Sampler

[fgrunewald]

Initial draft for a sampler class that takes fragments and makes random molecules according to bonding probabilities

[pckroon]

Some small comments, looks mostly fine to me.

[pckroon]

I like most of the changes, but I'm not happy with how you seed the random generator. You should seed it only once, at the start of your process. Otherwise your seeded random will not be random enough.
You can also, for reproducibility, generate a seed if None is given (time.time_ns()) and log that somewhere the user can find it.

[pckroon]

For the Sampler examples, I'm not super clear on what the difference between *reactivities and *probabilities is. Also, for some examples you do from_string, for other from_fragment_string. Is this intentional?
Lastly, does it make sense to completely replace the terminal logic with the fragment reactivities?

[fgrunewald]

@pckroon All reactivities are probabilities. Reactivities is used when the decision is based on a bonding descriptor. I've added a clarification in the docs.

they should all be from_fragment_string but some examples were older from before

Not entirely because at the CG level branch termination cannot be determined by the bonding descriptor when no end-group is added. Otherwise, yes one could do as suggest. Of course one other option is to simply define an empty or fake terminal. But not sure that makes it easier.

[pckroon]

Thanks for the clarification. The docstring(s) also help me :)
Still found a typo though.

As for the terminals, I don't have a super strong preference for one or the other. I just know from experience that the more complex an API is, the harder it is to use. It requires more mental space to keep track of all the arguments, their different formats/types/conventions, and how it all interacts.

[pckroon]

Still found some small points :)