[Feature Request] Be able to set the preferred "origin" corner

[bbeaulant]

Hello,

Even if it can produce a less optimized result, it would fit to real needs to be able to choose the preferred "origin" corner for irregular problem types.

I changed the directions variable values there and it seems to do the job. Is it the right location to look ?

[fontanf]

Hi,

That's not enough, one needs to change how the leftover value is computed in solutions and nodes.
I'll try to look at this in the next days

[fontanf]

I started the implementation here https://github.com/fontanf/packingsolver/tree/feature/irregular-leftover-corner

The instance parameters have a new parameter leftover_corner.

I haven't tested. Could you add some tests for each corner? You can do this by creating the small instances file (3 simple items), then running the solver to generate the solution, check the solution manually, store the instance and solution files here https://github.com/fontanf/packingsolver/tree/feature/irregular-leftover-corner/data/irregular/tests and add the tests at the end of this file https://github.com/fontanf/packingsolver/blob/feature/irregular-leftover-corner/test/irregular/irregular_test.cpp

[bbeaulant]

Hi,

Wow, nice. But I don't find the leftover_corner parameter on irregular instance parameters. Do I miss something ?

[fontanf]

Here

packingsolver/include/packingsolver/irregular/instance.hpp

Lines 50 to 51 in c88abf1

	/** Reference corner to compute the leftover value. */
	Corner leftover_corner = Corner::BottomLeft;

Are you on the right branch?

[bbeaulant]

Ok, but other options (like item_item_minimum_spacing) use a setter on the instance_builder.
And I think that there's no getter to retrieve parameters object on the instance builder.

[fontanf]

Indeed, sorry, I forgot to add the setter. I should be there now

[bbeaulant]

Good Job ! it seems to work in forth directions. I will add tests later in the afternoon.

[bbeaulant]

But it changes the corner only for the last not full bin, no ?

[fontanf]

Is it really important for the non-last bins as well?

[bbeaulant]

Yes, it is. The use cas of this kind of feature is to align items in one specific corner of a CNC machine = where we align the panel.

[fontanf]

Then it's more complex.

The objective BinPackingWithLeftovers optimizes first the number of bins and then the leftover value.

For the last bin, I can update the way the leftover value is computed to favor a given corner.

But the shape of the non-last bins is not part of the objective. One would need to define a measure of "cornerness" to include in the objective to be able to compare two solutions.

Maybe it would be easier to handle this on your side by shifting the items by (bin_type.x_max - solution_bin.x_max, bin_type.y_max - solution_bin.y_max) assuming the bins are rectangular. Or to implement this as a post-process in the solver

[bbeaulant]

I understand that it's not as simple as you thought.
However, the need expressed here is independent of any objective. It's even a kind of permanent need depending on the user's machine.
For other types of problems, I was able to perform the processing a posteriori since the pieces are rectangles. But for irregular, I can't use this trick. Flipping the solution would be like mirroring the items for the two adjacent corners.

Currently, the items are aligned bottom left. Isn't there already a search order that produces this solution, which can be changed depending on a "corner" parameter?

I'm aware, however, that bins can be something other than rectangles too...

[bbeaulant]

NestingCenter offers for example a parameter to customize the nesting direction :

[fontanf]

For other types of problems, I was able to perform the processing a posteriori since the pieces are rectangles. But for irregular, I can't use this trick. Flipping the solution would be like mirroring the items for the two adjacent corners.

You can translate instead of flipping.

Consider the following edge case where there exists no optimal solution touching a corner.

Is this physically impossible to cut with such machines? so that this instance would necessarily require two bins instead of one?