Is there a way to unfreeze a frozen instance?

[volkerjaenisch]

Dear @diegojromerolopez

Thank you for this nice piece of software!
I need a frozen instance that I later can make a unfrozen (deep-)copy of. This is used for IMHO typical (scientific) data processing chains:

1. Calculate a instance of an intermediate result A and store it frozen (e.g. load a dataset)
2. Calculate a result B that depend on A but shares the same metadata (e.g. b = filter(A), or B= smooth(A))
3. Store B frozen
   ...

In your documentation you stated that (deep-)copy of a frozen instance is not possible.

So I see two ways to achieve this goal:

1. There may be a way to unfreeze a frozen instance?
2. There may be a way to access the original instance below the frozen structural sharing?

Any help appreciated

Cheers,
Volker

[diegojromerolopez]

Hello volkerjaenisch

Glad you find this project interesting and useful!

At the moment there is no way to unfreeze an instance, nor a way to get the original instance. I suppose you could have a side-effect in the on_freeze function, but that's not an elegant solution at all.

However, it's an interesting feature so I'm going to take a look and see if I can have a solution for this in a couple of days.

Best regards.

[volkerjaenisch]

Hello @diegojromerolopez
Thank you very much for the fast response. I would be really happy if a solution could be found.
I am staying tuned.
Volker

[diegojromerolopez]

@volkerjaenisch I have a possible solution for your problem in this PR: #22 please, check it out and see if it suits your needs.

Otherwise, could you give me a mock example of the pipeline you're trying to implement (for example using a toy dataset)? That way I could have a more clear idea of your use-case.

[volkerjaenisch]

@diegojromerolopez
Perfect!
I will checkout your code tomorrow
Volker

[diegojromerolopez]

Nice, keep me posted!

[volkerjaenisch]

@diegojromerolopez
It seems to work.

But I learned that your code does not work inplace (build a readonly wrapper around the given instance) but returns a copy which is then made readonly. Sorry for not having understood that before I asked you. Since in that case of a not inplace operation we could use the original version of gelidum.

Also I noticed that you code does not support numpy arrays which I will open a separate issue for.

Sorry for the not so positive feedback. I really appreciate your effort

Volker

[diegojromerolopez]

Exactly, the code of the PR does not work in-place, because now I have not a clear idea of how to unfreeze a frozen object. I'll need to think about this.

This package has some limitations, like not supporting classes with __slots__, maybe numpy arrays have that or for other reason gelidum does not support them.

No worries, any constructive feedback is positive feedback. Thank you!