feat: add down sampling before luis publish

[lei9444]

Description

If the number of intent's utterances more than 10(can be customized) times the least number of the intent's utterances, do a re-sampling process to keep the proportion;

If detect > 15000 utterances, downsize the utterances.

add down sampling config to bot setting

Task Item

refs #2610

Screenshots

[boydc2014]

In terms of putting it into samples, we should give this a meaningful name

downsampling: {
max_imbalance_ratio: 10,
max_utterances_allowed: 15000
}

[lei9444]

In terms of putting it into samples, we should give this a meaningful name

downsampling: {
max_imbalance_ratio: 10,
max_utterance_allowed: 15000
}

Hi @boydc2014 I have add the config to client setting now. user can update it in setting page

[boydc2014]

Looks good to me, can you ask @hibrenda or someone from skill team to have a try?

[github-actions]

Coverage increased (+0.09%) to 41.033% when pulling 248cebe on lei9444:downsampling into 723785b on microsoft:master.

[hibrenda]

I tried this for todo skill. 2 findings.

• down sampling does not only resizing the big ratio intent (eg. ratio is more than 1:10) but also resizing the small ratio intent (eg. 1:8) which should not be happen
• For bot developer's awareness, hope to get the notification after down sampling for some intents

[lei9444]

@hibrenda @boydc2014 I have fix the first commit. For the second, I think we can have more discusses about this.

[hibrenda]

Verified the the fix. now the resizing only for the utterance number bigger than the ratio defined

[cwhitten]

thank you @hibrenda