search

olshena / COVIDNearTerm

1 stars 0 forks source link

Add unweighted kernel #16

Closed olshena closed 4 years ago

olshena commented 4 years ago

The 'unweighted' kernel would be like the 'equal' kernel with one difference. Instead of dropping the last value in the window when a new one is added, a random value in the kernel would be dropped. This should be made the default method.

kikapp commented 4 years ago

Added unweighted kernel method which drops a random value from the window instead of the oldest value and made it the default method. Please double check code to see if it makes sense.

olshena commented 4 years ago

I think for the unweighted kernel we should distinguish between the build and predict. In build I think the unweighted kernel should be just like the equal kernel, because we have actual data. I think it is for the prediction where there be a distinction between the two. Sorry for not clarifying.

I also think the unweighted kernel can be more efficient, but that is less of an issue.

kikapp commented 4 years ago

i've reverted the unweighted method change in predictAR.

olshena commented 4 years ago

The 'unweighted' kernel should equal the 'equal' kernel for buildAR, since we have data there. It should differs for the predictAR. Right now we have the opposite where they differ in buildAR but are the same for predictAR.

olshena commented 4 years ago

I think this is correct now.