-
Notifications
You must be signed in to change notification settings - Fork 501
Support training models from psql SET commands #1585
Conversation
Minor Decrease in PerformanceBe warned: this PR may have decreased the throughput of the system slightly.
|
Codecov Report
@@ Coverage Diff @@
## master #1585 +/- ##
==========================================
- Coverage 81.75% 81.66% -0.10%
==========================================
Files 739 739
Lines 52045 52112 +67
==========================================
+ Hits 42550 42556 +6
- Misses 9495 9556 +61
Continue to review full report at Codecov.
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks good! I'm just wondering what would happen if we specify something like --train_forecast_model
when we start the system? At least we should make sure that it doesn't crash. Then will the system train the model or just ignore it? I think either way it's probably fine. But we should document what would be the behavior.
Minor Decrease in PerformanceBe warned: this PR may have decreased the throughput of the system slightly.
|
Minor Decrease in PerformanceBe warned: this PR may have decreased the throughput of the system slightly.
|
Documented the behavior. We just ignore it if it is specified during startup. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM!
Minor Decrease in PerformanceBe warned: this PR may have decreased the throughput of the system slightly.
|
Description
PR adds support to perform the training of the OU model, the interference model, and the forecast model over psql by exposing three dedicated
train_X_model
knobs that can be triggered with theSET
command.The following additional settings are introduced (although some could possibly be coalesced) that can also be altered over psql via the
SET
command to adjust the training of the models:interference_model_input_path
: input path to directory for data to train interference model withinterference_model_train_methods
: comma-delimited methods to use for training interference model (i.e., "rf")interference_model_train_timeout
: timeout for training the interference modelinterference_model_pipeline_sample_rate
: sampling rate of pipeline metric OUs for interference modelou_model_input_path
: input path for directory to data for training OU modelou_model_train_methods
: comma-delimited methods to use to train OU model (i.e., "lr,rf")ou_model_train_timeout
: timeout for training the OU model