Skip to content
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.

Sign up for GitHub

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

Jump to bottom

Bug: Exploit Structure in get_fantasy_strategy #2494

Open
wants to merge 3 commits into
base: master
Choose a base branch
from
Open

Bug: Exploit Structure in get_fantasy_strategy #2494

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
33662d1
Pass linear operator to fantasy strategy constructor instead of tensor
naefjo Jan 23, 2024
e19a57e
Fix failing test
naefjo Mar 19, 2024
93d87cd
Remove conversion of new_root to dense operator in get_fantasy_strategy
naefjo Mar 19, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
9 changes: 4 additions & 5 deletions gpytorch/models/exact_prediction_strategies.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,6 @@
AddedDiagLinearOperator,
BatchRepeatLinearOperator,
ConstantMulLinearOperator,
DenseLinearOperator,
InterpolatedLinearOperator,
LinearOperator,
LowRankRootAddedDiagLinearOperator,
Expand Down Expand Up @@ -211,8 +210,8 @@ def get_fantasy_strategy(self, inputs, targets, full_inputs, full_targets, full_

# now update the root and root inverse
new_lt = self.lik_train_train_covar.cat_rows(fant_train_covar, fant_fant_covar)
new_root = new_lt.root_decomposition().root.to_dense()
new_covar_cache = new_lt.root_inv_decomposition().root.to_dense()
new_root = new_lt.root_decomposition().root
new_covar_cache = new_lt.root_inv_decomposition().root

# Expand inputs accordingly if necessary (for fantasies at the same points)
if full_inputs[0].dim() <= full_targets.dim():
Expand All @@ -222,7 +221,7 @@ def get_fantasy_strategy(self, inputs, targets, full_inputs, full_targets, full_
full_inputs = [fi.expand(fant_batch_shape + fi.shape) for fi in full_inputs]
full_mean = full_mean.expand(fant_batch_shape + full_mean.shape)
full_covar = BatchRepeatLinearOperator(full_covar, repeat_shape)
new_root = BatchRepeatLinearOperator(DenseLinearOperator(new_root), repeat_shape)
new_root = BatchRepeatLinearOperator(new_root, repeat_shape)
# no need to repeat the covar cache, broadcasting will do the right thing

if isinstance(full_output, MultitaskMultivariateNormal):
Expand All @@ -238,7 +237,7 @@ def get_fantasy_strategy(self, inputs, targets, full_inputs, full_targets, full_
inv_root=new_covar_cache,
)
add_to_cache(fant_strat, "mean_cache", fant_mean_cache)
add_to_cache(fant_strat, "covar_cache", new_covar_cache)
add_to_cache(fant_strat, "covar_cache", new_covar_cache.to_dense())
return fant_strat

@property
Expand Down