Bias Reformulation for One-shot Function Induction

Revision: 
1
Contributing Authors: 
  • Dianhuan Lin
  • Eyal Dechter
  • Kevin Ellis
  • Joshua B. Tenenbaum
  • Stephen H. Muggleton
Description: 

Instructions for running the experiment:

$ vagrant init SomeName http://recomputation.org/ecai2014/experiments/dianhuan/metagolDF.box
$ vagrant up

The two subdirectories 'metagolDF_noRec' and 'metagolDF_withRec' correspond to different declarative bias. Specifically, 'metagolDF_noRec' is not given the recursive metarule, while 'metagolDF_withRec' is provided with the recursive metarule, thus a much larger hypothesis space. Therefore 'metagolDF_noRec' and 'metagolDF_withRec' are mainly different in the file 'flashfill_prim2.pl' corresponding to background knowledge.

The details of the files within each subdirectory are as follows:

1.'gen_metagol.pl': MetagolDF
2.'./examples_oneShot': Examples
3.'flashfill_prim2.pl': Background knowledge including declarative bias
4. 'scripts_oneShot_prim2_dependent' and 'scripts_oneShot_prim2_independent' contain the scripts for comparing independent and dependent learning.

Running the following commands will reproduce the results as those in the subdirectory 'scripts_oneShot_prim2_dependent/results' or 'scripts_oneShot_prim2_independent/results'.
$ bash ./script.sh

Note: The indexes of the severnteen problems are renamed as below when presented in the paper.
ep01 => ep01
ep02 => ep02
ep03 => ep03
ep04 => ep04
ep06 => ep05
ep08 => ep06
ep10 => ep07
ep13 => ep08
ep15 => ep09
ep16 => ep10
ep18 => ep11
ep19 => ep12
ep21 => ep13
ep28 => ep14
ep29 => ep15
ep30 => ep16
ep31 => ep17