The pycddlib installation problem cannot be resolved with the commands provided at the ReadMe unfortunately. I've done some digging at the github page of pycddlib and it seems like the currently available version (2.1.6) isn't compatible with MacOS. It seems like they experimented with new versions that are compatible (2.1.7a0) but aren't available to install yet. I've inferred all these information from the page https://github.com/mcmtroffaes/pycddlib/actions

So, how can we work around this issue and be able to install & test LNN regardless?

Currently running the code brew install gmp; env "CFLAGS=-I/usr/local/include -L/usr/local/lib" pip install pycddlib returns the following error:

`Collecting pycddlib Using cached pycddlib-2.1.6.tar.gz (159 kB) Preparing metadata (setup.py) ... done Building wheels for collected packages: pycddlib Building wheel for pycddlib (setup.py) ... error error: subprocess-exited-with-error

× python setup.py bdist_wheel did not run successfully. │ exit code: 1 ╰─> [18 lines of output] running bdist_wheel running build running build_ext cythoning cdd.pyx to cdd.c building 'cdd' extension creating build creating build/temp.macosx-10.9-x86_64-3.9 creating build/temp.macosx-10.9-x86_64-3.9/cddlib creating build/temp.macosx-10.9-x86_64-3.9/cddlib/lib-src clang -Wno-unused-result -Wsign-compare -Wunreachable-code -DNDEBUG -fwrapv -O2 -Wall -fPIC -O2 -isystem /Users/ub/opt/anaconda3/envs/lnn/include -arch x86_64 -I/Users/ub/opt/anaconda3/envs/lnn/include -fPIC -O2 -isystem /Users/ub/opt/anaconda3/envs/lnn/include -arch x86_64 -I/usr/local/include -L/usr/local/lib -DGMPRATIONAL -Icddlib/lib-src -I/Users/ub/opt/anaconda3/envs/lnn/include/python3.9 -c cdd.c -o build/temp.macosx-10.9-x86_64-3.9/cdd.o clang: warning: argument unused during compilation: '-L/usr/local/lib' [-Wunused-command-line-argument] In file included from cdd.c:749: In file included from cddlib/lib-src/cdd.h:17: cddlib/lib-src/cddmp.h:30:11: fatal error: 'gmp.h' file not found #include "gmp.h" ^~~~~~~ 1 error generated. error: command '/usr/bin/clang' failed with exit code 1 [end of output]

note: This error originates from a subprocess, and is likely not a problem with pip. ERROR: Failed building wheel for pycddlib Running setup.py clean for pycddlib Failed to build pycddlib Installing collected packages: pycddlib Running setup.py install for pycddlib ... error error: subprocess-exited-with-error

× Running setup.py install for pycddlib did not run successfully. │ exit code: 1 ╰─> [18 lines of output] running install running build running build_ext skipping 'cdd.c' Cython extension (up-to-date) building 'cdd' extension creating build creating build/temp.macosx-10.9-x86_64-3.9 creating build/temp.macosx-10.9-x86_64-3.9/cddlib creating build/temp.macosx-10.9-x86_64-3.9/cddlib/lib-src clang -Wno-unused-result -Wsign-compare -Wunreachable-code -DNDEBUG -fwrapv -O2 -Wall -fPIC -O2 -isystem /Users/ub/opt/anaconda3/envs/lnn/include -arch x86_64 -I/Users/ub/opt/anaconda3/envs/lnn/include -fPIC -O2 -isystem /Users/ub/opt/anaconda3/envs/lnn/include -arch x86_64 -I/usr/local/include -L/usr/local/lib -DGMPRATIONAL -Icddlib/lib-src -I/Users/ub/opt/anaconda3/envs/lnn/include/python3.9 -c cdd.c -o build/temp.macosx-10.9-x86_64-3.9/cdd.o clang: warning: argument unused during compilation: '-L/usr/local/lib' [-Wunused-command-line-argument] In file included from cdd.c:749: In file included from cddlib/lib-src/cdd.h:17: cddlib/lib-src/cddmp.h:30:11: fatal error: 'gmp.h' file not found #include "gmp.h" ^~~~~~~ 1 error generated. error: command '/usr/bin/clang' failed with exit code 1 [end of output]

note: This error originates from a subprocess, and is likely not a problem with pip. error: legacy-install-failure

× Encountered error while trying to install package. ╰─> pycddlib

note: This is an issue with the package mentioned above, not pip. hint: See above for output from the failure.`

In [1]: from lnn import Predicate, And
   ...: 
   ...: P = Predicate("P", arity=0)
   ...: 

In [2]: And(P(), P())
Out[2]: <lnn.symbolic.logic.And at 0x1107e4d30>

In [3]: And(P(), P)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Input In [3], in <module>
----> 1 And(P(), P)

File ~/Gits/LNN/lnn/symbolic/logic.py:1110, in _NAryNeuron.__init__(self, *formula, **kwds)
   1109 def __init__(self, *formula, **kwds):
-> 1110     super().__init__(*formula, arity=len(formula), **kwds)

File ~/Gits/LNN/lnn/symbolic/logic.py:976, in _ConnectiveNeuron.__init__(self, *formula, **kwds)
    975 def __init__(self, *formula, **kwds):
--> 976     super().__init__(*formula, **kwds)
    977     self.neuron = _NeuralActivation(kwds.get("neuron", {}).get("type"))(
    978         self.propositional, self.arity, self.world, **kwds.get("neuron", {})
    979     )
    980     self.func = self.neuron.function(
    981         self.__class__.__name__, direction=Direction.UPWARD
    982     )

File ~/Gits/LNN/lnn/symbolic/logic.py:971, in _ConnectiveFormula.__init__(self, *formula, **kwds)
    970 def __init__(self, *formula: _Formula, **kwds):
--> 971     super().__init__(*formula, **kwds)

File ~/Gits/LNN/lnn/symbolic/logic.py:156, in _Formula.__init__(self, name, arity, world, *formula, **kwds)
    154     for subformula in formula:
    155         if isinstance(subformula, Predicate):
--> 156             raise ValueError(
    157                 f"predicate {subformula} inside formula must be called"
    158             )
    160 # formula naming
    161 self.name = (
    162     name
    163     if name
    164     else (f"{self.class_name}_{_Formula.unique_num[self.class_name]}")
    165 )

ValueError: predicate P inside formula must be called

The given examples and tutorial shows LNN can do the logical reasoning from given data and root formulae. Can LNN generate logic hypothesis from the given data? Like, can it output the possible logic rules induced from background knowledges?

Thank you.

Compound formulae expect predicates to be called (call)

e.g. given:

P = Predicate('P')
Q = Predicate('Q', arity=2)

correct

And(P(x), Q(x, y))

incorrect

And(P, Q)

not having called compound formulae is only the syntax for propositional inputs e.g.

P = Proposition('P')
Q = Proposition('Q')
And(P, Q)

right now it seems to fail silently by reverting to the default configuration

When **kwds is not part of the argument list, the following error is thrown when training a network which contains a negation: "TypeError: upward() got an unexpected keyword argument 'losses'"

This script generates this error in the original code: https://drive.google.com/file/d/16mUagX7-zZw4iF8x972D6B-Ct9_l0cWB/view?usp=sharing

Hello Team, There are 2 out of 3 predicates in add_knowledge sentence, it would be best to add all 3 of them or none. I made a fix to add none.

Signed-off-by: michal [email protected]

Signed-off-by: HannaAbiAkl [email protected]

Update README for Windows Graphviz conda install as per official PyGraphviz documentation: https://pygraphviz.github.io/documentation/stable/install.html

Quickstart instructions

"3. Install the LNN as a library: pip install git+https://github.com/IBM/LNN.git"

should be updated to agree with the Box Note instructions;

"3. Install the LNN as a library: pip install git+https://github.com/GitHubUserName/LNN.git"

If you run in this error running the command pip install git+https://github.com/IBM/LNN.git on Windows

  Building wheel for pygraphviz (setup.py) ... error
  ERROR: Command errored out with exit status 1:
  ....

install pygraphvizwith the command below and then install LNN

pip install --global-option=build_ext --global-option="-IC:\Program Files\Graphviz\include" --global-option="-LC:\Program Files\Graphviz\lib" pygraphviz

Hi,

**kwargs is probbaly more common in Python projects, but even standard libs sometimes uses **kwds. So feel free to close this PR if you prefer **kwds.

Tomas

Hi,

I have reordered imports according to PEP8 (https://www.python.org/dev/peps/pep-0008/#imports). Not sure if there is some specific reason to have them in original order, feel free to close this PR :)

Also I can also fix (if needed) following:

• there are places where import itertools and same time from itertools import chain is used in other file, maybe first should be preferred?
• in general I am missing more blank lines in the code, maybe some strict formatter can be use (like black)?

TM

Hi there, sorry for another question:

This is the full program for the Missle logic program:

from lnn import (Predicate, Variable, Join, And,
                 Exists, Implies, ForAll, Model, Fact, World)

model = Model()  # Instantiate a model.
x, y, z, w = map(Variable, ['x', 'y', 'z', 'w'])

missile = Predicate('missile')
american = Predicate('american')
hostile = Predicate('hostile')
criminal = Predicate('criminal')
weapon = Predicate('weapon')
owns = Predicate('owns', 2)    # binary predicate
enemy = Predicate('enemy', 2)  # binary predicate
sells = Predicate('sells', 3)  # ternary predicate

americaenemyishostile = ForAll(x, Implies(enemy(x, 'america'), hostile(x)))
missleareweapon = ForAll(x, Implies(missile(x), weapon(x)))
nonomisslesoldbywest = ForAll(x, Implies(And(missile(x), owns('nono', x)),sells('west', x, 'nono')))
youarecriminalif = ForAll(x,y,z, Implies(And(american(x), sells(x,y,z), weapon(y), hostile(z)), criminal(x)))

query = Exists(x, criminal(x))
model.add_knowledge(americaenemyishostile, missleareweapon, nonomisslesoldbywest, youarecriminalif, query)

model.add_data({
    owns: {('nono', 'm1'): Fact.TRUE},
    missile: {'m1': Fact.TRUE},
    american: {'west': Fact.TRUE},
    enemy: {('nono', 'america'): Fact.TRUE, ('wakanda', 'america'): Fact.TRUE, ('gotham', 'america'): Fact.TRUE},
})

model[query].print()
model.infer()
model[query].print()
model[query].true_groundings

Running with

model.add_data({
    owns: {('nono', 'm1'): Fact.TRUE},
    missile: {'m1': Fact.TRUE},
    american: {'west': Fact.TRUE},
    enemy: {('nono', 'america'): Fact.TRUE, ('wakanda', 'america'): Fact.TRUE, ('gotham', 'america'): Fact.TRUE},
})

yields the correct result in less than 2s:

OPEN Exists: (∃0, criminal(0))                              TRUE (1.0, 1.0)

{'west'}

However, running with two extra facts that should not impact the results

model.add_data({
    owns: {('nono', 'm1'): Fact.TRUE}, ('wakanda', 'm2'): Fact.TRUE}, ('gotham', 'm3'): Fact.TRUE}
    missile: {'m1': Fact.TRUE, 'm2': Fact.TRUE, 'm3': Fact.TRUE},
    american: {'west': Fact.TRUE},
    enemy: {('nono', 'america'): Fact.TRUE, ('wakanda', 'america'): Fact.TRUE, ('gotham', 'america'): Fact.TRUE},
})

stalls for > 1hr.

Is this common and do you know if LNN in general will be able to scale to much larger problems?

From the tutorial, it seems like a CONTRADICTION is when L > U: however, the results bellow happens quite often when L > U is considered APPROX TRUE.

Can someone help to clarify this?

Model before inference:

                            LNN Model

OPEN And: (A ∧ B) APPROX_TRUE (0.51, 0.52)

OPEN Proposition: B APPROX_TRUE (0.9, 0.99)

OPEN Proposition: A APPROX_TRUE (0.9, 0.99)

Model after inference:

                            LNN Model

OPEN And: (A ∧ B) APPROX_TRUE (0.8, 0.52)

OPEN Proposition: B APPROX_TRUE (0.9, 0.99)

OPEN Proposition: A APPROX_TRUE (0.9, 0.99)

Hello, I am interested in doing a final year project for my bachelor's degree in this topic, and I need more guidance and tutorials.

Within the tutorials/README.md of this repo, an image of 9 jupyter notebook tutorials is shown, but the repo only has 3.

What happened to the other .ipynb files? Furthermore is there a tutorial on applying LNN's to knowledge graphs?

Thank you.

From the Usage page I have abstracted the following code

from lnn import Predicates
from lnn import Variables
from lnn import Implies, Equivalent
from lnn import World
from lnn import Fact

Smokes, Cancer = Predicates('Smokes', 'Cancer')
Friends = Predicates('Friends', arity=2)

x, y = Variables('x', 'y')

Smoking_causes_Cancer = Implies(Smokes(x), Cancer(x))
Smokers_befriend_Smokers = Implies(Friends(x, y), Equivalent(Smokes(x), Smokes(y)))

formulae = [
    Smoking_causes_Cancer,
    Smokers_befriend_Smokers
]
model.add_knowledge(*formulae, world=World.AXIOM)

# add data to the model
model.add_data({
    Friends: {
        ('Anna', 'Bob'): Fact.TRUE,
        ('Bob', 'Anna'): Fact.TRUE,
        ('Anna', 'Edward'): Fact.TRUE,
        ('Edward', 'Anna'): Fact.TRUE,
        ('Anna', 'Frank'): Fact.TRUE,
        ('Frank', 'Anna'): Fact.TRUE,
        ('Bob', 'Chris'): Fact.TRUE},
    Smokes.name: {
        'Anna': Fact.TRUE,
        'Edward': Fact.TRUE,
        'Frank': Fact.TRUE,
        'Gary': Fact.TRUE},
    Cancer.name: {
        'Anna': Fact.TRUE,
        'Edward': Fact.TRUE}
    })
model.print()

I added a comma in the formulae. When I run this, I get

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In [17], line 27
     23 model.add_knowledge(*formulae, world=World.AXIOM)
     26 # add data to the model
---> 27 model.add_data({
     28     Friends: {
     29         ('Anna', 'Bob'): Fact.TRUE,
     30         ('Bob', 'Anna'): Fact.TRUE,
     31         ('Anna', 'Edward'): Fact.TRUE,
     32         ('Edward', 'Anna'): Fact.TRUE,
     33         ('Anna', 'Frank'): Fact.TRUE,
     34         ('Frank', 'Anna'): Fact.TRUE,
     35         ('Bob', 'Chris'): Fact.TRUE},
     36     Smokes.name: {
     37         'Anna': Fact.TRUE,
     38         'Edward': Fact.TRUE,
     39         'Frank': Fact.TRUE,
     40         'Gary': Fact.TRUE},
     41     Cancer.name: {
     42         'Anna': Fact.TRUE,
     43         'Edward': Fact.TRUE}
     44     })
     45 model.print()

File ~/src/pas-notebooks/env/lib64/python3.9/site-packages/lnn/model.py:346, in Model.add_data(self, data)
    344 for formula, fact in data.items():
    345     if not isinstance(formula, Formula):
--> 346         raise TypeError(
    347             "formula expected of type Formula, received "
    348             f"{formula.__class__.__name__}"
    349         )
    350     _exceptions.AssertFormulaInModel(self, formula)
    351     if formula.propositional:

TypeError: formula expected of type Formula, received str

Hi, We had a discussion with the IBM Neuro-Symbolic AI Team regarding contributing to LNN. One particular application type I became interested in is using them with traditional NN-based approaches. I suggest LNN repository should have some examples with public datasets like keras and Theseus. These examples help practitioners get a head-first experience of the framework. I look forward to contributing such one-click examples. Thanks.