def run(self):
global DEBUG_FLAG
data = self.preprocess_metafunctions()
if DEBUG_FLAG:
self.display_preprocessed_code(data)
model = PrologString(data)
if DEBUG_FLAG:
print '=' * 80
print "BEFORE ESCAPING"
print '=' * 80
for elem in model: print clause2str(elem)
print '=' * 80
model = escape_metafunctions(model)
if DEBUG_FLAG:
print '=' * 80
print "AFTER ESCAPING"
print '=' * 80
for elem in model: print clause2str(elem)
print '=' * 80
engine = DefaultEngine(label_all=True)
engine.add_builtin(METAMETAFUNCTION_FUNCTOR, 2, BooleanBuiltIn(self.builtin_metafunction))
engine.add_builtin('declare', 2, BooleanBuiltIn(self.builtin_declare))
engine.add_builtin('declare', 3, BooleanBuiltIn(self.builtin_declare))
db = engine.prepare(model)
if DEBUG_FLAG:
print "=" * 80
print "DATABASE"
print "=" * 80
for elem in db: print elem
print "=" * 80
gp = LogicFormula(
keep_all=True,
keep_order=True,
keep_duplicates=True,
avoid_name_clash=True
)
gp = engine.ground_all(db, target=gp)
if DEBUG_FLAG:
print "=" * 80
print "GROUND PROGRAM (GROUNDER)"
print "=" * 80
for elem in gp.enum_clauses():
print elem
print "=" * 80
clauses = []
facts = []
for clause in unescape_metafunctions(gp.enum_clauses()):
if isinstance(clause, Clause):
clauses.append(clause)
else:
facts.append(clause)
query_atoms = gp._names['query']
return self.builtin_declare.declarations, clauses + facts, query_atoms
def sample(self, model, queries=None):
"""Sample one assignment to the queries of the given model.
Returns a tuple containing:
queries : list of pairs (name, value)
facts: list of probabilistic facts with their sampled value (name, value)
probability: overall probability of the assignment
"""
engine = self
db = engine.prepare(model)
if queries == None:
queries = [q[0] for q in engine.query(db, Term('query', None))]
# if evidence == None :
# evidence = engine.query(db, Term( 'evidence', None, None ))
target = LogicFormula()
for query in queries:
target = engine.ground(db, query, target, label=target.LABEL_QUERY)
# for query in evidence :
# if str(query[1]) == 'true' :
# target = engine.ground(db, query[0], target, label=LABEL_EVIDENCE_POS)
# elif str(query[1]) == 'false' :
# target = engine.ground(db, query[0], target, label=LABEL_EVIDENCE_NEG)
# else :
# target = engine.ground(db, query[0], target, label=LABEL_EVIDENCE_MAYBE)
# Take into account remaining probabilities of no-choice nodes for annotated disjunctions.
for k, p in self.groups.items():
if p != None:
self.probability *= p
translate = lambda x: (x[0], x[1] == 0)
facts = []
for f, v in self.facts.items():
if v:
if type(f) == tuple:
node = db.get_node(f[0])
args = f[1]
else:
node = db.get_node(f)
args = node.args
if node.functor != 'query':
if args:
facts.append('%s(%s)' %
(node.functor, ', '.join(map(str, args))))
else:
facts.append('%s' % (node.functor))
result = map(translate, target.queries()), facts, self.probability
self.reset()
return result
def init_db(engine, model, propagate_evidence=False):
db = engine.prepare(model)
if propagate_evidence:
evidence = engine.query(db, Term("evidence", None, None))
evidence += engine.query(db, Term("evidence", None))
ev_target = LogicFormula()
engine.ground_evidence(db, ev_target, evidence)
ev_target.lookup_evidence = {}
ev_nodes = [
node for name, node in ev_target.evidence()
if node != 0 and node is not None
]
ev_target.propagate(ev_nodes, ev_target.lookup_evidence)
evidence_facts = []
for index, value in ev_target.lookup_evidence.items():
node = ev_target.get_node(index)
if ev_target.is_true(value):
evidence_facts.append((node[0], 1.0) + node[2:])
elif ev_target.is_false(value):
evidence_facts.append((node[0], 0.0) + node[2:])
else:
evidence_facts = []
ev_target = None
return db, evidence_facts, ev_target
def evaluate_custom_weights(self, eval_name=None):
class TestSemiringProbabilityNSP(SemiringProbability):
def is_nsp(self):
return True
program = """
0.25::a.
query(a).
"""
pl = PrologString(program)
lf = LogicFormula.create_from(pl,
label_all=True,
avoid_name_clash=True)
semiring = TestSemiringProbabilityNSP()
kc_class = get_evaluatable(name=eval_name, semiring=semiring)
kc = kc_class.create_from(lf)
a = Term('a')
# without custom weights
results = kc.evaluate(semiring=semiring)
self.assertEqual(0.25, results[a])
# with custom weights
weights = {a: 0.1}
results = kc.evaluate(semiring=semiring, weights=weights)
self.assertEqual(0.1, results[a])
def ground_problog_program(program):
""" Grounds a Problog program using the problog library. """
lf = LogicFormula.create_from(program,
avoid_name_clash=True,
keep_order=True,
label_all=True)
return lf.to_prolog()
def run_tests_with_static_methods():
from problog.program import PrologString
from problog.engine import DefaultEngine
from problog.logic import Term, Var
p = PrologString("""
coin(c1). coin(c2).
0.4::heads(C); 0.6::tails(C) :- coin(C).
win :- heads(C).
""")
qs, evs = ([Term("win")], [(Term("heads", Term("c1")), False)]) # For now
engine = DefaultEngine()
db = engine.prepare(p)
labels = (LogicFormula.LABEL_QUERY, LogicFormula.LABEL_EVIDENCE_POS,
LogicFormula.LABEL_EVIDENCE_NEG)
lf = LogicFormula()
lf = AMCQuery.ground_query_evidence(engine, db, qs, evs, lf, labels)
circuit = AMCQuery.compile_to_circuit(lf, "ddnnf")
prob_sr = SemiringProbability()
results, ground_evidence = AMCQuery.evaluate_circuit(
circuit, labels, prob_sr)
print("evidence: ", ground_evidence)
for r in results:
print(r)
print("---")
def find_all_prob():
ps = ""
with open("prolog/problog_predicates.pl", "r") as f:
for line in f:
ps += line
# Calcolo probabilità tramite problog
ps += "query(infect(_))."
p = PrologString(ps)
dbp = engine.prepare(p)
lf = LogicFormula.create_from(p) # ground the program
dag = LogicDAG.create_from(lf) # break cycles in the ground program
cnf = CNF.create_from(dag) # convert to CNF
ddnnf = DDNNF.create_from(cnf) # compile CNF to ddnnf
r = ddnnf.evaluate()
# Siccome Problog restituisce un dizionario struttrato in questa maniera:
# {query(infect(2)): 0.67, query(infect(3)): 0.8, ...}
# Bisogna estrarre ogni id dalla chiave nel seguente modo
items = []
if len(RedNode.query.all()) > 0:
for key, value in r.items():
start = "infect("
end = ")"
result = str(key)[len(start):-len(end)]
try:
u = User.query.get(int(result))
items.append((u, value))
except ValueError:
continue
return items
def main_mpe_semiring(args):
inputfile = args.inputfile
init_logger(args.verbose)
if args.web:
result_handler = print_result_json
else:
result_handler = print_result
if args.output is not None:
outf = open(args.output, 'w')
else:
outf = sys.stdout
with Timer("Total"):
try:
pl = PrologFile(inputfile)
lf = LogicFormula.create_from(model, label_all=True)
prob, facts = mpe_semiring(lf, args.verbose)
result_handler((True, (prob, facts)), outf)
except Exception as err:
trace = traceback.format_exc()
err.trace = trace
result_handler((False, err), outf)
def add_and(self, content, **kwdargs):
i = 0
for c in content:
if c is not None and c != 0:
i += 1
if i > 1:
raise ValueError("Can't combine sampled predicates.")
return LogicFormula.add_and(self, content)
def add_or(self, content, **kwd):
i = 0
for c in content:
if c is not None and c != 0:
i += 1
if i > 1:
raise ValueError(
"Bodies for same head should be mutually exclusive.")
return LogicFormula.add_or(self, content, **kwd)
def __init__(self, **kwargs):
LogicFormula.__init__(self, **kwargs)
self.facts = {}
self.groups = {}
self.probability = 1.0 # Try to compute
self.values = []
self.distributions = {
'normal': random.normalvariate,
'gaussian': random.normalvariate,
'poisson': sample_poisson,
'exponential': random.expovariate,
'beta': random.betavariate,
'gamma': random.gammavariate,
'uniform': random.uniform,
'triangular': random.triangular,
'vonmises': random.vonmisesvariate,
'weibull': random.weibullvariate,
}
def __init__(self, **kwargs):
LogicFormula.__init__(self, **kwargs)
self.facts = {}
self.groups = {}
self.probability = 1.0 # Try to compute
self.values = []
self.distributions = {
"normal": random.normalvariate,
"gaussian": random.normalvariate,
"poisson": sample_poisson,
"exponential": random.expovariate,
"beta": random.betavariate,
"gamma": random.gammavariate,
"uniform": random.uniform,
"triangular": random.triangular,
"vonmises": random.vonmisesvariate,
"weibull": random.weibullvariate,
"in_range": random.randint,
}
def main():
p = PrologString("""
increaseOsteoblasts :- calcium.
0.5::\+increaseOsteoblasts :- calcium, bispho.
reduceOsteoclasts :- bispho.
1.0::\+reduceOsteoclasts :- calcium , bispho.
osteoprosis :- initialOsteoprosis.
0.85::\+osteoprosis :- reduceOsteoclasts. % Bisphosphonates
0.15::\+osteoprosis :- increaseOsteoblasts. % Calcium
% Prior probabilities
0.5::calcium. 0.5::bispho. 0.5::initialOsteoprosis.
% Query probability of effect
evidence(initialOsteoprosis, true).
evidence(calcium, true).
evidence(bispho, false).
query(osteoprosis).
""")
#1.3: Create the CNF of the problog
lf = LogicFormula.create_from(p,avoid_name_clash=True, keep_order=True, label_all=True) # ground the program
print("Ground program")
print(LogicFormula.to_prolog(lf))
dag = LogicDAG.create_from(lf,avoid_name_clash=True, keep_order=True, label_all=True) # break cycles in the ground program
cnf = CNF.create_from(dag) # convert to CNF
print(CNF.to_dimacs(cnf))
ddnnf = DDNNF.create_from(cnf) # compile CNF to ddnnf
test = DDNNF.get_weights(ddnnf)
print(test)
print(ddnnf.evaluate())
#3.1: Create 4 interpretations
print("--Create 4 interpretations--")
interpretations = create_interpretations(p_without_evidence, 4)
for i in interpretations: print(i)
#3.2: Create 100, 1000, 10000 interpretations and estimate p_n
print("--Estimate parameters--")
estimate_parameters(100)
estimate_parameters(1000)
estimate_parameters(10000)
def call_theorem_prover(theorem_prover,
instance_id,
question_id,
theory,
assertion,
gold_label,
print_log=True):
"""Function that takes a single theory/assertion example and runs it through the theorem prover
to obtain a label. Returns the obtained label, elapsed time to solve it, and exception returned
by the engine, if any.
"""
obtained_result = False
millisecs_elapsed = 0
if print_log: print("=======ORIGINAL THEORY=========")
theory_as_txt = theory.program(theorem_prover)
if print_log: print(theory_as_txt)
theory.preprocess(theorem_prover)
theory_as_txt = theory.program(theorem_prover)
if theorem_prover == "problog":
assertion_lf = assertion.logical_form(theorem_prover, False)
assertion_lf = f"query({assertion_lf})."
program = f"{theory_as_txt}\n{assertion_lf}"
if print_log:
print("=======PROGRAM FROM PREPROCESSED THEORY=========")
print(program)
print("=======EXPECTED LABEL=========")
print(f" {gold_label}")
start_millisecs = current_milli_time()
try:
lf = LogicFormula.create_from(program) # ground the program
dag = LogicDAG.create_from(
lf) # break cycles in the ground program
sdd = SDD.create_from(dag)
result = sdd.evaluate()
end_millisecs = current_milli_time()
elapsed_millisecs = end_millisecs - start_millisecs
result_tuples = [(k, v) for k, v in result.items()]
obtained_result = result_tuples[0][1] != float(0)
return obtained_result, elapsed_millisecs, None
except (NegativeCycle, NonGroundProbabilisticClause,
UnknownClause) as e:
end_millisecs = current_milli_time()
elapsed_millisecs = end_millisecs - start_millisecs
if print_log:
print(
f"!!!Encountered Exception at instance id {instance_id}, question id {question_id}: {e}"
)
exception_name = str(type(e)).lstrip("<class '").rstrip("'>")
return None, elapsed_millisecs, exception_name
return obtained_result, elapsed_millisecs, None
def ad_atom_duplicate(self, eval_name=None):
"""
This test must pickup the case where during the transformation, additional _extra atoms are created because
add_atom(..., cr_extra=True) is used instead of cr_extra=False.
"""
program = """
0.2::a ; 0.8::b.
query(a).
query(b).
"""
pl = PrologString(program)
lf = LogicFormula.create_from(pl, label_all=True, avoid_name_clash=True)
semiring = SemiringProbability()
kc_class = get_evaluatable(name=eval_name, semiring=semiring)
kc = kc_class.create_from(lf) # type: LogicFormula
self.assertEqual(3, kc.atomcount)
def run_theory_in_problog(theory, assertion):
"""Run the given theory and assertion through ProbLog engine to obtain a True/False label.
If an exception is encountered, return None so that this example will not be part of output."""
theorem_prover = "problog"
try:
program = theory.program(theorem_prover, assertion)
lf = LogicFormula.create_from(program) # ground the program
dag = LogicDAG.create_from(lf) # break cycles in the ground program
sdd = SDD.create_from(dag)
result = sdd.evaluate()
result_tuples = [(k, v) for k, v in result.items()]
if len(result_tuples) == 0:
return False
return result_tuples[0][1] != 0.0
except (NegativeCycle, NonGroundProbabilisticClause, UnknownClause) as e:
return None
return None
def find_user_prob(uid):
ps = ""
with open("prolog/problog_predicates.pl", "r") as f:
for line in f:
ps += line
# Pulizia dei nodi dinamici date/1 all'interno di problog
p = PrologString(ps)
dbp = engine.prepare(p)
query = Term("clean")
res = engine.query(dbp, query)
# Calcolo probabilità tramite problog
ps += "query(infect(" + str(uid) + "))."
p = PrologString(ps)
dbp = engine.prepare(p)
lf = LogicFormula.create_from(p) # ground the program
dag = LogicDAG.create_from(lf) # break cycles in the ground program
cnf = CNF.create_from(dag) # convert to CNF
ddnnf = DDNNF.create_from(cnf) # compile CNF to ddnnf
r = ddnnf.evaluate()
# Salvataggio nel database SQLite della data del nodo rosso più vecchio con cui è stato a contatto
term = Term("date", None)
database = problog_export.database # Database interno di Problog dove vengono salvati i fatti con assertz()
node_key = database.find(term)
if node_key is not None:
node = database.get_node(node_key)
dates = node.children.find(
term.args) # Tutti i fatti date/1 inseriti con assertz/1
vals = []
if dates:
for date in dates:
n = database.get_node(date)
vals.append(int(n.args[0]))
min_val = min(vals) # Trova la data (in millisecondi) minima
u = User.query.get(uid)
u.oldest_risk_date = min_val
db.session.commit()
return r
def __init__(self, db):
self.db = db
self.lf = LogicFormula()
self._current_version = 0
self._compiled = {}
def evaluate_custom_weights(self, eval_name=None):
class TestSemiringProbabilityNSP(SemiringProbability):
def is_nsp(self):
return True
def pos_value(self, a, key=None):
if isinstance(a, tuple):
return float(a[0])
else:
return float(a)
def neg_value(self, a, key=None):
if isinstance(a, tuple):
return float(a[1])
else:
return 1 - float(a)
program = """
0.25::a.
query(a).
"""
pl = PrologString(program)
lf = LogicFormula.create_from(pl,
label_all=True,
avoid_name_clash=True)
semiring = TestSemiringProbabilityNSP()
kc_class = get_evaluatable(name=eval_name, semiring=semiring)
kc = kc_class.create_from(lf)
a = Term('a')
# without custom weights
results = kc.evaluate(semiring=semiring)
self.assertEqual(0.25, results[a])
# with custom weights
weights = {a: 0.1}
results = kc.evaluate(semiring=semiring, weights=weights)
self.assertEqual(0.1, results[a])
# with custom weights
weights = {a: (0.1, 0.1)}
results = kc.evaluate(semiring=semiring, weights=weights)
self.assertEqual(0.5, results[a])
# with custom weights based on index
weights = {kc.get_node_by_name(a): 0.2}
results = kc.evaluate(semiring=semiring, weights=weights)
self.assertEqual(0.2, results[a])
# Testing with weight on node 0 (True)
weights = {0: 0.3, a: (0.1, 0.1)}
results = kc.evaluate(semiring=semiring, weights=weights)
self.assertEqual(0.5, results[a])
# Testing query on node 0 (True)
class TestSemiringProbabilityIgnoreNormalize(SemiringProbabilityNSPCopy
):
def normalize(self, a, z):
return a
weights = {0: (0.3, 0.7), a: (0.1, 0.1)}
results = kc.evaluate(
index=0,
semiring=TestSemiringProbabilityIgnoreNormalize(),
weights=weights)
self.assertEqual(0.06, results)
def __init__(self) :
LogicFormula.__init__(self)
self.facts = {}
self.groups = {}
self.probability = 1.0
m = model.read()
times = []
door_num = range(3, 10)
for i in door_num:
start = timeit.default_timer()
model = m.format(door_num=i)
p = PrologString(model)
formula = get_evaluatable().create_from(p)
print(formula.evaluate())
stop = timeit.default_timer()
times.append(stop - start)
for i in door_num:
model = m.format(door_num=i)
p = PrologString(model)
lf = LogicFormula.create_from(p)
lfs.append(lf)
dag = LogicDAG.create_from(lf)
dags.append(dag)
cnf = CNF.create_from(dag)
cnfs.append(cnf)
for i in door_num:
model = m.format(door_num=i)
p = PrologString(model)
lf = LogicFormula.create_from(p)
lfs.append(lf)
dag = LogicDAG.create_from(lf)
dags.append(dag)
cnf = CNF.create_from(dag)
cnfs.append(cnf)
def evaluate(self, examples, engine):
# UITLEG subst
# bij elke refinement wordt er 1 predicaat toegevoegd aan de body of head (in ons geval enkel aan de body, aangezien we enkel in de modes '\+' gebruiken)
# het toegevoegde predicaat zit vervat in 'rule_literal'
# elke failing bevat een aantal lijsten, subst is telkens een pointer naar een van deze lijsten
# het i-de element van subst is telkens de waarde van parameter Ai
# Bv example: p(a). p(b). q(a,b)
# huidige hypothese: p(A1) -> false
# ----> rule_literal is dus p(A1)!!
# de originele failing zou dan zijn [[]] (zie evaluate van EmptyClause)
# de nieuwe failing is nu [[a],[b]] aangezien er in het voorbeeld twee keer p(A1) voorkomt
# aangezien de body nu true en de head false (want die is altijd false in dit voorbeeld) dekt dit het voorbeeld niet
# daarom staat er een 'if not head_literal and i == None' wat betekent dat als de rule_literal een body_literal is en als die true is in het voorbeeld
# dan wordt het voorbeeld niet gedekt
# gather the failings from the parent
# the failings list contains for every example information about that example
# if a failing element is just an empty list ([]) then the clause covers that example
if self.parent.failings is not None:
failings = self.parent.failings
else:
self.parent.evaluate(examples, engine)
failings = self.parent.failings
# check whether the literal is a body or a head literal
# body literals are negated
rule_literal = abs(self.literal)
if not self.literal.is_negated():
head_literal = True
else:
head_literal = False
# boolean indicating whether there is an example which is not covered by this clause
fails = False
# the list that will contain the updated failings
# later on self.failings will be set to this list
new_failings = []
# boolean indicating whether this clause is an improvement of its parent clause
improvement = False
if head_literal:
improvement = True
# loop through the examples and failings at the same time
# 'example' contains the example and 'failing' contains information about coverage of the example by the parent
for k, (example, failing) in enumerate(zip(examples, failings)):
#print("-----------------------------------------------------")
#print("testing",self,"on example",k+1)
if failing: # equivalent to if 'failing != []' so if this if succeeds, the parent does not cover the example
# TODO: find out why this is necessary
if hasattr(example, '_ground'):
formula = example._ground
formula.clear_queries()
else:
formula = LogicFormula() # Load from cache
# formula = LogicFormula()
# construct a new list to store the failing information for this example
# later on 'failing' will be replaced by 'new_failing'
new_failing = []
# calculate the contents of new_failing
for subst in failing:
#print("rule_literal:",rule_literal)
#print("original subst:",subst)
# for every new variable this clause introduces, add a None element to the substances
if len(subst) < self.variable_count:
subst = subst + [None
] * (self.variable_count - len(subst))
improvement = True # adds a variable
#print("new subst:",subst)
# subst now contains a list where the ith element contains the value for the ith parameter of the rule_literal
# bv. instantiate(pred(A1,A2,A3),[a,b,None]) would result in pred(a,b,_)
literal = instantiate(rule_literal, subst)
#print("literal:",literal)
# formula contains a list queries
# this list contains all the facts in the example of the literal
# bv example: pred1(a). pred1(b). pred2(b). pred3(a,b).
# literal = pred1(_)
# queries: [pred1(a),pred1(b)]
formula = engine.ground(example.database,
literal,
target=formula,
label='query')
#print("formula:",formula)
# q is the query
# if i == 0, the query is true
# if i == None, the query is false
# de uitbreiding van een clause dekt het positieve voorbeeld als rule_literal een body_literal is die naar false evalueert of
# of als rule_literal een head_literal is die naar true evalueert
for q, i in formula.queries():
# create a copy of the subst_new
subst_new = subst[:]
#print("q:",q)
try:
# try to unify the query with the rule literal and put the unifications in subst_new
unify(q, rule_literal, subst_new)
#print("subst_new:",subst_new)
if len(set(subst_new)) == len(subst_new):
# TODO check subst_new is all different
# the head_literal evaluated to false
if head_literal and i is None:
if None in subst_new:
get_logger('claudette').error(' '.join(
map(str,
(list(example.database), self,
failing, formula))))
raise RuntimeError(
'This shouldn\'t happen!')
#print("head query",q,"evaluated to false")
new_failing.append(subst_new)
# the body literal evaluated in true
elif not head_literal and i == 0:
new_failing.append(subst_new)
#print("body query",q,"evaluated to true")
except UnifyError:
# if we have a unify error then the literal is not unifyable with the query
#print("unify error")
pass
#print("==========")
example._ground = formula
new_failings.append(new_failing)
#print("failing:",failing)
#print("new_failing:",new_failing)
# we check whether new_failing contains any items
# if it doesn't then we know that the refinement covers the example
if new_failing:
# if new_failing contains less items than failing than this clause covers it 'more' than the parent so there is an improvement
if len(new_failing) < len(
failing
) or self.variable_count == 0: # the last statement is needed for terminal (otherwise terminal gives no contribution)
improvement = True
fails = True
else: # new_failing == [] thus this example is now covered
assert failing
improvement = True # failing eliminated
else:
# the parent covers this example, so the refinement covers this to
new_failings.append([])
self.failings = new_failings
self.improvement = improvement
# if fails is set to True, then the clause didn't cover a positive example so it is not a valid one and needs to be refined
return not fails
def main():
program = PrologFile('model.txt')
formula = LogicFormula.create_from(program)
modelEvaluatable = get_evaluatable().create_from(formula)
print modelEvaluatable.evaluate()
#
p2 = PrologString("""
person(a).
person(b).
person(c).
0.2::stress(X) :- person(X).
0.1::friends(X,Y) :- person(X), person(Y).
0.3::smokes(X) :- stress(X).
0.4::smokes(X) :- friends(X,Y), smokes(Y).
evidence(friends(a,b), true).
evidence(friends(a,c), true).
query(smokes(a)).
""")
lf2 = LogicFormula.create_from(p2,
avoid_name_clash=True,
keep_order=True,
label_all=True)
# print(LogicFormula.to_prolog(lf2))
dag2 = LogicDAG.create_from(lf2,
avoid_name_clash=False,
keep_order=True,
label_all=True)
# # print(dag2)
# # print(LogicFormula.to_prolog(dag2))
cnf2 = CNF.create_from(dag2)
# # print(cnf2.to_dimacs(weighted=True, invert_weights=True))
ddnnf2 = DDNNF.create_from(cnf2)
#print(ddnnf2.evaluate())
#
# import PyBool_public_interface as Bool
def verify(self, examples, engine):
if self.parent.verifications is not None:
verifications = self.parent.verifications
else:
self.parent.verify(examples, engine)
verifications = self.parent.verifications
# check whether the literal is a body or a head literal
# body literals are negated
rule_literal = abs(self.literal)
if not self.literal.is_negated():
head_literal = True
else:
head_literal = False
verifies = False
new_verifications = []
for k, (example,
verification) in enumerate(zip(examples, verifications)):
#print("-----------------------------------------------------")
#print("testing",self,"on example",k+1)
if verification:
if hasattr(example, '_ground'):
formula = example._ground
formula.clear_queries()
else:
formula = LogicFormula()
new_verification = []
for subst in verification:
#print("rule_literal:",rule_literal)
#print("original subst:",subst)
if len(subst) < self.variable_count:
subst = subst + [None
] * (self.variable_count - len(subst))
#print("new subst:",subst)
literal = instantiate(rule_literal, subst)
#print("literal:",literal)
formula = engine.ground(example.database,
literal,
target=formula,
label='query')
#print("formula:",formula)
for q, i in formula.queries():
subst_new = subst[:]
#print("q:",q)
try:
unify(q, rule_literal, subst_new)
#print("subst_new:",subst_new)
#if len(set(subst_new)) == len(subst_new):
if head_literal and i == 0:
if None in subst_new:
get_logger('claudette').error(' '.join(
map(str, (list(example.database), self,
failing, formula))))
raise RuntimeError(
'This shouldn\'t happen!')
#print("head query",q,"evaluated to true")
new_verification.append(subst_new)
# the body literal evaluated in true
elif not head_literal and i == 0:
new_verification.append(subst_new)
#print("body query",q,"evaluated to true")
except UnifyError:
# if we have a unify error then the literal is not unifyable with the query
#print("unify error")
pass
#print("==========")
example._ground = formula
if str(rule_literal)[0] == '#':
new_verifications.append([[]])
else:
new_verifications.append(new_verification)
#print("verification:",verification)
#print("new_verification:",new_verification)
if new_verification:
verifies = True
else:
new_verifications.append([])
self.verifications = new_verifications
return verifies
from problog.program import PrologString
from problog.formula import LogicFormula, LogicDAG
from problog.logic import Term
from problog.ddnnf_formula import DDNNF
from problog.cnf_formula import CNF
p = PrologString("""
coin(c1). coin(c2).
0.4::heads(C); 0.6::tails(C) :- coin(C).
win :- heads(C).
evidence(heads(c1), false).
query(win).
query(coin(X)).
""")
lf = LogicFormula.create_from(p) # ground the program
dag = LogicDAG.create_from(lf) # break cycles in the ground program
cnf = CNF.create_from(dag) # convert to CNF
ddnnf = DDNNF.create_from(cnf) # compile CNF to ddnnf
results = ddnnf.evaluate()
print(results)
# 0.4::b.
# 0.3::a :- b.
# 0.5::b :- a.
# query(a).
# evidence(b).
# """)
p = PrologString("""
0.4::heads(1). 0.7::heads(2). 0.5::heads(3).
win :- heads(1).
win :- heads(2), heads(3).
query(win).
""")
lf = LogicFormula.create_from(p,
avoid_name_clash=True,
keep_order=True,
label_all=True) # ground the program
# print(lf)
# print(LogicFormula.to_prolog(lf))
dag = LogicDAG.create_from(
lf, avoid_name_clash=True, keep_order=True,
label_all=True) # break cycles in the ground program
# print(dag)
# print(LogicFormula.to_prolog(dag))
cnf = CNF.create_from(dag) # convert to CNF
# for clause in cnf._clauses:
# print(clause)
ddnnf = DDNNF.create_from(cnf) # compile CNF to ddnnf
# Outcome for the a/b thing with query(a) is 0,2+(0,8*0,3*0,4)
# but if evidence(b) : (0,2*0,4+0,2*0,5*0,6+0,4*0,3*0,8) / (0,4+0,6*0,5*0,2) [Formula for conditional probability P(a|b) = ...]
# For coin thing : 1-0,6*0,3*0,5-0,6*0,3*0,5-0,6*0,7*0,5 = 0,61
return new_weights, iters
# Read in the progam with tunable parameters
#ground_progam = file_to_string("test_learn.pl")
# Assign to each tunable parameter a random initialization value
#ground_progam_tunable = init_tunable(ground_progam)
# Write back the prolog file with random initialized tunable probabilities
#f = open('test_learn_tunable.pl', 'w')
#f.write(ground_progam_tunable)
pf = PrologFile("test_learn_tunable.pl")
formula = LogicFormula.create_from(pf)
sdd = SDD.create_from(formula)
queries = [
Term('stress', Term('a')),
Term('stress', Term('b')),
Term('stress', Term('c')),
Term('smokes', Term('a')),
Term('smokes', Term('b')),
Term('smokes', Term('c'))
]
# File which holds all the evidence examples
examples = file_to_string('data.pl')
# Retrieve the amount of interpretations that has been specified