functools.partial对象的集成

Question

我正在试着做一个随机模拟器。然而，我想在一个确定性的设置中做一些健全的检查，因此需要集成一个ODE系统。具体地说，就是integrand函数中声明的self.stoich_mat和prop_cont的乘积。由于某些原因，被积函数可以工作，输出为[15. -15.]，但scipy的integrate.odeint不起作用，并抛出错误：

[ 15. -15.] #THIS IS THE OUTPUT FROM THE INTEGRAND FUNCTION

Traceback (most recent call last):
  File "C:/Users/Tomek/Documents/MasterThesis/sim_method5.py", line 474, in <module>
    Simulator2.run()
  File "C:/Users/Tomek/Documents/MasterThesis/sim_method5.py", line 455, in run
    full_output=True)
  File "C:\ProgramData\Anaconda2\lib\site-packages\scipy\integrate\odepack.py", line 215, in odeint
    ixpr, mxstep, mxhnil, mxordn, mxords)
TypeError: Cannot cast array data from dtype('O') to dtype('float64') according to the rule 'safe'

我已经构建了MarkoV类来处理不同的模拟类型，因此self.propensity_generator看起来可能有点复杂。下面的代码是仍然显示错误的最小数量。我尝试过使用lambda表达式而不是functools have，但这似乎没有帮助。

代码：

class MarkoV(object):
    def __init__(self,rates,stoich_mat):
        self.stoich_mat=stoich_mat
        self.rates=rates
        self.nr_reactions=rates.shape[1]

    def init(self,init_state,tf,t0=0.0):
        self.state=init_state
        self.init_state=np.copy(init_state)
        self.nr_species=init_state.shape[0]
        self.tf=tf
        self.t_init=t0
        self.propensity_generator()

    def uni_het_bi_mol_const(self,state,t,k,el):
        return k*np.prod(state[el])
    def uni_het_bi_mol_cat_t(self,state,t,k,cat,el):
        return k*np.prod(state[el])*self.init_state[cat][0](t=t)

    def propensity_generator(self):
        self.propensity_funcs=range(self.stoich_mat.shape[1])
        for i,reac in enumerate(self.stoich_mat.T):
            j=np.where(reac<0)[0]
            if isinstance(self.rates[0, i], float):
                k=self.rates[0, i]
                self.propensity_funcs[i] = functools.partial(self.uni_het_bi_mol_const,k=k,el=j)
            else:
                k,cat=self.rates[0,i]
                #HERE THE DETERMINISTIC INPUT IS DEFINED, ESSENTIAL TO MY PROBLEM, BUT MAYBE A HURDLE
                name_func,func_kwargs=self.init_state[cat][0]
                self.init_state[cat][0]=functools.partial(getattr(self,name_func),**func_kwargs)
                #self.propensity_funcs[i]=lambda state,t:self.uni_het_bi_mol_cat_t(state=state,t=t,k=k,cat=cat,el=j) #TRY LAMBDA
                self.propensity_funcs[i] = functools.partial(self.uni_het_bi_mol_cat_t,k=k,cat=cat,el=j)
        self.propensity_funcs=np.array(self.propensity_funcs)

class deterministic(MarkoV):
    def init_sym(self,init_state,tf,species_continuous,t0=0.0):
        self.species_continuous=species_continuous
        self.init(init_state,tf,t0)
        self.species_discrete=[i for i in np.arange(self.nr_species) if i not in self.species_continuous]

    def sinput(self,t,amplitude=6.0,frequency=0.05,offset=1.0):
        return amplitude*np.sin(frequency*t)+amplitude+offset

    def integrand(self,state,t):
        prop_cont=np.zeros(self.nr_reactions)
        for i,func in enumerate(self.propensity_funcs):
            prop_cont[i]=func(state=state,t=t)
        return np.dot(self.stoich_mat[self.species_discrete,:],prop_cont)

    def run(self,t_step=1e-2):
        t=np.arange(self.t_init,self.tf,t_step)

        h=integrate.odeint(func=self.integrand,
                           y0=self.state[self.species_discrete],
                           t=t,
                           full_output=True)

if __name__ == '__main__':
    import scipy.integrate as integrate
    import numpy as np
    import functools
    import numbers

    rates=np.array([(1.2,0),6.0],dtype=object,ndmin=2)
    stoich_mat=np.array([[ 0, 0], # input
                         [-1, 1], # A
                         [ 1,-1]])# A*
    init_state=np.array([("sinput",{"amplitude":2.0,"frequency":0.05,"offset":0.5}),5.0,5.0],ndmin=2).T
    species_continuous=[0]

    Simulator2=deterministic(rates=rates,stoich_mat=stoich_mat)
    Simulator2.init_sym(init_state=init_state,tf=20.0,species_continuous=species_continuous)
    print Simulator2.integrand(state=init_state,t=0)
    print ""
    Simulator2.run()

Answer 1

In [126]: init_state=np.array([("sinput",{"amplitude":2.0,"frequency":0.05,"offs
     ...: et":0.5}),5.0,5.0],ndmin=2).T
In [127]: init_state
Out[127]: 
array([[('sinput', {'frequency': 0.05, 'amplitude': 2.0, 'offset': 0.5})],
       [5.0],
       [5.0]], dtype=object)
In [128]: init_state.shape
Out[128]: (3, 1)
In [129]: species_discrete=[i for i in np.arange(3) if i not in [0]]
In [130]: species_discrete
Out[130]: [1, 2]
In [131]: init_state[species_discrete]
Out[131]: 
array([[5.0],
       [5.0]], dtype=object)
In [132]: _.shape
Out[132]: (2, 1)

据我所知，您正在将odeint中的y0设置为一个(2,1)对象数组。

我怀疑你的代码会运行，或者至少不会引发这个错误，如果它是：

In [133]: init_state[species_discrete].astype(float).ravel()
Out[133]: array([ 5.,  5.])

您还需要确保

Simulator2.integrand(y0, 0)

运行，并返回类似的(2，)数组。