Should I use a Custom solver phase to initialize planning variables?

[kentbill]

Hi, Teams,

When I want to initialize planning variables (including genuine and shadow variable) in a dataset with my own logic before commit to solver, what is the difference between using a Custom solver phase or not?

I found that when I commit an initialized dataset to the solver, it skips the CH phase and starts the LS phase immediately.
There is another approach that completes the initialization of planning variables through a Custom solver phase and then starts the LS phase. What is the difference between using a Custom solver phase or not in the dataset initialization?

Thanks in advance！

[triceo]

Typically a custom phase is used when you want to do something between your CH and your LS phases.

If you want to have a completely custom CH and not use the solver's CH implementations, there is no need for you to use a custom phase at all - initialize your solution in whichever way you like, and only give it to the solver initialized. That will skip the CH and go straight into LS.

Our CH implementations exist so that people do not need to write their own. But sometimes, there is a better algorithm which is better suitable for that particular problem. In that case, you really don't need to use our CH, and you can do any initialization before the solver even starts.

[kentbill]

Thank you! @triceo