New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
MMD unbiased estimate issue #4595
Comments
Hi @saeidnp. Thanks for letting us know. I'll check and get back to you on this. |
any updates @lambday ? |
I just tried and I get
I guess there is an issue with your random number generator? |
@karlnapf in your code less samples are used than mine. |
That seems strange to me but I’ll check! |
I get exactly the same magic number. Crazy! This is a bug and I will dig deeper once I have a moment. @lambday might also be able to help |
This is an overflow issue caused by a |
|
@saeidnp can you please confirm whether this closed issue fixed the "magic number" problem? The issue is closed but I don't see a message here saying that the bug is fixed. |
@gbaydin yes it is solved via my hotfix. Could you confirm if you are using it? It (unfortunately) will now a tiny bit slower than before due to the lack of low level vectorization in Eigen3 when summing the kernel matrix, now a c++ loop rather than SIMD summations of the columns. But that should only be noticeable when sampling from the null distribution (repeated computation of the test). We will run benchmarks and potentially try to improve it. |
Great, thanks |
I run the following code which computes MMD unbiased estimate multiple times. Since both the distributions are Normal(0,1) and the estimate is unbiased, I expect the average output to be 0. However, it always prints the exact same number (-0.6577018139068969)
The text was updated successfully, but these errors were encountered: