docs

Author: wwcohen

datasets/README.txt

@@ -0,0 +1,2 @@
+These are various sample datasets for tensorlog - check the individual
+directories for more information.  Some of these are used for testing.

datasets/amie-qa/README.txt

@@ -0,0 +1,10 @@
+AMIE is a system that mines rules from KBs.  Part of what was
+distributed with this system was a small sample KB.
+/afs/cs.cmu.edu/user/wcohen/shared-home/data/dialog-toy
+contains code to generate questions (using templates) from 
+this KB.
+
+Luis Galárraga, Christina Teflioudi, Katja Hose, and Fabian
+M. Suchanek. 2015. Fast rule mining in ontological knowledge bases
+with AMIE$$+$$+. The VLDB Journal 24, 6 (December 2015),
+707-730. DOI=http://dx.doi.org/10.1007/s00778-015-0394-1

datasets/amie/README.txt

@@ -0,0 +1,10 @@
+TODO (Katie): docs
+
+AMIE is a system that mines rules from KBs.  Part of what was
+distributed with this system was a largish set of rules and some small
+KBs, used here to ....
+
+Luis Galárraga, Christina Teflioudi, Katja Hose, and Fabian
+M. Suchanek. 2015. Fast rule mining in ontological knowledge bases
+with AMIE$$+$$+. The VLDB Journal 24, 6 (December 2015),
+707-730. DOI=http://dx.doi.org/10.1007/s00778-015-0394-1

datasets/cora/README.txt

@@ -0,0 +1,10 @@
+This is the CORA bibliography-matching problem described original in 
+
+Poon, H., & Domingos, P. (2007, July). Joint inference in information
+extraction. In AAAI (Vol. 7, pp. 913-918).
+
+and later adapted for PROPPR 
+
+William Yang Wang, Kathryn Mazaitis, William W. Cohen (2013):
+Programming with Personalized PageRank: A Locally Groundable
+First-Order Probabilistic Logic in CIKM-2013

datasets/fb15k-237/README.txt

@@ -0,0 +1,8 @@
+FB15K-237 Knowledge Base Completion Dataset from
+https://www.microsoft.com/en-us/download/details.aspx?id=52312
+
+Using a set of rules learned by ISG using PROPPR.
+
+Experiment does rule weight learning.
+
+

datasets/fb15k-speed/README.txt

@@ -0,0 +1,10 @@
+FB15K-237 Knowledge Base Completion Dataset from
+https://www.microsoft.com/en-us/download/details.aspx?id=52312
+
+Using a set of rules learned by ISG using PROPPR.
+
+Experiment does some performance tests for inference and such.  The
+speed numbers don't seem very stable - they vary a lot, maybe due to
+system load? so the tests are kindof unreliable.
+
+TODO: decide if the system got slower sometime around v1.3.6 or not.

datasets/grid/README.txt

@@ -0,0 +1,30 @@
+Various artificial learning tasks inspired by experiments in:
+
+Dries, Anton, et al. "ProbLog2: Probabilistic logic programming."
+Joint European Conference on Machine Learning and Knowledge Discovery
+in Databases. Springer, Cham, 2015.
+
+code for running scalability experiments in JAIR submission
+ - bigexpt.py
+ - bigtfexpt.py
+well-documented code to demo building a TF model
+ - demo.py
+
+learning an approximation of the problog2 semantics, by learning
+probabilities defined by a biased logistic on top of the proof-count
+function, in JAIR submission
+ - distlearning.py
+
+learning to process queries with >1 target ouput, by learning
+probabilities defined by a biased logistic on top of the proof-count
+function, in JAIR submission
+ - multiclass.py
+
+demo of integration with TF - embedding learning is in JAIR submission
+ - tfintegration.py
+
+automated tests
+ - expt.py
+ - testexpt.py
+ - tfexpt.py
+

datasets/grid/bigexpt.py

@@ -1,3 +1,5 @@
+# code for running scalability experiments in JAIR submission
+
 import sys
 import numpy as NP
 import random

datasets/grid/distlearning.py

@@ -16,7 +16,6 @@
 # at a different scale.  euclidean distance? jenson-shannon?
 # TODO: output test-set and full-data y's for visualization
 # TODO: why do the actual y's include non-zeros for the filler entities?
-# TODO: should I normalize this? why?
 #
 #
 # --gendata M - draw M sample interpretations (ie grids, where edges

datasets/grid/tfintegration.py

@@ -5,8 +5,6 @@
 import random
 import getopt
 
-# todo: for visualization, sort everything by cell name
-#
 # 1) demonstrates adding logical inference as an input to a tensorflow
 #  function, and training thru the function to modify fact confidences
 #  (with the --corner soft option)