reductionista commented on a change in pull request #462: DL: Add asymmetric
cluster support for fit and evaluate
URL: https://github.com/apache/madlib/pull/462#discussion_r352940338
##########
File path:
src/ports/postgres/modules/deep_learning/madlib_keras_fit_multiple_model.sql_in
##########
@@ -63,11 +63,13 @@ CREATE OR REPLACE FUNCTION
MADLIB_SCHEMA.fit_transition_multiple_model(
model_architecture TEXT,
compile_params TEXT,
fit_params TEXT,
+ dist_key INTEGER,
+ dist_key_mapping INTEGER[],
current_seg_id INTEGER,
- seg_ids INTEGER[],
- images_per_seg INTEGER[],
- gpus_per_host INTEGER,
segments_per_host INTEGER,
+ images_per_seg INTEGER[],
+ use_gpus BOOLEAN,
+ gpus_per_seg INTEGER[],
Review comment:
I notice you changed `gpus_per_host` to `gpus_per_seg` everywhere. I get
that the size of this array is the number of segments, but it bothers me that
we're referring to the number of gpus per host as gpus_per_seg.
To me, gpus_per_seg should only refer to the number of gpus per segment, ie
gpus_per_host / segments_per_host. Each element of this array still represents
the gpus_per_host seen from that segment, so I think `gpus_per_host` would be
better. But maybe there is a 3rd alternative that's better than either of
these?
Personally, I'd be perfectly fine with just calling it `gpus_per_host`, but
I admit it could also potentially lead to some confusion.
See further comments later on for more...
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
