fcmaes 1.3.17__py3-none-any.whl → 1.6.9__py3-none-any.whl

fcmaes/pgpecpp.py

@@ -0,0 +1,340 @@
+# Copyright (c) Dietmar Wolz.
+#
+# This source code is licensed under the MIT license found in the
+# LICENSE file in the root directory.
+
+"""
+Eigen based implementation of PGPE see http://mediatum.ub.tum.de/doc/1099128/631352.pdf .
+Derived from https://github.com/google/evojax/blob/main/evojax/algo/pgpe.py .
+"""
+
+import sys
+import os
+import math
+import ctypes as ct
+import numpy as np
+from numpy.random import PCG64DXSM, Generator
+from scipy.optimize import OptimizeResult, Bounds
+from fcmaes.evaluator import _check_bounds, _get_bounds, callback_par, parallel, call_back_par, libcmalib
+
+from typing import Optional, Callable, Union
+from numpy.typing import ArrayLike
+
+os.environ['MKL_DEBUG_CPU_TYPE'] = '5'
+
+def minimize(fun: Callable[[ArrayLike], float],
+             bounds: Optional[Bounds] = None,
+             x0: Optional[ArrayLike] = None,
+             input_sigma: Optional[Union[float, ArrayLike, Callable]] = 0.1,
+             popsize: Optional[int] = 32,
+             max_evaluations: Optional[int] = 100000,
+             workers: Optional[int] = None,
+             stop_fitness: Optional[float] = -np.inf,
+             rg: Optional[Generator] = Generator(PCG64DXSM()),
+             runid: Optional[int] = 0,
+             normalize: Optional[bool] = True,
+             lr_decay_steps: Optional[int] = 1000,
+             use_ranking: Optional[bool] = True,
+             center_learning_rate: Optional[float] = 0.15,
+             stdev_learning_rate: Optional[float] = 0.1,
+             stdev_max_change: Optional[float] = 0.2,
+             b1: Optional[float] = 0.9,
+             b2: Optional[float] = 0.999,
+             eps: Optional[float] = 1e-8,
+             decay_coef: Optional[float] = 1.0,
+             ) -> OptimizeResult:
+       
+    """Minimization of a scalar function of one or more variables using a 
+    C++ PGPE implementation called via ctypes.
+     
+    Parameters
+    ----------
+    fun : callable
+        The objective function to be minimized.
+            ``fun(x) -> float``
+        where ``x`` is an 1-D array with shape (dim,)
+    bounds : sequence or `Bounds`, optional
+        Bounds on variables. There are two ways to specify the bounds:
+            1. Instance of the `scipy.Bounds` class.
+            2. Sequence of ``(min, max)`` pairs for each element in `x`. None
+               is used to specify no bound.
+    x0 : ndarray, shape (dim,)
+        Initial guess. Array of real elements of size (dim,),
+        where 'dim' is the number of independent variables.  
+    input_sigma : float or np.array, optional
+        Initial step size.
+    popsize = int, optional
+        CMA-ES population size.
+    max_evaluations : int, optional
+        Forced termination after ``max_evaluations`` function evaluations.
+    workers : int or None, optional
+        If workers > 1, function evaluation is performed in parallel for the whole population. 
+        Useful for costly objective functions but is deactivated for parallel retry.  
+    stop_fitness : float, optional 
+         Limit for fitness value. If reached minimize terminates.
+    rg = numpy.random.Generator, optional
+        Random generator for creating random guesses.
+    runid : int, optional
+        id used by the is_terminate callback to identify the CMA-ES run.     
+    normalize : boolean, optional
+        if true pheno -> geno transformation maps arguments to interval [-1,1] 
+    lr_decay_steps int, optional - ADAM optimizer configuration
+    use_ranking : boolean, optional - Should we treat the fitness as rankings or not.       
+    center_learning_rate : float, optional - Learning rate for the Gaussian mean.
+    stdev_learning_rate : float, optional - Learning rate for the Gaussian stdev.
+    init_stdev : float, optional - Initial stdev for the Gaussian distribution.
+    stdev_max_change : float, optional - Maximum allowed change for stdev in abs values.    
+    b1 : float, optional - ADAM optimizer configuration
+    b2 : float, optional - ADAM optimizer configuration
+    eps : float, optional - ADAM optimizer configuration
+    decay_coef : float, optional - ADAM optimizer configuration
+        
+    Returns
+    -------
+    res : scipy.OptimizeResult
+        The optimization result is represented as an ``OptimizeResult`` object.
+        Important attributes are: ``x`` the solution array, 
+        ``fun`` the best function value, 
+        ``nfev`` the number of function evaluations,
+        ``nit`` the number of CMA-ES iterations, 
+        ``status`` the stopping critera and
+        ``success`` a Boolean flag indicating if the optimizer exited successfully. """
+    
+    lower, upper, guess = _check_bounds(bounds, x0, rg)      
+    dim = guess.size   
+    if popsize is None:
+        popsize = 32      
+    if popsize % 2 == 1: # requires even popsize
+        popsize += 1
+    if lower is None:
+        lower = [0]*dim
+        upper = [0]*dim
+    if callable(input_sigma):
+        input_sigma=input_sigma()
+    if np.ndim(input_sigma) == 0:
+        input_sigma = [input_sigma] * dim 
+    parfun = None if (workers is None or workers <= 1) else parallel(fun, workers)
+    array_type = ct.c_double * dim   
+    c_callback_par = call_back_par(callback_par(fun, parfun))
+    res = np.empty(dim+4)
+    res_p = res.ctypes.data_as(ct.POINTER(ct.c_double))
+    try:
+        optimizePGPE_C(runid, c_callback_par, dim, array_type(*guess), 
+                array_type(*lower), array_type(*upper), 
+                array_type(*input_sigma), max_evaluations, stop_fitness,
+                popsize, int(rg.uniform(0, 2**32 - 1)), 
+                lr_decay_steps, use_ranking, center_learning_rate,
+                stdev_learning_rate, stdev_max_change, b1, b2, eps, decay_coef,
+                normalize, res_p)
+        x = res[:dim]
+        val = res[dim]
+        evals = int(res[dim+1])
+        iterations = int(res[dim+2])
+        stop = int(res[dim+3])
+        res = OptimizeResult(x=x, fun=val, nfev=evals, nit=iterations, status=stop, success=True)
+    except Exception as ex:
+        res = OptimizeResult(x=None, fun=sys.float_info.max, nfev=0, nit=0, status=-1, success=False)
+    if not parfun is None:
+        parfun.stop()
+    return res
+
+class PGPE_C:
+
+    def __init__(self,
+        dim: int,
+        bounds: Optional[Bounds] = None,
+        x0: Optional[ArrayLike] = None,
+        input_sigma: Optional[Union[float, ArrayLike, Callable]] = 0.1,
+        popsize: Optional[int] = 32,
+        rg: Optional[Generator] = Generator(PCG64DXSM()),
+        runid: Optional[int] = 0,
+        normalize: Optional[bool] = True,
+        lr_decay_steps: Optional[int] = 1000,
+        use_ranking: Optional[bool] = False, 
+        center_learning_rate: Optional[float] = 0.15,
+        stdev_learning_rate: Optional[float] = 0.1, 
+        stdev_max_change: Optional[float] = 0.2, 
+        b1: Optional[float] = 0.9,
+        b2: Optional[float] = 0.999, 
+        eps: Optional[float] = 1e-8, 
+        decay_coef: Optional[float] = 1.0, 
+        ):
+       
+        """Minimization of a scalar function of one or more variables using a 
+        C++ CR-FM-NES implementation called via ctypes.
+         
+        Parameters
+        ----------
+        dim : int
+            dimension of the argument of the objective function
+        bounds : sequence or `Bounds`, optional
+            Bounds on variables. There are two ways to specify the bounds:
+                1. Instance of the `scipy.Bounds` class.
+                2. Sequence of ``(min, max)`` pairs for each element in `x`. None
+                   is used to specify no bound.
+        x0 : ndarray, shape (dim,)
+            Initial guess. Array of real elements of size (dim,),
+            where 'dim' is the number of independent variables.  
+        input_sigma : float, optional
+            Initial step size.
+        popsize = int, optional
+            CMA-ES population size.
+        max_evaluations : int, optional
+            Forced termination after ``max_evaluations`` function evaluations.
+        workers : int or None, optional
+            If workers > 1, function evaluation is performed in parallel for the whole population. 
+            Useful for costly objective functions but is deactivated for parallel retry.  
+        stop_fitness : float, optional 
+             Limit for fitness value. If reached minimize terminates.
+        rg = numpy.random.Generator, optional
+            Random generator for creating random guesses.
+        runid : int, optional
+            id used by the is_terminate callback to identify the CMA-ES run.     
+        normalize : boolean, optional
+            if true pheno -> geno transformation maps arguments to interval [-1,1]
+        lr_decay_steps int, optional - ADAM optimizer configuration
+        use_ranking : boolean, optional - Should we treat the fitness as rankings or not.       
+        center_learning_rate : float, optional - Learning rate for the Gaussian mean.
+        stdev_learning_rate : float, optional - Learning rate for the Gaussian stdev.
+        init_stdev : float, optional - Initial stdev for the Gaussian distribution.
+        stdev_max_change : float, optional - Maximum allowed change for stdev in abs values.    
+        b1 : float, optional - ADAM optimizer configuration
+        b2 : float, optional - ADAM optimizer configuration
+        eps : float, optional - ADAM optimizer configuration
+        decay_coef : float, optional - ADAM optimizer configuration"""
+
+        lower, upper, guess = _get_bounds(dim, bounds, x0, rg)      
+        if popsize is None:
+            popsize = 32      
+        if popsize % 2 == 1: # requires even popsize
+            popsize += 1
+        if lower is None:
+            lower = [0]*dim
+            upper = [0]*dim
+        if callable(input_sigma):
+            input_sigma=input_sigma()
+        if np.ndim(input_sigma) == 0:
+            input_sigma = [input_sigma] * dim 
+        array_type = ct.c_double * dim   
+        try:
+            self.ptr = initPGPE_C(runid, dim, array_type(*guess), 
+                           array_type(*lower), array_type(*upper), 
+                    array_type(*input_sigma), popsize, int(rg.uniform(0, 2**32 - 1)),
+                    lr_decay_steps, use_ranking, center_learning_rate,
+                    stdev_learning_rate, stdev_max_change, b1, b2, eps, decay_coef, 
+                    normalize)
+            self.popsize = popsize
+            self.dim = dim            
+        except Exception as ex:
+            print (ex)
+            pass
+    
+    def __del__(self):
+        destroyPGPE_C(self.ptr)
+            
+    def ask(self) -> np.array:
+        """ask for popsize new argument vectors.
+            
+        Returns
+        -------
+        xs : popsize sized list of dim sized argument lists."""
+        
+        try:
+            lamb = self.popsize
+            n = self.dim
+            res = np.empty(lamb*n)
+            res_p = res.ctypes.data_as(ct.POINTER(ct.c_double))
+            askPGPE_C(self.ptr, res_p)
+            xs = np.empty((lamb, n))
+            for p in range(lamb):
+                xs[p,:] = res[p*n : (p+1)*n]
+            return xs
+        except Exception as ex:
+            print (ex)
+            return None
+
+    def tell(self, 
+             ys: np.ndarray) -> int:      
+        """tell function values for the argument lists retrieved by ask().
+        
+        Parameters
+        ----------
+        ys : popsize sized list of function values 
+ 
+        Returns
+        -------
+        stop : int termination criteria, if != 0 loop should stop."""        
+        
+        try:
+            array_type_ys = ct.c_double * len(ys)
+            return tellPGPE_C(self.ptr, array_type_ys(*ys))
+        except Exception as ex:
+            print (ex)
+            return -1        
+
+    def population(self) -> np.array:
+        try:
+            lamb = self.popsize
+            n = self.dim
+            res = np.empty(lamb*n)
+            res_p = res.ctypes.data_as(ct.POINTER(ct.c_double))
+            populationPGPE_C(self.ptr, res_p)
+            xs = np.array(lamb, n)
+            for p in range(lamb):
+                xs[p] = res[p*n : (p+1)*n]
+                return xs
+        except Exception as ex:
+            print (ex)
+            return None
+
+    def result(self) -> OptimizeResult:
+        res = np.empty(self.dim+4)
+        res_p = res.ctypes.data_as(ct.POINTER(ct.c_double))
+        try:
+            resultPGPE_C(self.ptr, res_p)
+            x = res[:self.dim]
+            val = res[self.dim]
+            evals = int(res[self.dim+1])
+            iterations = int(res[self.dim+2])
+            stop = int(res[self.dim+3])
+            res = OptimizeResult(x=x, fun=val, nfev=evals, nit=iterations, status=stop, success=True)
+        except Exception as ex:
+            res = OptimizeResult(x=None, fun=sys.float_info.max, nfev=0, nit=0, status=-1, success=False)
+        return res
+
+if not libcmalib is None: 
+
+    optimizePGPE_C = libcmalib.optimizePGPE_C
+    optimizePGPE_C.argtypes = [ct.c_long, call_back_par, ct.c_int, \
+                ct.POINTER(ct.c_double), ct.POINTER(ct.c_double), ct.POINTER(ct.c_double), \
+                ct.POINTER(ct.c_double), ct.c_int, ct.c_double, ct.c_int, 
+                ct.c_long, ct.c_int, 
+                ct.c_bool, ct.c_double, ct.c_double, ct.c_double, 
+                ct.c_double, ct.c_double, ct.c_double, ct.c_double, 
+                ct.c_bool, ct.POINTER(ct.c_double)]
+                
+    initPGPE_C = libcmalib.initPGPE_C
+    initPGPE_C.argtypes = [ct.c_long, ct.c_int, ct.POINTER(ct.c_double), 
+                ct.POINTER(ct.c_double), ct.POINTER(ct.c_double), ct.POINTER(ct.c_double), 
+                ct.c_int, ct.c_long, ct.c_int, 
+                ct.c_bool, ct.c_double, ct.c_double, ct.c_double, 
+                ct.c_double, ct.c_double, ct.c_double, ct.c_double, 
+                ct.c_bool]
+    
+    initPGPE_C.restype = ct.c_void_p   
+    
+    destroyPGPE_C = libcmalib.destroyPGPE_C
+    destroyPGPE_C.argtypes = [ct.c_void_p]
+    
+    askPGPE_C = libcmalib.askPGPE_C
+    askPGPE_C.argtypes = [ct.c_void_p, ct.POINTER(ct.c_double)]
+    
+    tellPGPE_C = libcmalib.tellPGPE_C
+    tellPGPE_C.argtypes = [ct.c_void_p, ct.POINTER(ct.c_double)]
+    tellPGPE_C.restype = ct.c_int
+    
+    populationPGPE_C = libcmalib.populationPGPE_C
+    populationPGPE_C.argtypes = [ct.c_void_p, ct.POINTER(ct.c_double)]
+    
+    resultPGPE_C = libcmalib.resultPGPE_C
+    resultPGPE_C.argtypes = [ct.c_void_p, ct.POINTER(ct.c_double)]

fcmaes/pygmoretry.py

@@ -6,12 +6,12 @@
 import math
 import os
 import sys
-from numpy.random import Generator, MT19937, SeedSequence
+import numpy as np
+from numpy.random import Generator, PCG64DXSM, SeedSequence
 from scipy.optimize import OptimizeResult, Bounds
 import multiprocessing as mp
 from multiprocessing import Process
 from fcmaes.retry import Store
-from fcmaes.optimizer import logger
 
 os.environ['MKL_DEBUG_CPU_TYPE'] = '5'
 os.environ['MKL_NUM_THREADS'] = '1'
@@ -19,12 +19,11 @@ os.environ['OPENBLAS_NUM_THREADS'] = '1'
 
 def minimize(prob, 
              algo,
-             value_limit = math.inf,
+             value_limit = np.inf,
              num_retries = 100*mp.cpu_count(),
-             logger = None,
              workers = mp.cpu_count(),
              popsize = 1, 
-             ):   
+             ) -> OptimizeResult:   
     """Minimization of a scalar function of one or more variables using parallel retry.
        Similar to fcmaes.retry but works with pygmo / pagmo problems + algorithms.
        For problems with equality/inequality contraints or multiple objectives fcmaes.retry cannot 
@@ -44,10 +43,6 @@ def minimize(prob,
         Upper limit for optimized function values to be stored. 
     num_retries : int, optional
         Number of optimization retries.    
-    logger : logger, optional
-        logger for log output of the retry mechanism. If None, logging
-        is switched off. Default is a logger which logs both to stdout and
-        appends to a file ``optimizer.log``.
     workers : int, optional
         number of parallel processes used. Default is mp.cpu_count()
     popsize = int, optional
@@ -63,30 +58,26 @@ def minimize(prob,
 
     lb, ub = prob.get_bounds()
     bounds = Bounds(lb, ub)
-    store = Store(bounds, logger = logger)
+    store = Store(bounds)
     return retry(store, prob, algo, num_retries, value_limit, popsize, workers)
                  
-def retry(store, prob, algo, num_retries, value_limit = math.inf, popsize=1, workers=mp.cpu_count()):
+def retry(store, prob, algo, num_retries, value_limit = np.inf, popsize=1, workers=mp.cpu_count()):
     try:
         import pygmo as pg
     except ImportError as e:
         raise ImportError("Please install PYGMO (pip install pygmo) to use PAGMO optimizers") from e
     sg = SeedSequence()
-    rgs = [Generator(MT19937(s)) for s in sg.spawn(workers)]
+    rgs = [Generator(PCG64DXSM(s)) for s in sg.spawn(workers)]
     proc=[Process(target=_retry_loop,
             args=(pid, rgs, store, prob, algo, num_retries, value_limit, popsize, pg)) for pid in range(workers)]
     [p.start() for p in proc]
     [p.join() for p in proc]
-    store.sort()
+    store.sort(store.get_xs())
     store.dump()
     return OptimizeResult(x=store.get_x_best(), fun=store.get_y_best(), 
                           nfev=store.get_count_evals(), success=True)
         
 def _retry_loop(pid, rgs, store, prob, algo, num_retries, value_limit, popsize, pg):
-
-   #reinitialize logging config for windows -  multi threading fix
-    if 'win' in sys.platform and not store.logger is None:
-        store.logger = logger()
     
     while store.get_runs_compare_incr(num_retries):      
         try: