EmpiricalCopula usage

[M-Soundouss]

Hello,

I am trying to use the EmpiricalCopula class and I am having some difficulties.

Example code :

from copulae import EmpiricalCopula
import numpy as np

np.random.seed(8)
data = np.random.uniform(size=(3000, 3))
cop = EmpiricalCopula(dim=3, data=data)

print(cop.random(10))  # simulate random number
print(cop.cdf(np.array([[0.1, 0.5, 1]])))

I find that the cop.cdf always gives the same number no matter what I use in the input. Am I using it wrong?

Thanks for writing this package. Best,

[DanielBok]

Hello,

Addressing 2 issues:

1. You see the same random variables because of the seed
2. The cdf is a 1-to-1 transformation so given any input, it should return deterministically the same output

That said, after reviewing the code, I noticed I made a mistake with the random generation in that I should have sampled the data with replacement. I'll fix it in the next update.

[M-Soundouss]

Hello,

Thanks for your response. Sorry I was not clear enough, what I meant is that :

print(cop.cdf(np.array([[0.1, 0.5, 1.]])))

# Gives [0.13366667]

print(cop.cdf(np.array([[0.1, 1., 1.]])))

# Gives [0.13366667]

print(cop.cdf(np.array([[1., 1., 1.]])))

# Gives [0.13366667]

I understand that cdf is deterministic but it does not mean that it should map all possible inputs into 0.13366667, right?

This behavior is different from other types of copulae (like the normal one) where the cdf function gives the output I expected.

[DanielBok]

Yes, you're right. Thank you for pointing out the bug.

The issue arises because I was trying to make it convenient by auto-adjusting the data into pseudo-observations when it is not already so. When I take it away, it'll be correct. I'll release the fix in the next copy.

[M-Soundouss]

Thank you very much! Looking forward to it.

[DanielBok]

@M-Soundouss I pushed up a fix in 0.7.0.

[M-Soundouss]

Thank you very much!