Hi,

You are right; currently the node dimensions are ignored. We will change this in the next release.

Stoyan

In 2-3 weeks... But we'll try to upload here a preview version with that implented in a couple of days.

Stoyan

Hi,

It seems that something went wrong with this fix.
The distribution is sparser due to this change - but again it brings us back the problem of large distance between boxes.
If in previous posts we've discussed this problem and discovered that running the layout several time solves all problem, now suddenly even running it a few time doesn't solve the problem - the layout stays sparse.

These are the parameters I am using:

AnnealLayout layout = new AnnealLayout();
layout.setIterationsPerStage(50);
layout.setStages(15);
layout.setInitialTemperature(200f);
layout.setTemperatureScale(0.75f);
layout.setArrowLengthFactor(0.001);
layout.setDistributionFactor(40000);
layout.setBoundaryFactor(50000);
layout.setNodeArrowDistFactor(0.35);
layout.setLayoutArea(this.getDocBounds());
layout.setCrossingArrowsCost(100000);
layout.arrange(this);

Any ideas?
Thanks,
Yoav

Hi,

I've tried different parameters. First, changing the setArrowLengthFactor to a value closer to 1 didn't look well - the arrows became much shorter (almost disappearing).
Changing other parameters didn't do well also.
After inverstigating the issues further, it seems that when we don't consider the distance between nodes as the distance between the centers, we may have overlapping. If we do consider that - we have a sparse graph.
Is there a way to deterimine the method that is used? I nean, if we're using the center distance or not, or anything that will allow us to choose what we would prefer to use?

Thanks,
Yoav

Hi,

As I understand, I can determine the arrow length in code using the ArrowLengthFactor parameter as you described.
What currently can't be done is determining if the setDistributionFactor uses node dimensions in its calculations. Apparently this affects the distribution of the nodes.
Could this be determined by the user somehow?

Thanks,
Yoav

The reason to calculate distances from center was to improve the layout.
Layout maladies were:
1) Nodes were sometimes overlapping
2) Graph was sparse (meaning - unconnected nodes were very far from each other). Re-running the algorithm a few times seems to fix it.

A suggestion was made, that maybe using the node's size will improve the layout.
After applying this fix, it seems that we don't have ovelrlapping nodes anymore, but on the other hand the distance between unconnected nodes became much larger, and re-runnnig the layout takes more rime to solve.

So this solution may not fix the layout problem, but it simply fixes one symptom (overlap) and makes one worse (spaces).

This is why I asked if user can decide they don't want to use this solution - since there's a price using it.

Maybe you start with a big document size, and it gets smaller and smaller after each time running the layout?

Stoyan