Caching

[pat-s]

In mlr we use the memoise pkg for caching.
I saw that Tasks are cached in mlr3 already via this PR.

@mllg Would this concept also be preferrable for the filters or does it have any limitations? And should we rely on the memoise pkg for caching?

[mb706]

btw, mlr3 attaches a $hash to most things, so there may be a way to avoid calculating the digest for input tasks (as memoise would probably do it) and speed things up.

[pat-s]

memoise caches the whole function call and all of its arguments. Not sure if the hashed task would have any effect here.

[mb706]

The return object of memoise::cache_memory and memoise::cache_filesystem contain the hash function that the these cache mechanisms should use in the $digest slot. It would be possible to modify this to consider the $hash of input tasks instead of running digest::digest on them.

[pat-s]

@mb706 AFAIR caching is done in pipelines, isn't it? Hence I would close here if we can find/already have an overarching solution.

[mb706]

Caching is on the way to be implemented in pipelines (here), but not there yet. But I agree that it is probably redundant to have filtering here once it is also in pipelines.

[pat-s]

Cool, thanks. Then I won't put more thoughts about this in here 🙂