Skip to content

Eliminate double dataload loop learning status #105

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

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

Eliminate double dataload loop learning status #105

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1839abc
show status time
casper2002casper Mar 2, 2022
39232c3
single for loop
casper2002casper Mar 2, 2022
4de5a26
Revert "show status time"
casper2002casper Mar 2, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 6 additions & 4 deletions src/learning.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -154,6 +154,7 @@ end
function learning_status(tr::Trainer, samples)
# As done now, this is slighly inefficient as we solve the
# same neural network inference problem twice
samples = Network.convert_input_tuple(tr.network, samples)
W, X, A, P, V = samples
regws = Network.regularized_params(tr.network)
Ls = losses(tr.network, regws, tr.params, tr.Wmean, tr.Hp, samples)
Expand All @@ -167,11 +168,12 @@ end
function learning_status(tr::Trainer)
batchsize = min(tr.params.loss_computation_batch_size, num_samples(tr))
batches = Flux.Data.DataLoader(tr.data; batchsize, partial=true)
reports = map(batches) do batch
batch = Network.convert_input_tuple(tr.network, batch)
return learning_status(tr, batch)
reports = []
ws = []
for batch in batches
push!(reports, learning_status(tr, batch))
push!(ws, sum(batch.W))
end
ws = [sum(batch.W) for batch in batches]
return mean_learning_status(reports, ws)
end

Expand Down