[feature] Support log-uniform scale in search space definition

[gaocegege]

/kind feature

Describe the solution you'd like [A clear and concise description of what you want to happen.]

https://github.com/kubeflow/katib/pull/1205#issuecomment-639200522

Anything else you would like to add: [Miscellaneous information that will assist in solving the issue.]

[issue-label-bot[bot]]

Issue Label Bot is not confident enough to auto-label this issue. See dashboard for more details.

[issue-label-bot[bot]]

Issue-Label Bot is automatically applying the labels:

Label	Probability
area/front-end	0.98

Please mark this comment with :thumbsup: or :thumbsdown: to give our bot feedback! Links: app homepage, dashboard and code for this bot.

[gaocegege]

/cc @sperlingxx @andreyvelich

[sperlingxx]

@gaocegege In my opinion, we need to add additional field Distribution in FeasibleSpace. The FeasibleSpace definition could be changed to:

type ParameterDistribution string

const (
	Uniform    ParameterDistribution = "uniform"
	LogUniform ParameterDistribution = "log_uniform"
	Normal     ParameterDistribution = "normal"
	LogNormal  ParameterDistribution = "log_normal"
	Discrete   ParameterDistribution = "discrete"
)

type FeasibleSpace struct {
	Max          string                `json:"max,omitempty"`
	Min          string                `json:"min,omitempty"`
	List         []string              `json:"list,omitempty"`
	Step         string                `json:"step,omitempty"`
	Distribution ParameterDistribution `json:"distribution,omitempty"`
}

[gaocegege]

SGTM. I think it is one possible approach.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[andreyvelich]

/lifecycle frozen

[tenzen-y]

/assign

I will submit a small proposal for this feature.

[yanwu2014]

Any updates on this? Would love to be able to search over a log-uniform scale for tuning learning rates

[tenzen-y]

[WIP] Add enhancement for Parameter Distribution #2059

@yanwu2014 Thank you for being interested in this feature. I'm writing the proposal. But we need to cut a new Experiment API version (v1beta2) since we need to introduce significant and breaking changes.

So it will take a little bit of time to release.

[yanwu2014]

That's great, thanks so much :)!

[andreyvelich]

/area gsoc