Dead, unused code that might still be interesting. A museum of old stuff.
-
lg-export: Export AtomSpace contents into Link Grammar SQLite3 dictionaries. It "works" but is pointless; it's just an extra step of complexity that is not needed.
-
cluster: Assorted attempts at clustering words into clusters. None worked out, due to complexity, bugs, mis-design, mis-understanding what the actual problem was.
-
mst-parse: The older, scheme-based MST parser built with the opencog sheaf code. The LG parser can now do MST and MPG parsing, and so provides a better platform. So, this older parser is now obsolete. BTW, so is the sheaf code.
-
mpg-parse: The newer, matrix-api batch-processing MST/MPG parsing and counting code. The LG parser does all the counting. The problem with this code is it uses the matrix API, which is truly great, but has issues: (1) slow, causes oceans of scheme GC, (2) does not have an Atomese API, and cannot be pipeline processing.
-
pair-count: older pair-counting code. It tried to do too much, was too general. Thus, too complex. We now know exacly how to do pair-counting: simply and directly. The older proof-of-concept code used this.
-
pair-count-new: newer pair-counting code, that replaced the above. Implements
matrix-API for batch processing. But batch processing turns out to be unwieldy and is being phased out as a compute approach. -
fibers-sim.scm: Attempt to use scheme/guile fibers to obtain parallelism in the similarity calculations. Fails to improve performance. Plus fibers are buggy, somehow.
-
cheesy-thread.scm: Another attempt at parallelizing similarity calculations.
Stuff below should be deleted a few years after being declared dead.
-
singletons.scm: Code for working with single words. Not used anywhere. Well, used in the summary reports below.
-
summary.scm: Stale, boring report printer. The reports are anemic. Other than that, they do work.