Adding a remove/replace feature

[SamuelBrand1]

Hi there,

I'm now submitting a model to the covid forecast hub that is a thin wrap around AutoGP https://github.com/CDCgov/covid19-forecast-hub/tree/main/model-output/CFA-EpiAutoGP .

At the moment, I'm doing a full refit on the data but obviously I'm aiming to transition to sequential fitting as new covid hospitalisation data arrives in real time.

However, often data gets revised in these settings. This means it would be handy to have the functionality to remove or replace data from the model so I can refit the model on most recent estimates as well latest revisions from data in the past.

In an ideal world I could also change the underlying variance of the observation noise so it isn't uniform along the time series but that might be a feature too far?

[fsaad]

This means it would be handy to have the functionality to remove or replace data

Is the remove_data! function the feature you are looking for?

AutoGP.jl/src/api.jl

Lines 422 to 443 in 11b2b3a

	"""
	remove_data!(model::GPModel, ds::IndexType, y::Vector{<:Real})
	Remove existing observations `ds` from `model`.
	"""
	function remove_data!(model::GPModel, ds::IndexType)
	# Find the data point.
	indexes = findall(x->x in ds, model.ds)
	if length(indexes) == 0
	error("No such time points $(ds).")
	end
	# Append the data.
	deleteat!(model.ds, indexes)
	deleteat!(model.y, indexes)
	# Convert to numeric.
	ds_numeric = Transforms.apply(model.ds_transform, to_numeric.(model.ds))
	y_numeric = Transforms.apply(model.y_transform, model.y)
	# Prepare observations.
	observations = Gen.choicemap((:xs, y_numeric))
	!isnothing(model.config.noise) && (observations[:noise] = trace[:noise])
	# Run SMC step.
	Inference.smc_step!(model.pf_state, (ds_numeric, model.config), observations)
	end

[SamuelBrand1]

Hahaha I completely blanked this from my mind when I was checking your docs! Apols.