Question

为了使建模代码更整洁，我一直在使用namedtuples管理模型参数。我想使用SciPy的并行化implementation of differential evolution使我的模型适合数据，但是我只能使其串行工作。

differential_evolution的文档规定，对于并行优化，目标函数必须“可插入”。在目标函数参数中使用namedtuples似乎违反了这一要求。有没有不完全重写我的建模代码如何处理参数的解决方法？

下面是一个简化的示例。

代码：

from collections import namedtuple
from scipy.optimize import differential_evolution

def rosenbrock(x, par):
    """Rosenbrock function for testing optimization algorithms"""
    return (par.a - x[0])**2 + par.b*(x[1] - x[0]**2)**2

if __name__ == '__main__':
    # Define a namedtuple generator object for creating model parameter namedtuples.
    parameters_nt = namedtuple('parameters', 'a b')

    # Create a model parameter namedtuple with a=2 and b=3 (global minimum at [2, 4]).
    par01 = parameters_nt(2, 3)

    # Define optimization bounds.
    bounds = [(0, 10), (0, 10)]

    # Attempt to optimize in series.
    series_result = differential_evolution(rosenbrock, bounds, args=(par01,))
    print(series_result.x)

    # Attempt to optimize in parallel.
    parallel_result = differential_evolution(rosenbrock, bounds, args=(par01,),
                                             updating='deferred', workers=-1)
    print(parallel_result.x)

程序输出：

[2. 4.]
Traceback (most recent call last):
  File "parallel_test.py", line 23, in <module>
    parallel_result = differential_evolution(rosenbrock, bounds, args=(par01,), updating='deferred', workers=-1)
  File "/home/jack/miniconda3/lib/python3.7/site-packages/scipy/optimize/_differentialevolution.py", line 276, in differential_evolution
    ret = solver.solve()
  File "/home/jack/miniconda3/lib/python3.7/site-packages/scipy/optimize/_differentialevolution.py", line 688, in solve
    self.population)
  File "/home/jack/miniconda3/lib/python3.7/site-packages/scipy/optimize/_differentialevolution.py", line 789, in _calculate_population_energies
    parameters_pop[0:nfevs]))
  File "/home/jack/miniconda3/lib/python3.7/site-packages/scipy/_lib/_util.py", line 412, in __call__
    return self._mapfunc(func, iterable)
  File "/home/jack/miniconda3/lib/python3.7/multiprocessing/pool.py", line 268, in map
    return self._map_async(func, iterable, mapstar, chunksize).get()
  File "/home/jack/miniconda3/lib/python3.7/multiprocessing/pool.py", line 657, in get
    raise self._value
  File "/home/jack/miniconda3/lib/python3.7/multiprocessing/pool.py", line 431, in _handle_tasks
    put(task)
  File "/home/jack/miniconda3/lib/python3.7/multiprocessing/connection.py", line 206, in send
    self._send_bytes(_ForkingPickler.dumps(obj))
  File "/home/jack/miniconda3/lib/python3.7/multiprocessing/reduction.py", line 51, in dumps
    cls(buf, protocol).dump(obj)
_pickle.PicklingError: Can't pickle <class '__main__.parameters'>: attribute lookup parameters on __main__ failed

Answer 1

我修改了代码，以便目标函数将参数作为字典，然后将该字典转换为一个命名元组。

代码

from collections import namedtuple
from scipy.optimize import differential_evolution

def rosenbrock(x, par):
    """Rosenbrock function for testing optimization algorithms"""

    # Convert parameter dictionary to namedtuple.
    par = convert_par_type(par)

    return (par.a - x[0])**2 + par.b*(x[1] - x[0]**2)**2

def convert_par_type(par):
    """converts a parameter namedtuple to a dictionary and vice versa"""
    if type(par)==parameters_nt:
        par = dict(par._asdict())
    elif type(par)==dict:
        par = parameters_nt(**par)
    else:
        raise TypeError
    return par

if __name__ == '__main__':
    # Define a namedtuple factory object for generating model parameter namedtuples.
    parameters_nt = namedtuple('parameters', 'a b')

    # Create a model parameter namedtuple with a=2 and b=3 (global minimum at [2, 4]).
    par01 = parameters_nt(2, 3)

    # Convert model parameter namedtuple to dictionary.
    par02 = convert_par_type(par01)

    # Define optimization bounds.
    bounds = [(0, 10), (0, 10)]

    # Attempt to optimize in series.
    series_result = differential_evolution(rosenbrock, bounds, args=(par02,))
    print(series_result.x)

    # Attempt to optimize in parallel.
    parallel_result = differential_evolution(rosenbrock, bounds, args=(par02,),
                                             updating='deferred', workers=-1)
    print(parallel_result.x)

输出

[2. 4.]
[2. 4.]

如果目标函数具有namedtuple参数，则scipy.optimize.differential_evolution无法并行运行

1 个答案: