Integrate over multiple limits at once using Quadpy about quadpy HOT 8 CLOSED

The function is missing a signature: what is the input shape and what is the output shape?

from quadpy.

I would blame the function signature. Answered in your original question on SO.

from quadpy.

Your loss_function is a bit weird: Whenever p is less than a given value, the return value is an array, otherwise it's 0. That doesn't make sense. If your function really is 0-or-1, Gaussian integration isn't well suited to the problem either.

Also, you probably mean to use quadpy.quad instead of quadpy.c1.integrate_adaptive.

I'll release a new quadpy version shortly which will feature better error messages.

from quadpy.

Thanks for checking. The loss function works for the cubature-version. How can I rewrite it in such a way it also works for quadpy?

from quadpy.

Thanks for checking. The loss function works for the cubature-version. How can I rewrite it in such a way it also works for quadpy?

Let's see the working variant.

from quadpy.

Thanks for checking. The loss function works for the cubature-version. How can I rewrite it in such a way it also works for quadpy?

Let's see the working variant.

The code can be found in the first post of this topic. Here it is

from cubature import cubature
import numpy as np


class Kelly:
    def __init__(self):
        self.odds = 1.952
        self.kelly = 0.08961344537815132
        self.i = 0.001
        self.f = np.arange(0, 1 + self.i, self.i).flatten()
        self.c1 = 1
        self.c2 = 2
        self.k = 1.5

    def loss_function(self, p):
        p = p[:, 0]
        loss_function = np.where(np.less(p[:, None], (self.f * self.odds - 1) + 1 / self.odds),
                                (self.c1 + self.c2) * abs(self.f - self.kelly) ** self.k, 
                                 0)
        return loss_function

    def integrate_vec(self):
        xmin = np.zeros(len(self.f))
        xmax = np.array([self.f * (self.odds - 1) + 1 / self.odds]).flatten()
        vals, errors = cubature(func=self.loss_function,
                                ndim=1,
                                fdim=len(self.f),
                                xmin=xmin,
                                xmax=xmax,
                                vectorized=True)
        return vals, errors


kelly = Kelly()
vals, errors = kelly.integrate_vec()
print(vals)
# [0.         0.         0.         ... 1.3305999  1.3327977  1.33499671]

from quadpy.

The function is missing a signature: what is the input shape and what is the output shape?

The input- and output shape are equal to (1, len(self.f))

from quadpy.

I just spent some time looking at the functions and it's not entirely clear to me what you're trying to do mathematically. Your loss_function gets a vector of shape (n,), and returns an array of shape (n, 1001). Those are the characteristics of a vector-valued function on a 1D interval. Is this correct?

I've also improved the error messages; they should be more helpful for your use case now.

from quadpy.