand_to_come("mean") | comps.fig_demand_to_come("std")

Out[19]:

We can even take arbitrary functions that apply pandas tools, and have them run automatically against multiple summary objects. For example, we can look at the variance-covariance matrix of aggregate demand by passenger type, and compare those matrices for both the simulation and the target.

In [20]:

comps.apply(lambda s: s.aggregate_demand_history(by_segment=True).unstack("segment").cov())

comps.apply(lambda s: s.aggregate_demand_history(by_segment=True).unstack("segment").cov())

Out[20]:

	segment	business	leisure
source	segment
simulation	business	3665.319273	2224.550439
	leisure	2224.550439	5590.625157
target	business	3249.598992	1955.419868
	leisure	1955.419868	5359.555328