-
Jérôme Euzenat authored
In case of dependent link keys, it may happens that the context contains several link key conditions subsuming each others. So far, it selected one at random (the subsumed method was not finished). I changed that so that it could return concept\wedge concept. This is due to the uneasiness to test subsumption. This is only a hack because there should have several such concepts (not just two) and they cannot be dealt with like that.
Jérôme Euzenat authoredIn case of dependent link keys, it may happens that the context contains several link key conditions subsuming each others. So far, it selected one at random (the subsumed method was not finished). I changed that so that it could return concept\wedge concept. This is due to the uneasiness to test subsumption. This is only a hack because there should have several such concepts (not just two) and they cannot be dealt with like that.
Dev.README 962 B
Dev.README 2018-03-25
* concept subsumption
Concept reduction... when a concept refers through others as keys. It should be good to identify the smallest one... but this is not very possible and not done.
I only implemented a trick with the subsumed test in exported: this is not satisfying
If fact all this work should be done as well within the whole algorithm...
* Alignment output
TODO
* concept naming
Several names may currently be given to the same concept leading to ugly display and, at least, too many iterations.
This has to be fixed.
* intent/extent
The link key extension of the initial FCA implementation is based on the encoding of intent and extent into string.
This somewhat explain the confusion between intent and extent.
A definitive improvement would be to generalise the FCA implementation to manipulate structures and not strings.
This could then support specific implementations such as those of link key candidate extraction.