Tracing back single class sample sizes of a random forest classification model (.pkl-file) [QGIS 3.16 - EnMAP Box - RFC]

Question

A couple of weeks ago, I created some prediction models with the EnMAP Box Classification Workflow. In some of the models I reduced the sample size for some classes as shown in the image below. Here, the classes WF and WW were reduced by 12 and 24 sampling points, respectively. But this is just an example which I created some minutes ago.

Now I would like to refine my models. But my issue is that I do not remember of which classes I reduced the sample sizes, and to which amount. (I did not make any notes, as I relied on the files transporting the information). If loading the pkl-file into EnMap Box (Project > Data source) I can derive that my total sample size was 306 (see second image). I switched through all nodes of this somehow cryptic pkl-file-view. But I could find any hints for the sample size of each separate class.
So I don't know for example I reduced both WW and WF by each 18 (instead of 12/24), or whether I reduced ohter classes two.

Are the single class sample sizes somehow retracable or is there a better option to view the pkl-file?

---

I found a solution for my core issue by loading the pkl-file to another tool. I posted an according answer below.
However, this solution makes obvious that the class-sample-size-information must somehow be contained in the pkl-file. The more now I am interested where/how to find these informations directly in the pkl-file, and what tool/software I would have to use to view the whole file content in clear text?

Answer 1

After some more research I could solve my main issue and will give the answer by myself. The solution is not very spectecular. However, maybe somebody else, being as new as me to that topic, will save some time or nerves.

So, in case you do not remember your settings of the actually applied sample sizes of your classes you can load the RFC-model-pkl-file into another EnMAP-tool called Regressor Fit/Training Performance and running the tool. In the output you can view these class-wise actually applied sample sizes.
(But just to avoid misunderstandings: If your issue should also be not to remember the right RFC-training-input-files: There you cannot see the maximum available sample sizes of your classes. So referring to my example you will know then that you used 36 samples of class WF (what solves my issue), but you don't know out of which maximal available samples (48, 55, 60...). This you may derive of your RFC-training files, if you still remember the right one :-))