-
Notifications
You must be signed in to change notification settings - Fork 0
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
translate point estimates to quantile #4
Comments
a little more detail on one possible approach here ... may be a naive way of thinking about it, but what if we used the point estimate from a model prediction along with its standard error to seed a simulated distribution of predicted values? to do so, we would have to assume the predicted values follow a certain distribution ... but from there we could cut the simulated values at relevant quantiles below is some code to do that with a linear model of
|
So the 50th percentile here, the median -- how does that relate to the point estimate? |
Few other approaches in the fable framework:
|
good point. but in this case the 50th percentile from quantile is generated from a random normal distribution that is centered on the point estimate. looks like it is (using objects created from example code above):
|
This section of the FPP3 book is a good primer on forecast distributions, and later on shows how to use bootstrapping to get prediction intervals. https://otexts.com/fpp3/prediction-intervals.html Also, from: https://otexts.com/fpp3/forecasting-using-transformations.html
|
This is demonstrated around f4a58da#diff-da4c2dfe4ffa338da3407b8db34eac89ce01bcfc58032a6688085c6452350d0fR108-R115 |
we have this working (see https://github.com/signaturescience/focustools/blob/master/scratch/fable-submission-mockup-allmetrics.R#L28-L38) the quantiles meet the valid entry file format once bound, formatted with target names, etc. closing for now, though we may need to revisit if/when we implement another kind of model outside of the |
the covid 19 forecast hub requires point estimates and pre-determined quantile values for each target.
the quantile values required are enumerated in the technical readme:
https://github.com/reichlab/covid19-forecast-hub/blob/master/data-processed/README.md#quantile
once we decide on an initial model / targets to pursue, we will need establish how to go about translating a point estimate to quantiles.
The text was updated successfully, but these errors were encountered: