Using custom background knowledge¶
One of the key novelties of autoBOT is the use of triplet-based graph databases as background knowledge (i.e. knowledge graphs). As shown in the examples, when doing the initialization:
autoBOTLibObj = autoBOTLib.GAlearner(
train_sequences, # input sequences
use_concept_features = True,
train_targets, # target space
memory_storage=
"./memory", # tripled base for concept features
representation_type="neurosymbolic") # or symbolic or neural
There is a dedicated parameter called memory_storage, which is a path that links to a list of files that contain triplets. You can use one or more gzipped tsv files! Example below.
schlau\tantonym\tdumm schlau\tantonym\tnaiv schlau\tantonym\ttrottelig schlau\tantonym\tunwissend schlauerweise\tantonym\tblöderweise schlauerweise\tantonym\tdum
The databases are simple tab-separated subject-predicate-object triplets. The files need to be Gzipped (there can be more in the memory folder!).