Popper is an inductive logic programming (ILP) system.
If you use Popper, please cite the paper:
Andrew Cropper and Rolf Morel. Learning programs by learning from failures. Mach. Learn. 110(4): 801-856 (2021)
-
pyswip (You must install pyswip from the master branch!)
- use the command:
pip install git+https://github.com/yuce/pyswip@master#egg=pyswip
- use the command:
-
SWI-Prolog (8.4.2 or above)
-
Clingo (5.5.0 or above)
The latest release of Popper is here.
To install the master branch, run the command:
pip install git+https://github.com/logic-and-learning-lab/Popper@main
You can run Popper with the command python popper.py <input dir>
.
For instance, the command python popper.py examples/dropk
produces:
********** SOLUTION **********
Precision:1.00 Recall:1.00 TP:10 FN:0 TN:10 FP:0 Size:7
f(A,B,C):- tail(A,C),one(B).
f(A,B,C):- decrement(B,E),tail(A,D),f(D,E,C).
******************************
The command python popper.py examples/trains1
produces:
********** SOLUTION **********
Precision:1.00 Recall:1.00 TP:394 FN:0 TN:606 FP:0 Size:6
f(A):- has_car(A,C),has_car(A,B),long(B),three_wheels(C),roof_closed(B).
******************************
Look at the examples for guidance.
You can import Popper and use it in your Python code like so:
from popper.util import Settings, print_prog_score
from popper.loop import learn_solution
settings = Settings(kbpath='input_dir')
prog, score, stats = learn_solution(settings)
if prog != None:
print_prog_score(prog, score)
Popper requires three files:
- an examples file
- a background knowledge (BK) file
- a bias file
An examples file contains positive and negative examples of the relation you want to learn:
pos(grandparent(ann,amelia)).
pos(grandparent(steve,amelia)).
pos(grandparent(ann,spongebob)).
pos(grandparent(steve,spongebob)).
pos(grandparent(linda,amelia)).
neg(grandparent(amy,amelia)).
A BK file contains other information about the problem:
mother(ann,amy).
mother(ann,andy).
mother(amy,amelia).
mother(linda,gavin).
father(steve,amy).
father(steve,andy).
father(gavin,amelia).
father(andy,spongebob).
A bias file contains information necessary to restrict the search space of Popper. Predicate declarations tell Popper which predicate symbols it can use in the head or body of a rule, such as:
head_pred(grandparent,2).
body_pred(mother,2).
body_pred(father,2).
These declarations say that each rule in a program must have the symbol grandparent with arity two in the head and mother and/or father in the body, also with arity two. If we call Popper with these three files it will produce the output:
grandparent(A,B):-mother(A,C),father(C,B).
grandparent(A,B):-father(A,C),mother(C,B).
grandparent(A,B):-father(A,C),father(C,B).
grandparent(A,B):-mother(A,C),mother(C,B).
% Precision:1.00, Recall:1.00, TP:5, FN:0, TN:1, FP:0
Popper has three main bias settings:
max_vars(N)
sets the maximum number of variables in a rule toN
(default: 6)max_body(N)
sets the maximum number of body literals in a rule toN
(default: 6)max_clauses(N)
sets the maximum number of rules/clauses toN
(default: 1 or 2 ifenable_recursion
is set)
These parameters are important. They greatly influence the search space. If the values are too high then Popper might struggle to learn a solution. If the settings are too low then the search space might be too small to contain a good solution.
You can set these settings in the bias file or through the command line (see --help
).
Finding suitable values can often be a process of trial and error. We are trying to automatically set these settings.
Do not supply max_clauses if you are learning non-recursive programs.
Popper is an anytime algorithm.
By default, it shows intermediate solutions.
For instance, the command python popper.py examples/dropk
produces:
08:08:54 Num. pos examples: 10
08:08:54 Num. neg examples: 10
08:08:54 Searching programs of size: 3
08:08:54 Searching programs of size: 4
08:08:54 Searching programs of size: 5
08:08:54 Searching programs of size: 6
08:08:54 ********************
08:08:54 New best hypothesis:
08:08:54 tp:1 fn:9 size:6
08:08:54 f(A,B,C):- tail(E,C),tail(D,F),tail(F,E),even(B),tail(A,D).
08:08:54 ********************
08:08:56 Searching programs of size: 7
08:08:57 ********************
08:08:57 New best hypothesis:
08:08:57 tp:10 fn:0 size:13
08:08:57 f(A,B,C):- tail(A,C),element(A,B).
08:08:57 f(A,B,C):- tail(E,C),tail(D,F),tail(F,E),even(B),tail(A,D).
08:08:57 f(A,B,C):- decrement(B,D),tail(E,C),f(A,D,E).
08:08:57 ********************
08:08:58 ********************
08:08:58 New best hypothesis:
08:08:58 tp:10 fn:0 size:7
08:08:58 f(A,B,C):- tail(A,C),one(B).
08:08:58 f(A,B,C):- f(A,E,D),tail(D,C),decrement(B,E).
08:08:58 ********************
********** SOLUTION **********
Precision:1.00 Recall:1.00 TP:10 FN:0 TN:10 FP:0 Size:7
f(A,B,C):- tail(A,C),one(B).
f(A,B,C):- decrement(B,E),f(A,E,D),tail(D,C).
******************************
To supress this information, run Popper with the --quiet
(-q
) flag.
To enable recursion add enable_recursion.
to the bias file.
Recursion allows Popper to learn programs where a predicate symbol appears in both the head and body of a rule, such as to find a duplicate element (python popper.py examples/find-dupl
) in a list:
f(A,B):-head(A,B),tail(A,C),element(C,B).
f(A,B):-tail(A,C),f(C,B).
Or to remove (python popper.py examples/filter
) non-even elements from a list:
f(A,B):-empty(A),empty(B).
f(A,B):-tail(A,D),head(A,C),odd(C),f(D,B).
f(A,B):-head(A,E),even(E),tail(A,C),f(C,D),prepend(E,D,B).
Recursion is expensive, so it is best to try without it first.
Popper supports optional type annotations in the bias file.
A type annotation is of the form type(p,(t1,t2,...,tk)
for a predicate symbol p
with arity k
, such as:
type(f,(list,list)).
type(head,(list,element)).
type(tail,(list,list)).
type(empty,(list,)).
type(odd,(element,)).
type(even,(element,)).
type(prepend,(element,list,list)).
These types are optional but can substantially reduce learning times.
Prolog often requires arguments to be ground. For instance, when asking Prolog to answer the query:
X is 3+K.
It throws an error:
ERROR: Arguments are not sufficiently instantiated
Moreover, we want to reduce the number of answers from a query. For instance, calling the length
predicate with only variables leads to an infinite set of answers.
To avoid this issues, Popper supports optional direction annotations.
A direction annotation is of the form direction(p,(d1,d2,...,dk)
for a predicate symbol p
with arity k
, where each di
is either in
or out
.
An in
variable must be ground when calling the relation.
By contrast, an out
variable need not be ground.
Here are example directions:
direction(head,(in,out)).
direction(tail,(in,out)).
direction(length,(in,out)).
direction(prepend,(in,in,out)).
direction(geq,(in,in)).
Again, directions are optional but can substantially reduce learning times.
Popper supports automatic predicate invention (PI).
To enable PI, add the setting enable_pi.
to the bias file.
With PI enabled, Popper (python popper.py examples/kinship-pi
) learns the following program:
grandparent(A,B):-inv1(C,B),inv1(A,C).
inv1(A,B):-mother(A,B).
inv1(A,B):-father(A,B).
% Precision:1.00, Recall:1.00, TP:5, FN:0, TN:1, FP:0
Predicate invention is currently very expensive so it is best to avoid it if possible.
To run with statistics use the flag --stats
(default: false)
To run in debug mode use the flag --debug
(default: false)
To run in quiet mode use the flag --quiet
(default: False)
To run with a maximum learning time use the flag --timeout
(default: 600 seconds)
To run with a maximum example testing (only applies when learning recursive programs) time use the flag --eval-timeout
(default: 0.001 seconds)
To allow non-Datalog clauses, where a variable in the head need not appear in the body, add ``non_datalog.` to your bias file.
To allow singleton variables (variables that only appear once in a clause), add allow_singletons.
to your bias file.
To set the maximum number of literals allow in a program use the flag --max-literals
(default: 40)
To set the maximum number of body literals allowed in the body of a rule use the flag --max-body
(default: 6)
To set the maximum number of variables allowed in a rule use the flag --max-vars
(default: 6)
To set the maximum number of examples to learn from use the flag --max-examples
(default: 10000)