"Weekly time" is probably not the best metric, because it's spikier than that, dependent on release schedules / vulnerability discovery. Personally, I try to keep dependencies minimal and limited to established projects known to follow semantic versioning. That eliminates most surprises, but there is still the issue of upgrading when the time comes.
I try to keep within one major version of latest, usually only upgrading when it is absolutely necessary. For example, maybe I need a feature in the new version, or maybe I'm about to write a bunch of code that touches the dependency and now is a good time to upgrade. In reality, what that means is I have a slowly building list of chores in the icebox to "upgrade X to 3.0, upgrade Y to 2.0, etc..."