Skip to content

MLOS lab 3 Leveldb tuning - #129

Merged
Brian Kroth (bpkroth) merged 10 commits into
microsoft:mainfrom
nithinv13:leveldb
Oct 14, 2020
Merged

Brian Kroth (bpkroth) merged 10 commits into
microsoft:mainfrom
nithinv13:leveldb

Conversation

@nithinv13

Copy link
Copy Markdown
Contributor

No description provided.

@ghost

Deleted user (ghost) commented Oct 7, 2020 •

Copy link
Copy Markdown

CLA assistant check
All CLA requirements met.

@amueller

Copy link
Copy Markdown
Contributor

Thanks for the PR. Can you please sign the CLA?

@amueller

Copy link
Copy Markdown
Contributor

As I had mentioned, I think for this example we don't need to use the grpc and can run the optimizer directly as in the Bayesian Optimization notebook.

@nithinv13

nithinv13 commented Oct 8, 2020 via email

Copy link
Copy Markdown
Contributor Author

@nithinv13

nithinv13 commented Oct 8, 2020 via email

Copy link
Copy Markdown
Contributor Author

@bpkroth Brian Kroth (bpkroth) left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@bpkroth

Copy link
Copy Markdown
Contributor

nithinv13 - I fixed the CI issue in #133. Can you please update your branch with the latest from main to force a rerun of the CI pipelines?

Thanks,
Brian

@bpkroth
Brian Kroth (bpkroth) merged commit 00efbf5 into microsoft:main Oct 14, 2020
@bpkroth

Copy link
Copy Markdown
Contributor

Thanks nithinv13 for your contribution!

As discussed, we will add another PR for integrating the notebook to the website menu and refine the notebook for additional "going forward" suggestions.

@bpkroth Brian Kroth (bpkroth) mentioned this pull request Oct 14, 2020
2 of 3 tasks
" \"experiment_designer_config.fraction_random_suggestions\": 0.5,\n",
" \"experiment_designer_config_fraction_random_suggestions\": 0.1\n",
"}.\n",
"10/08/2020 21:58:26 - BayesianOptimizerFactory - INFO - [BayesianOptimizerFactory.py: 40 - create_local_optimizer() ] Creating a bayesian optimizer with config: {\n",

@bpkroth Brian Kroth (bpkroth) Oct 15, 2020 •

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nithinv13 this optimizer creation output seems repeated a number of times. What's up with that?

" \"experiment_designer_config.fraction_random_suggestions\": 0.5,\n",
" \"experiment_designer_config_fraction_random_suggestions\": 0.1\n",
"}.\n",
"10/08/2020 21:58:27 - BayesianOptimizerFactory - INFO - [BayesianOptimizerFactory.py: 40 - create_local_optimizer() ] Creating a bayesian optimizer with config: {\n",

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nithinv13 this optimizer creation output seems repeated a number of times. What's up with that?

"#optimizer = bayesian_optimizer_factory.create_remote_optimizer(optimization_problem=optimization_problem)\n",
"def run_optimizer():\n",
" # Parameter 1: write_buffer_size: min_value = 1 MB, max_value = 128 MB\n",
" # Parameter 2: max_file_size: min_value = 1 MB, max_value = 128 MB\n",

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nithinv13 as mentioned: although these might be the experimentally chosen values you used, we could consider to expand the range to ensure that a different machine/workload doesn't find something else.

],
"source": [
"# Optimization Problem\n",
"# You might have to change the min and max value based on the objective that you are using \n",

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nithinv13 this note probably needs more details.
Actually, a better way would be to give the ContinuousDimension parameter some better names than objective so that it's more clear that if they're tuning for "latency" they want ranges from (0, n] some time units based on their hardware vs. if they're tuning for "throughput" they want ranges from (0, m] MB/s based on their hardware.

Either way, I think providing slightly higher bounds than are expected is good so that you don't accidentally end up capping the optimization problem's output if your benchmarks somehow exceed that.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants