ENH: optimize get_value_opt in class Function

[styris00]

Pull request type

• Code changes (bugfix, features)

Checklist

• Lint (black rocketpy/ tests/) has passed locally
• All tests (pytest --runslow) have passed locally

Current behavior

Calls to get_value_opt for shepard interpolation were slower.

New behavior

Calls to get_value_opt for shepard interpolation are quicker (using Numpy vectorized operations).

Breaking change

• No

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (13204ce) 71.17% compared to head (1a822a3) 71.16%.
Report is 7 commits behind head on develop.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #501      +/-   ##
===========================================
- Coverage    71.17%   71.16%   -0.02%     
===========================================
  Files           55       55              
  Lines         9277     9273       -4     
===========================================
- Hits          6603     6599       -4     
  Misses        2674     2674              

Flag	Coverage Δ
unittests	71.16% <100.00%> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[giovaniceotto]

Just sending the report made by @styris00 here so that all reviewers can see it.

Optimization_report.pdf

[giovaniceotto]

@styris00, can you share the code you used for your execution time reports? Specially the definition of the Function instances.

[styris00]

@styris00, can you share the code you used for your execution time reports? Specially the definition of the Function instances.

I can't manage to send a python file so I put the code below.

To perform the tests below and compare the two versions of get_value_opt, you have to modify the set_get_value_opt method so the Function instances have the attributs :

• get_value_opt1 : a function that correspond to the former get_value_opt attribut
• get_value_opt2 : a function that correspond to the new get_value_opt attribut

(You may have to also change the RocketPy import)

Note that the matrix MATRIX_RESULTS displayed by the tests function is the one we put in the table of the optimization report

import sys
sys.path.append('..')

import numpy as np
from RocketPy.rocketpy import Function  # May need to be changed
from timeit import Timer
import random as rd



def create_data(dim, size):  # Create a source to then create a Function Object
    list = []
    for s in range(size):
        tuple = ()
        for d in range(dim): # inputs
            tuple = tuple + (rd.uniform(-1000, 1000),)
        tuple = tuple + (rd.uniform(-50, 50),) # output
        list.append(tuple)
    array = np.array(list)
    return array

def create_point(dim):  # Create a point as a tuple to call get_value_opt
    point = ()
    for d in range(dim):
        point = point + (rd.uniform(-1000, 1000),)
    return point


dimensions = [2, 3, 4, 5]
sizes = [100, 1000, 10000]


def tests(nb = 10):
    MATRIX_RESULTS = [["Data Dim", "Data Length", "Time before", "Time after", "Improvement"]]
    correct_values = 0
    for dim in dimensions:
        for size in sizes:
            print("Dim", dim, "Size:", size)
            correct_values_case = 0
            mean_before = 0
            mean_after = 0
            mean_improvement = 0
            for i in range(nb):
                array = create_data(dim, size)
                point = create_point(dim)
                fun = Function(                           # Definition of the Function instance with array as a source
                    array,
                    interpolation="shepard",
                    extrapolation="natural",
                    title="test function"
                )
                fun.set_get_value_opt() # for fun.get_value_opt1 AND fun.get_value_opt2 to work

                value_ref = fun.get_value_opt1(*point)
                #print(value_ref)
                value_new = fun.get_value_opt2(*point)
                #print(value_new)
                same_value = np.isclose(value_ref, value_new, rtol = 1e-12) # check if the two values are nearly the same (1e-12)
                #print(same_value)
                if same_value:
                    correct_values += 1
                    correct_values_case += 1

                globals()['fun'] = fun
                globals()['np'] = np

                timer = Timer(lambda: fun.get_value_opt1(*point), 'from __main__ import fun, np')
                execution_time_ref = timer.timeit(number=1)
                mean_before = mean_before + execution_time_ref

                timer = Timer(lambda: fun.get_value_opt2(*point), 'from __main__ import fun, np')
                execution_time_new = timer.timeit(number=1)
                mean_after = mean_after + execution_time_new

                improvement = (execution_time_new/execution_time_ref)*100
                mean_improvement = mean_improvement + improvement

            mean_before = mean_before/nb
            mean_after = mean_after/nb
            mean_improvement = mean_improvement/nb
            LINE_RESULTS = [str(dim), str(size), mean_before, mean_after, str(mean_improvement)+" %"]
            MATRIX_RESULTS.append(LINE_RESULTS)
            print("Correct values", correct_values_case, "/", nb)

    print("CORRECT VALUES :", correct_values,"/",nb*len(dimensions)*len(sizes))
    MATRIX_RESULTS = np.array(MATRIX_RESULTS)
    print(MATRIX_RESULTS)

tests(100) # performs the tests

[Gui-FernandesBR]

Just sending the report made by @styris00 here so that all reviewers can see it.

Optimization_report.pdf

This report was really nice to read!

Just wondering, what were the computer specification when running all the tests? (for example: i7-7200U, running python 3.10 on ubuntu jammy)

[Gui-FernandesBR]

Well done! I'm happy seeing this speedup on the code.
Your work was amazing in this PR, thanks for contributions.

[styris00]

Just sending the report made by @styris00 here so that all reviewers can see it.
Optimization_report.pdf

This report was really nice to read!

Just wondering, what were the computer specification when running all the tests? (for example: i7-7200U, running python 3.10 on ubuntu jammy)

The table was created with an AMD Ryzen 7 6800H on Windows 11.
The same improvements (%) were found with an Intel i5-1035G1 on Windows 11.

If you've run these tests too, we wonder if you got roughly the same results ?

[Gui-FernandesBR]

Just sending the report made by @styris00 here so that all reviewers can see it.
Optimization_report.pdf

This report was really nice to read!
Just wondering, what were the computer specification when running all the tests? (for example: i7-7200U, running python 3.10 on ubuntu jammy)

The table was created with an AMD Ryzen 7 6800H on Windows 11. The same improvements (%) were found with an Intel i5-1035G1 on Windows 11.

If you've run these tests too, we wonder if you got roughly the same results ?

Actually I didn't run the same tests, I am trusting on your analysis.

[Gui-FernandesBR]

I'm comfortable enough with this PR, I think you can merge it if you are comfortable too, or you can wait for another review if you think it's necessary. The PR's main goal was achieved.