1
0
This repository has been archived on 2025-03-06. You can view files and clone it, but cannot push or open issues or pull requests.
on-restart-benchmarks/analyse_chuffed.py

156 lines
5.0 KiB
Python
Executable File

#!/usr/bin/env python3
"""
This script will produce the final value of a variable named 'objective' and its
area for every file ending in '.sol' in the directory provided. The files are
expected to be fzn-gecode output piped through solns2out with the
'--output-time' flag. Furthermore the file is expected to contain the initial
area which can be found by adding
'constraint trace("% init_area = \(ub(objective));\n", true);'
to the model in question.
"""
import csv
import os
import re
import sys
from statistics import stdev
def compute_area(file, time):
area = -1
objectives = []
times = [0]
timeout = -1
objectives.append(0)
for line in contents:
# match = re.match(r'%\sinit_area\s=\s(\d+)', line)
# if match:
# objectives.append(int(match.group(1)))
# continue
match = re.match(r"objective\s=\s(\d+)", line)
if match:
objectives.append(int(match.group(1)))
continue
match = re.match(r"%\stime elapsed:\s(\d+\.\d+)\ss", line)
if match:
times.append(float(match.group(1)))
continue
# Not proven optimal
if len(times) == len(objectives):
times.append(time)
assert len(objectives) > 0
assert len(objectives) + 1 == len(times)
area = 0
for i in range(len(objectives)):
area += (times[i + 1] - times[i]) * objectives[i]
return int(area)
folder = sys.argv[1]
statistics = dict()
instances = set()
for config in ["original", "restart"]:
for root, dirs, files in os.walk(folder + "/" + config):
for name in files:
if name.endswith(".sol"):
components = name[:-(4)].split(".")
data = components[0]
instances.add(data)
seed = 1
if len(components) > 1:
assert len(components) == 2
seed = components[1]
if data not in statistics:
statistics[data] = dict()
if config not in statistics[data]:
statistics[data][config] = []
with open(os.path.join(root, name)) as f:
contents = f.readlines()
nodes = None
solvetime = None
restarts = None
objective = None
for line in contents:
# Nodes
match = re.search(r"%%%mzn-stat: nodes=(\d+)", line)
if match:
nodes = int(match.group(1))
continue
# Solve time
match = re.search(r"%%%mzn-stat: solveTime=(\d+\.\d+)", line)
if match:
solvetime = float(match.group(1))
continue
# Restarts
match = re.search(r"%%%mzn-stat: restarts=(\d+)", line)
if match:
restarts = int(match.group(1))
continue
for line in contents[::-1]:
# Best objective
match = re.match(r"%%%mzn-stat: objective=(-?\d+)", line)
if match:
objective = int(match.group(1))
break
# Area
area = compute_area(contents, solvetime)
statistics[data][config].append(
(
area,
objective,
solvetime,
restarts,
nodes,
)
)
for data in instances:
for config in ["original", "restart"]:
stats = statistics[data][config]
cumulative = stats[0]
for i in range(1, len(stats)):
cumulative = tuple(map(sum, zip(cumulative, stats[i])))
avg = tuple([x / len(stats) for x in cumulative])
dev = stdev([x[1] for x in stats]) if len(stats) > 1 else 0
# (avg area, avg objective, stdev objective)
statistics[data][config] = (avg[0], avg[1], dev)
# Print header
print(
"""
\\begin{tabular}{l|rr|rr|rr}
\\toprule
& \multicolumn{2}{c|}{Chuffed} & \multicolumn{2}{c|}{Chuffed Restart} \\\\
Instance & $\intobj$ & $\minobj$ & $\intobj$ & $\minobj$ \\\\
\midrule
"""
)
sorted_instances = sorted(instances)
for data in sorted_instances:
print(f"{data}", end="")
for config in ["original", "restart"]:
print(
f" & {int(statistics[data][config][0] / 1000) }k & {int(statistics[data][config][1])}",
end="",
)
if statistics[data][config][2] != 0:
print("^{", end="")
print(
int(statistics[data][config][2] / statistics[data][config][1] * 100),
# int(statistics[data][config][2]),
end="",
)
print("}", end="")
print(" \\\\")
# Print footer
print("\n\\bottomrule\n\end{tabular}")