pseudo random numbers cannot be reproduced by fixing seeds

[toruseo]

UXsim uses pseudo random numbers to simulate stochastic route choice and traffic merging behaviors at nodes. I have implemented a function to fix the random seed by random_seed arg of World.__init__(), but it does not work properly under certain conditions.

UXsim/uxsim/uxsim.py

Lines 1381 to 1384 in 1e5fac5

	## parameter setting
	random.seed(random_seed)
	np.random.seed(random_seed)

In simple scenarios it works well, so I guess there are some small mistakes.

Example

Code:

from uxsim import *
from uxsim import Utilities

#########################################
# Iter 1
#########################################

W = World(
    name="",
    deltan=10,
    tmax=3600,
    print_mode=1, save_mode=1, show_mode=0,
    random_seed=42
)

n_nodes = 5
Utilities.generate_grid_network(W, n_nodes, n_nodes, length=1000)
od_pairs = [
    (f"n(0, 0)", f"n({n_nodes-1}, {n_nodes-1})"),
    (f"n({n_nodes-1}, 0)", f"n(0, {n_nodes-1})"),
    (f"n(0, {n_nodes-1})", f"n({n_nodes-1}, 0)"),
    (f"n({n_nodes-1}, {n_nodes-1})", f"n(0, 0)"),
]
for od_pair in od_pairs:
    W.adddemand(od_pair[0], od_pair[1], 0, 3000, 0.7)

W.exec_simulation()
W.analyzer.print_simple_stats()

#########################################
# Iter 2
#########################################

W = World(
    name="",
    deltan=10,
    tmax=3600,
    print_mode=1, save_mode=1, show_mode=0,
    random_seed=42
)

n_nodes = 5
Utilities.generate_grid_network(W, n_nodes, n_nodes, length=1000)
od_pairs = [
    (f"n(0, 0)", f"n({n_nodes-1}, {n_nodes-1})"),
    (f"n({n_nodes-1}, 0)", f"n(0, {n_nodes-1})"),
    (f"n(0, {n_nodes-1})", f"n({n_nodes-1}, 0)"),
    (f"n({n_nodes-1}, {n_nodes-1})", f"n(0, 0)"),
]
for od_pair in od_pairs:
    W.adddemand(od_pair[0], od_pair[1], 0, 3000, 0.7)

W.exec_simulation()
W.analyzer.print_simple_stats()

Results for Iter 1

results:
 average speed:	 16.7 m/s
 number of completed trips:	 8400 / 8400
 average travel time of trips:	 450.7 s
 average delay of trips:	 50.7 s
 delay ratio:			 0.112

Results for Iter 2

results:
 average speed:	 16.0 m/s
 number of completed trips:	 8400 / 8400
 average travel time of trips:	 471.8 s
 average delay of trips:	 71.8 s
 delay ratio:			 0.152

[EwoutH]

scientific-python/specs#180 might be interesting to keep an eye on.

[toruseo]

Finally I have solved this issue. The ultimate source of this bug was that set in Python does not preserve the order.

UXsim/uxsim/uxsim.py

Line 221 in d87a650

for outlink in {veh.route_next_link for veh in s.incoming_vehicles if veh.route_next_link != None}:

@EwoutH Thanks for the info. I have also updated the random number generators to the modern implementation.