1.
Some features, such as the link displacement, are computed on the fly, and are not serialized. But they might be interesting for other softwares that can read the mastodon file format and do not want to recompute them.
This PR introduces facilities to serialize them, and use them on the link displacement and velocity features.
2.
The de/serialization process is a bit complex in the case of branch features. We serialize only the core model, not
the branch graph. So we get access to a map the maps the core graph object (spots ot links) to file ID. But the
features are defined over a branch spot or a branch link.
There is now a hierarchy to can de/serialize the branch features.
1. Some features, such as the link displacement, are computed on the fly, and are not serialized. But they might be interesting for other softwares that can read the mastodon file format and do not want to recompute them.
This PR introduces facilities to serialize them, and use them on the link displacement and velocity features.
2. The de/serialization process is a bit complex in the case of branch features. We serialize only the core model, not the branch graph. So we get access to a map the maps the core graph object (spots ot links) to file ID. But the features are defined over a branch spot or a branch link.
There is now a hierarchy to can de/serialize the branch features.