celinehocquette / magicpopper Goto Github PK

C. Hocquette and A. Cropper, Learning programs with magic values.

Home Page: https://link.springer.com/article/10.1007/s10994-022-06274-w

Python 0.01% Prolog 99.95% Perl 0.05% Raku 0.01%

inductive-logic-programming program-synthesis

magicpopper's Introduction

Hi there 👋 Welcome to my Github page!

My name is Céline Hocquette, I am a researcher at the university of Oxford in the Logic and Learning group. My research focuses on inductive logic programming, a form of program synthesis based on logic. You can explore all my papers on my website.

Feel free to reach out if you have any questions!

magicpopper's People

Contributors

Stargazers

Watchers

Forkers

hakank

magicpopper's Issues

Is the example interval correct?

Running the interval example give the following solution:

python3 popper.py examples/interval        
NO MORE SOLUTIONS
********** SOLUTION **********
Precision:1.00 Recall:1.00 TP:26 FN:0 TN:22 FP:0 Size:26
interval(-36).
interval(-11).
interval(19).
interval(23).
interval(13).
interval(-20).
interval(-15).
interval(-19).
interval(-27).
interval(4).
interval(-29).
interval(-1).
interval(-28).
interval(-9).
interval(-34).
interval(-30).
interval(-8).
interval(-14).
interval(7).
interval(-35).
interval(2).
interval(5).
interval(-16).
interval(12).
interval(-21).
interval(9).
******************************

which is just a listing of the positive instances.

Is this the correct solution? bk.pl and bias.pl include a lot of predicates which seems to indicate that the solutions should use some of these predicates.

Program runs forever

Hello,

Thank you for making such nice work, which I think will benefit my research a lot. When playing with the software, I found something strange. I am running a toy problem of identifying root causes.

Here is my bk.pl

type_of(u1, ut1).
type_of(u2, ut1).
type_of(u3, ut2).

at_state(amf, ns1).
at_state(smf, ns1).
at_state(nrf, ns1).
at_state(ausf, ns2).
at_state(bsf, ns2).

has_user(c1, u1).
has_user(c2, u1).
has_user(c3, u2).
has_user(c4, u3).
has_user(c5, u3).
has_user(c6, u3).

calls_nf(c1, amf).
calls_nf(c2, amf).
calls_nf(c3, nrf).
calls_nf(c4, ausf).
calls_nf(c5, ausf).
calls_nf(c6, bsf).

and my exs.pl

pos(rootcause(c1, r1)).
pos(rootcause(c2, r1)).
pos(rootcause(c3, r1)).
pos(rootcause(c4, r2)).
pos(rootcause(c5, r2)).
pos(rootcause(c6, r2)).

If I impose such bias

head_pred(rootcause,2).
magic_value(rootcause, 1).

body_pred(has_user, 2).
body_pred(type_of, 2).
magic_value(type_of, 1).

It can successfully find the rules

$python ./magicpopper/popper.py ./magicpopper/examples/rca
********** SOLUTION **********
Precision:1.00 Recall:1.00 TP:6 FN:0 TN:0 FP:0 Size:6
rootcause(A,r1):- has_user(A,D),type_of(D,ut1).
rootcause(A,r2):- type_of(D,ut2),has_user(A,D).
******************************

Or if I use another set of bias

head_pred(rootcause,2).
magic_value(rootcause, 1).

body_pred(calls_nf, 2).
body_pred(at_state, 2).
magic_value(at_state, 1).

It can also successfully find the rules

$python ./magicpopper/popper.py ./magicpopper/examples/rca
********** SOLUTION **********
Precision:1.00 Recall:1.00 TP:6 FN:0 TN:0 FP:0 Size:6
rootcause(A,r1):- calls_nf(A,C),at_state(C,ns1).
rootcause(A,r2):- at_state(C,ns2),calls_nf(A,C).
******************************

However, when I put the 2 sets of bias together, hoping that it can induce both set of rules

head_pred(rootcause,2).
magic_value(rootcause, 1).

body_pred(has_user, 2).
body_pred(type_of, 2).
magic_value(type_of, 1).

body_pred(calls_nf, 2).
body_pred(at_state, 2).
magic_value(at_state, 1).

:-
	body_literal(Rule1,Q,_,_),
	body_literal(Rule2,Q,_,_),
	Rule1 != Rule2.

The program runs forever. Does any one know what was going wrong?

OS: ubuntu 20.04
Python: 3.9.7
Prolog: SWI-Prolog (threaded, 64 bits, version 8.5.15-97-g5c7d1f835)

Any comments will be appreciated.

Problem when running ilp-experiments

Next test is to run (MagicPopper's) ilp-experiments. Unfortunately throws an error:

$ cd magicpopper/ilp-experiments
$ python3 ilpexp.py FAST_TEST

Traceback (most recent call last):
  File "ilpexp.py", line 3, in <module>
    import ilpexp.experiment as experiment
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/ilp-experiments/ilpexp/experiment.py", line 2, in <module>
    from .problem import ALL_PROBLEMS, DEFAULT_PROBLEMS, DEFAULT_LIST_PROBLEMS, MINIMAL_DECAY
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/ilp-experiments/ilpexp/problem/__init__.py", line 2, in <module>
    from .trains import DEFAULT_TRAINS_PROBLEMS, ALL_TRAINS_PROBLEMS
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/ilp-experiments/ilpexp/problem/trains/__init__.py", line 1, in <module>
    from .trains import TrainsProblem
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/ilp-experiments/ilpexp/problem/trains/trains.py", line 6, in <module>
    from ...system import *
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/ilp-experiments/ilpexp/system/__init__.py", line 2, in <module>
    from .popper import BASIC_POPPER, MAGIC_POPPER, Popper, PopperTrainSettings
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/ilp-experiments/ilpexp/system/popper/__init__.py", line 1, in <module>
    from .popper import Popper, PopperTrainSettings, BASIC_POPPER, MAGIC_POPPER, generate_bias_file, generate_bk_file
ImportError: cannot import name 'MAGIC_POPPER' from 'ilpexp.system.popper.popper' (/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/ilp-experiments/ilpexp/system/popper/popper.py)

Both Python version 3.7 and 3.9 give the same error.

Note that this is not the same error that I get when running the "other" (real?) ilp-experiment reported in logic-and-learning-lab/ilp-experiments#18

Running plain Popper models does not work

After fixing the error reporting in #1 (commenting the lines regarding args.info in utils.py), the MagicPopper specific examples works, but for the plain Popper examples errors are thrown.

$ python3 popper.py examples/alleven       
Traceback (most recent call last):
  File "popper.py", line 7, in <module>
    settings = Settings()
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/magicpopper/popper/util.py", line 447, in __init__
    self.logger.debug(f'Max vars: {self.max_vars}')
AttributeError: 'Settings' object has no attribute 'max_vars'

After commenting the offending line and the next line (since it yields a similar error):

   # self.logger.debug(f'Max vars: {self.max_vars}')
   # self.logger.debug(f'Max body: {self.max_body}')

then the program throws this error:

Traceback (most recent call last):
  File "popper.py", line 8, in <module>
    prog, score, stats = learn_solution(settings)
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/magicpopper/popper/loop.py", line 316, in learn_solution
    timeout(settings, popper, (settings,), timeout_duration=int(settings.timeout),)
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/magicpopper/popper/util.py", line 78, in timeout
    raise moo
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/magicpopper/popper/util.py", line 70, in timeout
    result = func(*args, **kwargs)
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/magicpopper/popper/loop.py", line 67, in popper
    generator = Generator(settings, grounder)
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/magicpopper/popper/generate.py", line 40, in __init__
    encoding.append(f'max_body({settings.max_body}).')
AttributeError: 'Settings' object has no attribute 'max_body'

Running MagicInfo models yields "AttributeError: 'Namespace' object has no attribute 'info'"

When running any of the examples, e.g.

$ python3 popper examples/iggp-magiccoins

the following error is thrown:

Traceback (most recent call last):
  File "popper.py", line 7, in <module>
    settings = Settings()
  File "/home/hakank/inductive_programming/andrewcropper/popper/magicpopper/magicpopper/popper/util.py", line 364, in __init__
    elif args.info:
AttributeError: 'Namespace' object has no attribute 'info'

After commenting these lines

--- a/magicpopper/popper/util.py
+++ b/magicpopper/popper/util.py
@@ -361,9 +361,9 @@ class Settings:
         elif args.debug:
             log_level = logging.DEBUG
             logging.basicConfig(format='%(asctime)s %(message)s', level=log_level, datefmt='%H:%M:%S')
-        elif args.info:
-            log_level = logging.INFO
-            logging.basicConfig(format='%(asctime)s %(message)s', level=log_level, datefmt='%H:%M:%S')
+        # elif args.info:
+        #     log_level = logging.INFO
+        #     logging.basicConfig(format='%(asctime)s %(message)s', level=log_level, datefmt='%H:%M:%S')
 
         self.stats = Stats(info=info, debug=args.debug, stats_file=args.stats_file)
         self.stats.logger = self.logger

then he specific MagicPopper examples work, e,g,

python3 popper.py examples/iggp-magiccoins-goal

NO MORE SOLUTIONS
********** SOLUTION **********
Precision:1.00 Recall:1.00 TP:62 FN:0 TN:62 FP:0 Size:5
goal(A,you,100):- my_true_step(A,5).
goal(A,you,0):- my_true_step(A,E),different(5,E).
******************************

Note: running any "plain" Popper examples yields an error:

AttributeError: 'Settings' object has no attribute 'max_vars'

I report a separate issue about this.

Separate magicpopper and ilp-experiments?

The magicpopper repo is currently very large: about 9Gn, of which magicpopper is just a few Mb; the rest is for ilp-experiments (including the .git directories).

Would it be possible to separate these two directories so one can clone just the magicpopper directory?

Problem with ilp-experiments: python_cmd

Now ilp-experiments almost works. The only issue is the line

python_cmd = os.path.join(sys.exec_prefix, "bin", "python")

in the file ilp-experiments/ilpexp/system/popper/popper.py. It assumes that my Python executable is "/usr/local/bin/python", which not correct since it's "/usr/local/bin/python3".

The following fix seems to work, i.e. using sys.executable instead:

--- a/ilp-experiments/ilpexp/system/popper/popper.py
+++ b/ilp-experiments/ilpexp/system/popper/popper.py
@@ -72,7 +72,7 @@ class Popper(System):
         import sys
         # could break if the environment contains packages that have custom activation code
         # https://stackoverflow.com/questions/51819719/using-subprocess-in-anaconda-environment
-        python_cmd = os.path.join(sys.exec_prefix, "bin", "python")
+        python_cmd = sys.executable
         python_file = os.sep.join([self.install_dir, 'popper.py'])
         run_command([python_cmd, python_file], final_settings, timeout=self.timeout + 10)

At least it work on my machine both when running "plain" Pythons 3.7 as well as running Anaconda's Python 3.9.

Minor: "NO MORE SOLUTIONS" before the solution is a little confusing

When running a Magic example the line "NO MORE SOLUTIONS" is printed before the solutions which is a little confusing.

E.g.:

python3 popper.py examples/iggp-magic-minimal-decay 
NO MORE SOLUTIONS
********** SOLUTION **********
Precision:1.00 Recall:1.00 TP:8 FN:0 TN:46 FP:0 Size:6
next_value(A,5):- does(A,player,pressButton).
next_value(A,B):- does(A,player,noop),true_value(A,D),my_succ(B,D).
******************************

(Thanks for MagicPopper! I really like it and will play with some of my own examples. )

Place (working) examples of ilp-experiments in the examples directory

Here is a suggestion for the magicpopper examples directory.

It would be great if there was at least one (working) example from each of the problems in ilp-experments.
That would makes them more accessible and also show more of the power of MagicPopper.

Installation of Popper and magicpopper.

I installed Popper un Ubuntu 22.04.2 with Python 3.11.2 and run an example from https://github.com/logic-and-learning-lab/Popper/#library-usage with no problems. Afterwards, I installed magicpopper from https://github.com/celinehocquette/magicpopper/tree/main/magicpopper using pip install .. The installation overwrites popper-ilp-2.0.0 and the example no longer works. This is the error message:

(.venv) paul@desktop:~/st-python/prolog$  /usr/bin/env /home/paul/st-python/prolog/.venv/bin/python /home/paul/.vscode/extensions/ms-python.python-2023.4.1/pythonFiles/lib/python/debugpy/adapter/../../debugpy/launcher 36639 -- /home/paul/st-python/prolog/popper-0.py 
Traceback (most recent call last):
  File "/home/paul/st-python/prolog/popper-0.py", line 4, in <module>
    settings = Settings(kbpath='kinship')
               ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/paul/st-python/prolog/.venv/lib/python3.11/site-packages/popper/util.py", line 393, in __init__
    with open(self.bias_file) as f:
         ^^^^^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory: ''

Can these two, Popper and magicpopper, be installed together? Does it even make sense? I'm just learning about ILP, so these questions may be completely clueless.