OverflowError:数学范围错误 - 指数Python

时间:2017-09-17 04:44:28

标签: python artificial-intelligence

我正在尝试使用python创建一个简单的模拟退火搜索,但在使用math.exp计算指数时总是显示溢出错误。

这是我在python中的代码:

import random
import math

def getF(x1, x2):
    t1 = (4 - (2.1 * (x1 ** 2) + (x1 ** 4) / 3)) * (x1 ** 2)
    t2 = x1 * x2
    t3 = (-4 + (4 * (x2 ** 2))) * (x2 ** 2)
    t = t1 + t2 + t3
    return t


def getP(dE, t):
    return math.exp((-1*dE)/t)


def getNewRandInRange(x):
    newX = x + random.randint(-5, 5)
    while (newX > 10) or (newX < -10):
        newX = x + random.randint(-5, 5)
    return newX

initState1 = random.randint(-10, 10)
initState2 = random.randint(-10, 10)

currentState1 = initState1
currentState2 = initState2

BSF = getF(currentState1, currentState2)

T = 1000
Tmin = 1

while T > Tmin:
    print("T= %f" %T)
    newState1 = getNewRandInRange(currentState1)
    newState2 = getNewRandInRange(currentState2)

    currentF = getF(currentState1, currentState2)
    newF = getF(newState1, newState2)

    print("Current X1= %f" % currentState1)
    print("Current X2= %f" % currentState2)
    print("New X1= %f" % newState1)
    print("New X2= %f" % newState2)

    dE = currentF - newF
    print ("delta E: %f" %dE)

    if dE > 0:
        currentState1 = newState1
        currentState2 = newState2
        BSF = getF(newState1, newState2)
    else:
        randNumber = random.uniform(0, 1)
        p = getP(dE, T)
        if (randNumber < p):
            currentState1 = newState1
            currentState2 = newState2

    print("BSF: %f" %BSF)
    print("\n\n\n")
    T = T * 0.9

print(BSF) #final output

错误消息:

 Traceback (most recent call last): 
    return math.exp((-1*dE)/t)
OverflowError: math range error

我正在尝试使用try和catch但它不会返回指数编号,这会使结果出现问题,而且我正在尝试使用Google搜索,但找不到符合我要求的任何解决方案。

以前谢谢!

1 个答案:

答案 0 :(得分:0)

  

异常OverflowError

当算术运算的结果太大而无法表示时引发。对于长整数(这不会引起MemoryError而不是放弃)和大多数使用普通整数的操作都不会发生这种情况,而整数会返回一个长整数。由于C中缺少浮点异常处理的标准化,因此大多数浮点运算也不会被检查。 Ref

你试图计算一个大数字(大于710),这超出了双倍范围。

您可以使用try/except来处理它:

def getP(dE, t):
    try:
        return math.exp((-1*dE)/t)
    except:
        return -1 # or anything else :D

您可以在Python的代码中找到此评论:

/*
 * For the sake of simplicity and correctness, we impose an artificial
 * limit on ndigits, the total number of hex digits in the coefficient
 * The limit is chosen to ensure that, writing exp for the exponent,
 *
 *   (1) if exp > LONG_MAX/2 then the value of the hex string is
 *   guaranteed to overflow (provided it's nonzero)
 *
 *   (2) if exp < LONG_MIN/2 then the value of the hex string is
 *   guaranteed to underflow to 0.
 *
 *   (3) if LONG_MIN/2 <= exp <= LONG_MAX/2 then there's no danger of
 *   overflow in the calculation of exp and top_exp below.
 *
 * More specifically, ndigits is assumed to satisfy the following
 * inequalities:
 *
 *   4*ndigits <= DBL_MIN_EXP - DBL_MANT_DIG - LONG_MIN/2
 *   4*ndigits <= LONG_MAX/2 + 1 - DBL_MAX_EXP
 *
 * If either of these inequalities is not satisfied, a ValueError is
 * raised.  Otherwise, write x for the value of the hex string, and
 * assume x is nonzero.  Then
 *
 *   2**(exp-4*ndigits) <= |x| < 2**(exp+4*ndigits).
 *
 * Now if exp > LONG_MAX/2 then:
 *
 *   exp - 4*ndigits >= LONG_MAX/2 + 1 - (LONG_MAX/2 + 1 - DBL_MAX_EXP)
 *                    = DBL_MAX_EXP
 *
 * so |x| >= 2**DBL_MAX_EXP, which is too large to be stored in C
 * double, so overflows.  If exp < LONG_MIN/2, then
 *
 *   exp + 4*ndigits <= LONG_MIN/2 - 1 + (
 *                      DBL_MIN_EXP - DBL_MANT_DIG - LONG_MIN/2)
 *                    = DBL_MIN_EXP - DBL_MANT_DIG - 1
 *
 * and so |x| < 2**(DBL_MIN_EXP-DBL_MANT_DIG-1), hence underflows to 0
 * when converted to a C double.
 *
 * It's easy to show that if LONG_MIN/2 <= exp <= LONG_MAX/2 then both
 * exp+4*ndigits and exp-4*ndigits are within the range of a long.
 */

无论如何,您可以使用Decimal

import decimal
...
def getP(dE, t):
    return decimal.Decimal((-1*dE)/t).exp()