dietmarwo / fast-cma-es Goto Github PK

View Code? Open in Web Editor NEW

107.0 4.0 17.0 192.25 MB

A Python 3 gradient-free optimization library

License: MIT License

CMake 0.14% C++ 46.92% Shell 0.01% Python 52.92%

optimization multi-objective parallel quality-diversity

fast-cma-es's People

Contributors

Stargazers

Watchers

Forkers

nju-luke gitter-badger mingchengzuo cydia2018 timjim333 gfadini mohammedqaraad vishalbelsare alien-ware slamim8 chrinide solidahmad aaerox delaunay mrfox-jpg comunidadio worasuch

fast-cma-es's Issues

Stall iterations in BiteOpt

I've checked your fix to stall iterations handling. If stall_iterations is a value 0,1,2,3 (in documentation), you have to multiply it by 64. Please fix this. This applies to Java version as well, not only Python.
Orig:

            if (stall_iterations > 0 && stallCount > stall_iterations*dim) {
                stop = 2;
                break;
            }

Fixed:

            if (stall_iterations > 0 && stallCount > stall_iterations*64*dim) {
                stop = 2;
                break;
            }

Problem with running on Mac OS X

Hey, I am trying to use fcmaes on my MacBook, but when importing GTOP problems from fcmaes.astro I face this error:
no suitable image was found. Did find:
/Users/maryamsmacbookpro/opt/anaconda3/lib/python3.9/site-packages/fcmaes/lib/libgtoplib.dll: unknown file type, first eight bytes: 0x4D 0x5A 0x90 0x00 0x03 0x00 0x00 0x00

I have searched a lot but none of the solutions did work. I'd be very thankful if you help me!

biteopt stopping criteria

I see that biteopt stopping criteria is fixed

if (stallCount > 64*dim) {
stop = 2;
break;
}

it puts BiteOpt at a considerable disadvantage on complex problems. This should be switchable, and off by default. Also it would be better for stopping criteria specified as "stopc64dim", where "stopc" is an integer>0

Also please specify BiteOpt version in each instance its results are mentioned.

cannot find dll file on windows

I'm using "evolopy" project on my windows 10, but I got this error in "fcmaes":

FileNotFoundError: Could not find module 'D:\anaconda\envs\ML\lib\site-packages\fcmaes\lib\libgtoplib.dll' (or one of its dependencies).

However, I truly found this file in the right directory. Is that a bug on Windows? How can I deal with it? Must I use linux instead? Thx.

failed to use fcmaes on Debian 10

Hi, I'm trying to use fcmaes on Debian stable (10) but I get an error GLIBCXX_3.4.26 not found.

I guess libacmalib.so needs a more recent GCC but my GCC is not very old (8.3). Is it possible to have a more portable pre-built .so file, or its source code ?

Here is what I've done:

$ sudo apt install intel-mkl

$ virtualenv env -p python3

$ source env/bin/activate

(env) $ pip install fcmaes

(env) $ python
Python 3.7.3 (default, Dec 20 2019, 18:57:59)
[GCC 8.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.

>>> from fcmaes import cmaescpp
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/toto/test-fcmaes/env/lib/python3.7/site-packages/fcmaes/cmaescpp.py", line 129, in <module>
libcmalib = ct.cdll.LoadLibrary(basepath + '/lib/libacmalib.so')
File "/usr/lib/python3.7/ctypes/__init__.py", line 434, in LoadLibrary
return self._dlltype(name)
File "/usr/lib/python3.7/ctypes/__init__.py", line 356, in __init__
self._handle = _dlopen(self._name, mode)
OSError: /lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.26' not found (required by /home/toto/test-fcmaes/env/lib/python3.7/site-packages/fcmaes/lib/libacmalib.so)

CMA Cpp ignores x0 / guess

It seems that CMA Cpp completely ignores x0 /guess as this code shows:

fast-cma-es/_fcmaescpp/acmaesoptimizer.cpp

Lines 163 to 165 in 7952d2b

 bestValue = DBL_MAX; 

 // best parameters so far 

 bestX = guess;

Since guess is not evaluated, its value is assumed to be DBL_MAX, which means that it gets quickly removed.

What should happen is that guess is evaluated in the initialization and bestValue updated accordingly. And even better interface would allow to specify guessValue since this is often the best solution found in a previous iteration and we may wish to avoid evaluating it again.

[Question] Difference between Python/C++ functions

Hi, not an issue, just a question here!

First, thanks for what looks like a great library! I'm currently using SciPy's DE for the optimisation of an EHVI-based acquisition function - it's been working ok but is running fairly slowly for a 22-variable problem. Looking at your benchmarks and testing, I'm tempted to try out this library, as my next project will probably study 35-40 variables.

I was wondering, in your tests, you switch between calling the python and C++ versions of the functions. Are there any differences between them, other than the implementation language? I'm interested in the DE->CMA process, could I just call the C++ implementation (for speed)?

Thanks!

C++ algorithms return None on Windows

Hi Dietmar,

I installed fast-cma-es on Windows 11 following the instructions in README (X64 C++ runtime libraries were also installed).
Tried a small optimization:

import numpy as np
from scipy.optimize import Bounds
from fcmaes import cmaescpp, cmaes


def obj(x):
    return np.sum(x**2)

bounds = Bounds([-1, -1], [1, 1])
result = cmaescpp.minimize(obj, bounds)
print(result)

and got the following result:

 success: False
  status: -1
     fun: 1.7976931348623157e+308
       x: None
     nit: 0
    nfev: 0

The python algorithm works (cmaes). Output:

 success: True
  status: 4
     fun: 3.7194596397089263e-19
       x: [ 5.707e-10 -2.152e-10]
     nit: 78
    nfev: 1209

What do you think could be the cause? Thanks :)

Missing library loguru when installing fcmaes 1.6.3 with pip

Required package loguru is not installed automatically by pip in the 1.6.3 version.

I tested with python 3.9 and 3.11:

starting from a fresh virtual environment

~$ pip install fcmaes
~$ pip list
Package       Version
------------- -------
fcmaes        1.5.9
joblib        1.3.2
numpy         1.26.0
pip           23.2.1
scikit-learn  1.3.1
scipy         1.11.3
setuptools    68.2.0
threadpoolctl 3.2.0
wheel         0.41.2

>>> from fcmaes import cmaes
ModuleNotFoundError: No module named 'loguru'

Although loguru is specified as "required" in setup.py, it is not mentioned in the wheel on PyPi.
In the METADATA file of the wheel, we can read the dependencies:
"Requires-Dist: numpy
Requires-Dist: scipy
Requires-Dist: scikit-learn
Requires-Dist: threadpoolctl"

Of course, for now, I could install loguru manually to solve the issue.

I suppose the same issue will arise with missing numba package.

M1 Support

Hi! I tried to install fcmaes through both pip and conda, but it gave me this error

Output exceeds the [size limit](command:workbench.action.openSettings?%5B%22notebook.output.textLineLimit%22%5D). Open the full output data [in a text editor](command:workbench.action.openLargeOutput?91caeebd-c551-49fc-8a5d-5196a5210d61)
---------------------------------------------------------------------------
OSError                                   Traceback (most recent call last)
[/Users/ivy/src/experiments/squares/square-fcmaes.ipynb](https://file+.vscode-resource.vscode-cdn.net/Users/ivy/src/experiments/squares/square-fcmaes.ipynb) Cell 1 in 1
----> [1](vscode-notebook-cell:/Users/ivy/src/experiments/squares/square-fcmaes.ipynb#W0sZmlsZQ%3D%3D?line=0) from fcmaes.optimizer import logger
      [2](vscode-notebook-cell:/Users/ivy/src/experiments/squares/square-fcmaes.ipynb#W0sZmlsZQ%3D%3D?line=1) from fcmaes import retry

File [/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/optimizer.py:16](https://file+.vscode-resource.vscode-cdn.net/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/optimizer.py:16)
     13 import ctypes as ct
     14 import multiprocessing as mp 
---> 16 from fcmaes import cmaes, de, cmaescpp, decpp, dacpp, gcldecpp, lcldecpp, ldecpp, csmacpp, bitecpp
     18 _logger = None
     20 def logger(logfile = 'optimizer.log'):

File [/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/cmaescpp.py:19](https://file+.vscode-resource.vscode-cdn.net/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/cmaescpp.py:19)
     17 from scipy.optimize import OptimizeResult
     18 from fcmaes.cmaes import _check_bounds
---> 19 from fcmaes.decpp import mo_call_back_type, callback, libcmalib
     21 os.environ['MKL_DEBUG_CPU_TYPE'] = '5'
     23 def minimize(fun, 
     24              bounds=None, 
     25              x0=None, 
   (...)
     34              normalize = True,
     35              update_gap = None):   
...
--> 374     self._handle = _dlopen(self._name, mode)
    375 else:
    376     self._handle = handle

OSError: dlopen([/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/lib/libacmalib.dll](https://file+.vscode-resource.vscode-cdn.net/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/lib/libacmalib.dll), 0x0006): tried: '[/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/lib/libacmalib.dll](https://file+.vscode-resource.vscode-cdn.net/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/lib/libacmalib.dll)' (not a mach-o file), '[/System/Volumes/Preboot/Cryptexes/OS/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/lib/libacmalib.dll](https://file+.vscode-resource.vscode-cdn.net/System/Volumes/Preboot/Cryptexes/OS/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/lib/libacmalib.dll)' (no such file), '[/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/lib/libacmalib.dll](https://file+.vscode-resource.vscode-cdn.net/opt/homebrew/Caskroom/miniforge/base/envs/petri/lib/python3.9/site-packages/fcmaes/lib/libacmalib.dll)' (not a mach-o file

Is there some other way to install on M1 macs or is this just not supported? Thanks :)

Usage for diploma thesis

Dear Mr. Wolf,

Im a student from Kaiserslautern, Germany and Id like to involve your seemingly well done piece of work here in my diploma thesis. It is about direct benchmarking quantum solver results with proper up-to-date solvers on common hardware. As I read about MIT license here, am I allowed to use your code here to work with and publish it. Of course you as author will be called in the final thesis then. In that case, please tell me the correct source to mention for the job shop solver.

With kind regards,
Luca Graebenteich

JSON of the employee scheduling example

Thank you for the OptaPy comparison for Employee Scheduling problem.

Can you upload the JSON representation of the employee scheduling problem mentioned in the tutorial so we do not have to run the OptaPy solution.

Or even better would be to have an example with one of the nurse rostering benchmark problems.

illegal hardware instruction when importing mode on mac M1

Minimal steps to reproduce :
run python in a terminal (I am currently using python 3.9.7)
in the python terminal, run from fcmaes import mode

Error :
zsh: illegal hardware instruction python

Side note :
if I run the python terminal from a bash terminal (instead of zsh), I get
Illegal instruction: 4

If yo have any idea on how to fix this, that would be great :)

fcmaes.optimizer.de_cma throws SystemError and TypeError

While de_cma_py seems to work fine, when I try running de_cma, I get a flood of SystemError: null argument to internal routine on some iteration runs - it's hard to tell where it comes from (I suspect it's from the C++ CMA, and not the DE code). Other times, I'll see a flood of TypeError: object of type 'Process' has no len() errors but this is also intermittent. Do you have any thoughts on where this might be coming from?

Thanks and regards,
Tim

dietmarwo / fast-cma-es Goto Github PK

fast-cma-es's People

Contributors

Stargazers

Watchers

Forkers

fast-cma-es's Issues

Recommend Projects

Recommend Topics

Recommend Org

Jobs