【问题标题】:results of scipy.optimize minimize are offscipy.optimize 最小化的结果关闭
【发布时间】:2020-03-16 08:13:33
【问题描述】:

加载数据:

import numpy as np
import matplotlib.pyplot as plt
from scipy.optimize import minimize
from scipy.special import lambertw
import math

filelist = []
coords=[]

r0 = 0.1095/2 #in um
v0 = 4/3*math.pi*r0**3
ci0 = 0.19
cs = (0.522-ci0)/2+ci0
cw = 55
vf = v0*ci0/(ci0 + cs)


i=0
for fname in glob("*.txt"):
    x,y = np.genfromtxt(fname, unpack=True)
    i+=1
    # normalize
    ynorm = ((y-min(y))/((np.mean(y[-200:]))-min(y)))
    coords.append([x,ynorm])
plt.show()

init_guess=[1, 1, 1, 1, 1]

enter image description here

到目前为止一切顺利。问题是,函数 ssvn 的最小化,它应该用拟合参数 (rr, bsn, dsn, pf, pfn) 拟合函数 intn 到上面显示的曲线,导致值完全关闭。

for n in range(0,i): 

    vol = lambda pf: vf*(1+lambertw((cs/ci0)*np.exp(cs/ci0-(3*pf*(ci0+cs)**2)*(coords[n][0])/(r0*ci0*cw))))
    voln = lambda pfn: vf*(1+lambertw((cs/ci0)*np.exp(cs/ci0-(3*pfn*(ci0+cs)**2)*(coords[n][0])/(r0*ci0*cw))))
    totv = lambda rr, pf, pfn: rr*vol(pf) + (1 - rr)*voln(pfn)
    ifin = np.mean(coords[n][1][-200:])
    asn = lambda bsn, dsn: ifin - bsn*vf + dsn*vf**2
    intn = lambda rr, bsn, dsn, pf, pfn: asn(bsn, dsn) + bsn*totv(rr, pf, pfn) + dsn*(totv(rr, pf, pfn))**2
    def ssvn(var):
        rr, bsn, dsn, pf, pfn = var
        return sum(coords[n][1] - ((intn(rr, bsn, dsn, pf, pfn))**2)[0])
    bnds = ((0, 1), (None, 0), (0,None), (0,100), (0,1000))
    result = (minimize(ssvn, init_guess, bounds=bnds, options={"maxiter":5000}))
    print(result.x)

如果能得到一个提示,为什么在这种情况下最小化不起作用,我将不胜感激!

【问题讨论】:

  • 在不知道 glob("*.txt") 加载了什么的情况下很难说,因为您的示例无法复制。此外,如果“完全关闭的值”,预期结果是什么?你能把它包括进去吗?
  • 对!我已经添加了值。

标签: python function optimization scipy minimization


【解决方案1】:

当然!我已经减少了上面的计算,以便更清楚地解释拟合:

init_guess=[1, 1, 1]

for n in range(0,i): 

    vol = lambda pf: vf*(1+lambertw((cs/ci0)*np.exp(cs/ci0-(3*pf*(ci0+cs)**2)*(coords[n][0])/(r0*ci0*cw))))
    ifin = np.mean(coords[n][1][-200:])
    asn = lambda bsn, dsn: ifin - bsn*vf + dsn*vf**2
    intn = lambda bsn, dsn, pf: asn(bsn, dsn) + bsn*vol(pf) + dsn*(vol(pf))**2
    def ssvn(var):
        bsn, dsn, pf = var
        return sum(coords[n][1] - ((intn(bsn, dsn, pf))**2)[0])
    bnds = ((None, 0), (0,None), (0,100))
    result = (minimize(ssvn, init_guess, bounds=bnds, options={"maxiter":5000}))
    print(result.x)
    plt.plot(coords[n][0], (intn(result.x[0], result.x[1], result.x[2])))

模拟曲线的一些点(用 glob(".txt") 加载)是:

0. 0.000050914985470697376
0.1 0.00005226691119118905
0.2 0.000053273861157071714
0.3 0.00005399970642597728
0.4 0.000054509506785116066
0.5 0.000054860656948798976
0.6 0.00005509912334378207
0.7 0.00005525962182128574
0.8 0.000055366845246536884
0.9 0.0000554381500488244
1. 0.00005548540798339426
1.1 0.00005551666847593382
1.2 0.00005553732736787186
1.3 0.000055550957627270784
1.4 0.00005555995251553653
1.5 0.000055565880705824926
1.6 0.00005556979319469152
1.7 0.000055572367783992574
1.8 0.000055574071506690767
1.9 0.000055575203109605593
2. 0.000055575944218954186

预期的值是: bsn ~ -0.037, dsn ~ 1.8e-05, pf ~ 7

【讨论】:

    猜你喜欢
    • 2019-09-17
    • 1970-01-01
    • 1970-01-01
    • 1970-01-01
    • 2021-04-16
    • 2016-12-03
    • 2019-03-12
    • 1970-01-01
    • 1970-01-01
    相关资源
    最近更新 更多