Retrospective Analysis¶

In this example, we will demonstrate some analysis that is reading back results of prior completed simulation run. This is possible because PassengerSim can be configured to write out quite a bit of data to a SQLite database while running. Much of this output is optional, as writing out intermediate results can create very large database files and impose a significant runtime penalty on the simulation, but having the ability to do so is advantageous for research and development purposes.

In [1]:

import passengersim as pax
pax.versions()

import passengersim as pax pax.versions()

passengersim 0.18.1
passengersim.core 0.18.1

We can load in SummaryTables objects directly from the database created during a simulation run (assuming it was saved to disk). For this demo, we have created two small such databases by running two different models with on-disk database files, and then discarding the results (for now). We can then load summary tables of results from these two simulations by using the SummaryTables.from_sqlite constructor. This command will run queries against the database to reconstruct the same summary table objects we would have gotten as the output from running the simulation.

In [3]:

simple = pax.SummaryTables.from_sqlite("simple.sqlite", additional="*")
untrunc = pax.SummaryTables.from_sqlite("untruncated.sqlite", additional="*")

simple = pax.SummaryTables.from_sqlite("simple.sqlite", additional="*") untrunc = pax.SummaryTables.from_sqlite("untruncated.sqlite", additional="*")

In [4]:

from passengersim.contrast import Contrast

comps = Contrast(
    Simple=simple, Untruncated=untrunc
)

from passengersim.contrast import Contrast comps = Contrast( Simple=simple, Untruncated=untrunc )

All the same visualizations available in our prior demonstrations are also available for use with the pre-computed results. Since we are not actually running the simulation again, this analysis can even be done by users who don't have (or need) access to the PassengerSim core simulation code.

In [5]:

comps.fig_carrier_revenues()

comps.fig_carrier_revenues()

Out[5]:

In [6]:

comps.fig_carrier_load_factors()

comps.fig_carrier_load_factors()

Out[6]:

In [7]:

comps.fig_fare_class_mix()

comps.fig_fare_class_mix()

Out[7]:

In [8]:

comps.fig_bookings_by_timeframe(by_carrier="AL1", by_class=True, source_labels=True)

comps.fig_bookings_by_timeframe(by_carrier="AL1", by_class=True, source_labels=True)

Out[8]:

The results of the simulation are stored in database tables. We can write queries using typical SQLite syntax against the database directly if we like. For convenience, the SQLite database connection used to populate the summary tables is available as the cnx attribute of the SummaryTables object, and we can use its dataframe method to run arbitrary queries and get the results as a pandas DataFrame.

In [9]:

simple.cnx.dataframe("""
SELECT
  sample, auth, sold, forecast_mean, forecast_stdev
FROM 
  leg_bucket_detail 
WHERE
  flt_no = 101
  AND rrd = 21
  AND name = 'Y2'
  AND sample >= 100
LIMIT 10
""")

simple.cnx.dataframe(""" SELECT sample, auth, sold, forecast_mean, forecast_stdev FROM leg_bucket_detail WHERE flt_no = 101 AND rrd = 21 AND name = 'Y2' AND sample >= 100 LIMIT 10 """)

Out[9]:

	sample	auth	sold	forecast_mean	forecast_stdev
0	100	53	2	6.769231	2.518852
1	101	53	2	6.500000	2.745906
2	102	14	6	6.346154	2.770448
3	103	14	5	6.307692	2.782362
4	104	29	6	6.307692	2.782362
5	105	29	3	6.384615	2.772391
6	106	15	1	6.192308	2.828699
7	107	44	3	6.461538	2.745626
8	108	45	2	6.461538	2.745626
9	109	40	2	6.615385	2.772391