Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

weights question #86

Open
tischi opened this issue Feb 23, 2024 · 3 comments
Open

weights question #86

tischi opened this issue Feb 23, 2024 · 3 comments

Comments

@tischi
Copy link

tischi commented Feb 23, 2024

cfg.tracking_config.division_weight = -0.001

If we want to encourage that divisions are happening, should we put this to a low or high value?

@JoOkuma
Copy link
Member

JoOkuma commented Feb 23, 2024

Hi, @tischi, you should increase it.

When it's zero, it will not penalize divisions.

When positive, the ILP will try to maximize divisions, which is not ideal, and I don't recommend but you're free to do that.

@tischi
Copy link
Author

tischi commented Feb 23, 2024

Thanks @JoOkuma!

What would be the use-case for cfg.tracking_config.division_weight = -0.001? Would that still allow divisions? How would that compare to, e.g. cfg.tracking_config.division_weight = -0.1 and cfg.tracking_config.division_weight = -1?

@JoOkuma
Copy link
Member

JoOkuma commented Feb 23, 2024

@tischi

division_weight = -1 will have fewer divisions than division_weight = -0.1 which will have fewer divisions than division_weight = -0.01, etc.

Be aware that divisions compete with appearing cells, so if division_weight + w_ij < appear_weight, where w_ij is the association score between two cells, then it will favor creating a new cell rather than dividing it, the ILP selects the variables to maximize our objective.

As a rule of thumb, I usually have division_weight be an epsilon smaller than appear_weight, so appearance still takes priority, but a small overlap (w_ij) will make it favor divisions.

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

No branches or pull requests

2 participants