I need to rebase this but looks like interesting functionality. Late here but the only things that stick out a little at the moment is the SMARTS generation in the fingerprint, okay for now but I would prefer to centralise this but obviously would need to write that/use AMBITs one. Also wondering how you get rings in a circular fingerprint? Shouldn't all the subgraphs be trees, will probably become clear once I look more in depth.

• Fine to centralize SMARTS generation if possible, as here it uses internal CircularFingerprint structures;
• There are no AMBIT dependencies in this code (intentionally) - of course SMARTS should be valid at the end;
• Rings appear if the depth is large enough. @ntk73 may explain implementation details.
  As this feature is modeled after JCompoundMapper here is JCompoundMapper output for benzene

ECFP6 ( :1 means count , it's not part of the SMARTS)

[*]=1=C=1C=C2[*]2:1 
[*]=1=C=12([*]2):1  
[*]1C1=2(=[*]=2):1  
[*]1C1=CC=CC=2=[*]=2:1  
[*]=1=C=1C=CC=C2[*]2:1  
c1ccccc1:1  
[*]1c1=CC=Cc=2=[*]=2:1  
[*]1c1=CC=2=[*]=2:1 
[*]1C1=CC=CC=2=[*]=2:1

ECFP4 (no ring structures)

[*]1C1=2(=[*]=2):1  
[*]=1=C=1C=CC=C2[*]2:1  
[*]1C1=CC=CC=2=[*]=2:1  
[*]1C1=CC=2=[*]=2:1 
[*]=1=C=1C=C2[*]2:1 
[*]=1=C=12([*]2):1

our code generates these fragments for benzene (default CLASS_ECFP6)

c(:a):a
c(c:a)c:a
c(cc:a):a
c(:a)cc:a
c(cc:a)cc:a
c(ccc:a)c:a
c(cccc:a):a
c(:a)cccc:a
c(c:a)ccc:a
c1ccccc1

and for CLASS_ECFP4

c(:a):a
c(c:a)c:a
c(cc:a):a
c(:a)cc:a
c(cc:a)cc:a
c(ccc:a)c:a
c(cccc:a):a
c(:a)cccc:a
c(c:a)ccc:a

Rebased - #231