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

batch mode with Adam optimization #84

Merged
merged 60 commits into from
Nov 8, 2021
Merged

batch mode with Adam optimization #84

merged 60 commits into from
Nov 8, 2021

Conversation

jlmelville
Copy link
Owner

"Fixes" #83.
Additionally, a new epoch_callback parameter, also stolen from inspired by https://github.com/ltla/umappp

@jlmelville
extract in-place updating
7aa94ab
@jlmelville
use decltype in more places
7045e10
@jlmelville
move if-else blocks into factory
237aa22
@jlmelville
fix incorrect repulsion index
a2e4c93
@jlmelville
Remove un-needed vector
2ccc54f
@jlmelville
Extract head/tail coordinate handling
7c86e0b
@jlmelville
move Coords to uwot namespace
639ed9d
@jlmelville
Remove un-needed setters
be25d20
@jlmelville
remove un-used field
49e9442
@jlmelville
render unto gradient that which belongs to gradient
f7d08b7
@jlmelville
Move actual gradient calc into gradient header
b794dd0
@jlmelville
hide more dist/gradient calcs from opt loop
9e3b842
@jlmelville
move more update code into its own function
9bff860
@jlmelville
sgd loop doesn't need coords, update always knows
cba7311
@jlmelville
whitespace
53ef554
@jlmelville
update uses alpha
eae4ff6
@jlmelville
Rename n_epochs_per_sample_size -> n_edges
80497c2
@jlmelville
no need to move edges
141a862
@jlmelville
New function passing thread_id
5ebb69b
@jlmelville
Initial potentially workable batch implementation
6ea4377
@jlmelville
minor convenience overload
adbefe1
@jlmelville
template specialization for head gradient update
a2fd701
@jlmelville
document some things to myself
0a3934b
@jlmelville
Better layout optimization organization.
442b6be
@jlmelville
pfor overload
76493ce
@jlmelville
Reorg factory, rename workers, batch = FALSE
c5ad3d1
@jlmelville
build node worker for batch = TRUE
f9d6fde
@jlmelville
Move more epoch code into inst
3212a5b
@jlmelville
Add initial doc for batch param
2dcc8d3
@jlmelville
Down-scale the learning rate when transforming
74f235a
@jlmelville
Batch-compatible RNG for thread-independent results
c4aa4f1
@jlmelville
Remove R Progress from optimization loop
d883b17
@jlmelville
Abstract parallel processing
4737d2a
@jlmelville
Create explicit serial processor
24be498
@jlmelville
Store n_threads and grain_size
a51f069
@jlmelville
Split uwot optimization
5d92613
@jlmelville
Rename optimize.cpp
56e7381
@jlmelville
Move generic functions to the top
8a00533
@jlmelville
Move epoch knowledge into sampler
956120b
@jlmelville
remove unused variable, rename tail_nvert for consistency
bbecebb
@jlmelville
Adding "inline" appears to give 15-20% speed up
0a5bc37
@jlmelville
Initial optimizer work
01a6f8d
@jlmelville
create param objects
6037d2a
@jlmelville
Make sampler later + use verbose
d28ebd7
@jlmelville
QHM with varying nu/beta debias
76167aa
@jlmelville
simpler debiasing + initial QHAdam attempt
842a279
@jlmelville
Fix bad debiasing in QHAdam
e1dee3c
@jlmelville
Create sampler twice for batch and non-batch
c147938
@jlmelville
more experimentation
1ee1bff
@jlmelville
remove n_threads/grain_size overrides
65f4bb3
@jlmelville
simpler Adam impl
4609c0d
@jlmelville
epoch callback support
4e2d41e
@jlmelville
Add callback support
92b20fb 
also, remove some unneeded optimizer code
@jlmelville
remove opt_args from transform
4392a0c
@jlmelville
Make callback use a bit more convenient
225d3a1
@jlmelville
param docs for opt_args and epoch_callback
206162b
@jlmelville
Update R-CMD-check
597fabb
@jlmelville
remove mac oldrel, use ubuntu-latest
e8262a5
@jlmelville jlmelville merged commit 1fa8d5c into master Nov 8, 2021
@jlmelville jlmelville deleted the adamopt branch November 8, 2021 05:29
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@jlmelville