我认为NumPy有点过头了。

[p/10 for p in range(0, 10)]
[0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]

一般来说，要做阶跃1/x直到y

x=100
y=2
[p/x for p in range(0, int(x*y))]
[0.0, 0.01, 0.02, 0.03, ..., 1.97, 1.98, 1.99]

(1/x在我测试时产生的舍入噪声更少)。

令人惊讶的是，在Python 3文档中还没有人提到推荐的解决方案:

参见: linspace菜谱展示了如何实现适用于浮点应用程序的惰性版本的range。

一旦定义，recipe就很容易使用，不需要numpy或任何其他外部库，只需要numpy.linspace()这样的函数。注意，第三个num参数指定所需值的数量，而不是step参数，例如:

print(linspace(0, 10, 5))
# linspace(0, 10, 5)
print(list(linspace(0, 10, 5)))
# [0.0, 2.5, 5.0, 7.5, 10]

下面我引用了Andrew Barnert的完整Python 3配方的修改版本:

import collections.abc
import numbers

class linspace(collections.abc.Sequence):
    """linspace(start, stop, num) -> linspace object

    Return a virtual sequence of num numbers from start to stop (inclusive).

    If you need a half-open range, use linspace(start, stop, num+1)[:-1].
    """
    def __init__(self, start, stop, num):
        if not isinstance(num, numbers.Integral) or num <= 1:
            raise ValueError('num must be an integer > 1')
        self.start, self.stop, self.num = start, stop, num
        self.step = (stop-start)/(num-1)
    def __len__(self):
        return self.num
    def __getitem__(self, i):
        if isinstance(i, slice):
            return [self[x] for x in range(*i.indices(len(self)))]
        if i < 0:
            i = self.num + i
        if i >= self.num:
            raise IndexError('linspace object index out of range')
        if i == self.num-1:
            return self.stop
        return self.start + i*self.step
    def __repr__(self):
        return '{}({}, {}, {})'.format(type(self).__name__,
                                       self.start, self.stop, self.num)
    def __eq__(self, other):
        if not isinstance(other, linspace):
            return False
        return ((self.start, self.stop, self.num) ==
                (other.start, other.stop, other.num))
    def __ne__(self, other):
        return not self==other
    def __hash__(self):
        return hash((type(self), self.start, self.stop, self.num))

下面是我的解决方案，它可以很好地使用float_range(- 1,0,0.01)，并且没有浮点表示错误。它不是很快，但工作得很好:

from decimal import Decimal

def get_multiplier(_from, _to, step):
    digits = []
    for number in [_from, _to, step]:
        pre = Decimal(str(number)) % 1
        digit = len(str(pre)) - 2
        digits.append(digit)
    max_digits = max(digits)
    return float(10 ** (max_digits))


def float_range(_from, _to, step, include=False):
    """Generates a range list of floating point values over the Range [start, stop]
       with step size step
       include=True - allows to include right value to if possible
       !! Works fine with floating point representation !!
    """
    mult = get_multiplier(_from, _to, step)
    # print mult
    int_from = int(round(_from * mult))
    int_to = int(round(_to * mult))
    int_step = int(round(step * mult))
    # print int_from,int_to,int_step
    if include:
        result = range(int_from, int_to + int_step, int_step)
        result = [r for r in result if r <= int_to]
    else:
        result = range(int_from, int_to, int_step)
    # print result
    float_result = [r / mult for r in result]
    return float_result


print float_range(-1, 0, 0.01,include=False)

assert float_range(1.01, 2.06, 5.05 % 1, True) ==\
[1.01, 1.06, 1.11, 1.16, 1.21, 1.26, 1.31, 1.36, 1.41, 1.46, 1.51, 1.56, 1.61, 1.66, 1.71, 1.76, 1.81, 1.86, 1.91, 1.96, 2.01, 2.06]

assert float_range(1.01, 2.06, 5.05 % 1, False)==\
[1.01, 1.06, 1.11, 1.16, 1.21, 1.26, 1.31, 1.36, 1.41, 1.46, 1.51, 1.56, 1.61, 1.66, 1.71, 1.76, 1.81, 1.86, 1.91, 1.96, 2.01]

最佳解决方案:没有舍入误差

>>> step = .1
>>> N = 10     # number of data points
>>> [ x / pow(step, -1) for x in range(0, N + 1) ]

[0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]

或者，对于一组范围而不是一组数据点(例如，连续函数)，使用:

>>> step = .1
>>> rnge = 1     # NOTE range = 1, i.e. span of data points
>>> N = int(rnge / step
>>> [ x / pow(step,-1) for x in range(0, N + 1) ]

[0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]

实现一个函数:将x / pow(step， -1)替换为f(x / pow(step， -1))，并定义f。 例如:

>>> import math
>>> def f(x):
        return math.sin(x)

>>> step = .1
>>> rnge = 1     # NOTE range = 1, i.e. span of data points
>>> N = int(rnge / step)
>>> [ f( x / pow(step,-1) ) for x in range(0, N + 1) ]

[0.0, 0.09983341664682815, 0.19866933079506122, 0.29552020666133955, 0.3894183423086505, 
 0.479425538604203, 0.5646424733950354, 0.644217687237691, 0.7173560908995228,
 0.7833269096274834, 0.8414709848078965]

可以使用Numpy库来完成。Arange()函数允许浮动中的步骤。但是，它返回一个numpy数组，为了方便起见，可以使用tolist()将其转换为列表。

for i in np.arange(0, 1, 0.1).tolist():
   print i

这一行代码不会使代码变得混乱。step参数的符号很重要。

def frange(start, stop, step):
    return [x*step+start for x in range(0,round(abs((stop-start)/step)+0.5001),
        int((stop-start)/step<0)*-2+1)]