Action applynoisy

Application of rules using noisy-or aggregation approach:

\[score(e) = 1 - \prod_{i=1}^{k} (1 - conf(r_i))\]

where \(r_1, \dots, r_k\) are rules that predict entity \(e\).

Configuration file

Input :

• PATH_TRAININGValid path (file)

  Path to training file (absolute or relative), default: train.txt

• PATH_TESTValid path (file)

  Path to test file (absolute or relative), default: test.txt

• PATH_VALIDValid path (file)

  Path to validation file (absolute or relative), default: valid.txt

• PATH_RULESValid path (file)

  Path to rules file (absolute or relative), default: rules.txt

Properties :

• WORKER_THREADSint

  Number of threads that are used for computation. (-1 means all threads are used), default: -1

• DISCRIMINATION_BOUNDint

  Discriminates (omits) rules which predict more elements than this, 0 means no limit., default: 4000

• UNSEEN_NEGATIVE_EXAMPLESint

  The number of negative examples for which we assume that they exist, however, we have not seen them. Rules with high coverage are favoured the higher the chosen number, default: 5

• REFLEXIV_TOKENstring

  Token used for substitution of reflexive rules. (Used if AnyBURL ruleset was trained with REWRITE_REFLEXIV = TRUE), default: me_myself_i

• TOP_K_OUTPUTint

  The top-k results that are after filtering kept in the results, default: 10

• PREDICT_UNKNOWNint

  If set to 1, does not skip triples containing unkwown entities in the training set. F.e. generates predictions for john speaks UNKOWN if UNKNOWN is not in the training set. default 0

• ONLY_XYint

  If set to 1, only cyclic (XY) rules are read from the rules file, default: 0

Output :

• PATH_OUTPUTValid path (file)

  Path to file used for storing predictions, default: predictions.txt