def even_numbers_test():
B = [Atom([Term(False, '0')], 'zero')] + \
[Atom([Term(False, str(i)), Term(False, str(i + 1))], 'succ')
for i in range(0, 20)]
P = [Atom([Term(False, str(i))], 'target') for i in range(0, 21, 2)]
N = [Atom([Term(False, str(i))], 'target') for i in range(1, 21, 2)]
print(P)
term_x_0 = Term(True, 'X_0')
term_x_1 = Term(True, 'X_1')
p_e = [Atom([term_x_0], 'zero'), Atom([term_x_0, term_x_1], 'succ')]
p_a = [Atom([term_x_0, term_x_1], 'pred')]
target = Atom([term_x_0], 'target')
constants = [str(i) for i in range(0, 21)]
# Define rules for intensional predicates
p_a_rule = (Rule_Template(1, False), None)
target_rule = (Rule_Template(0, False), Rule_Template(1, True))
rules = {p_a[0]: p_a_rule, target: target_rule}
langage_frame = Language_Frame(target, p_e, constants)
program_template = Program_Template(p_a, rules, 10)
#program_template = Program_Template(p_a, rules, 300)
dilp = DILP(langage_frame, B, P, N, program_template)
dilp.train()
def less_than():
B = [Atom([Term(False, '0')], 'zero')] + \
[Atom([Term(False, str(i)), Term(False, str(i + 1))], 'succ')
for i in range(0, 9)]
P = []
N = []
for i in range(0, 10):
for j in range(0, 10):
if j >= i:
N.append(
Atom([Term(False, str(j)), Term(False, str(i))], 'target'))
else:
P.append(
Atom([Term(False, str(j)), Term(False, str(i))], 'target'))
term_x_0 = Term(True, 'X_0')
term_x_1 = Term(True, 'X_1')
p_e = [Atom([term_x_0], 'zero'), Atom([term_x_0, term_x_1], 'succ')]
p_a = []
target = Atom([term_x_0, term_x_1], 'target')
# target_rule = (Rule_Template(0, False), Rule_Template(1, True))
target_rule = (Rule_Template(0, False), Rule_Template(1, True))
rules = {target: target_rule}
constants = [str(i) for i in range(0, 10)]
langage_frame = Language_Frame(target, p_e, constants)
program_template = Program_Template(p_a, rules, 10)
dilp = DILP(langage_frame, B, P, N, program_template)
dilp.train()
def action_up_2():
B_0, P_up, P_sw, P_stay, N_up, N_sw, N_stay = get_B_N_P()
B = [Atom([Term(False, str(x[1]))], x[0]) for x in B_0]
N = [Atom([Term(False, str(id))], 'up') for id in P_up]
P = [Atom([Term(False, str(id))], 'up') for id in N_up]
term_x_0 = Term(True, 'X')
term_x_1 = Term(True, 'X1')
p_e = [
Atom([term_x_0], 'busy'),
Atom([term_x_0], 'NObusy'),
Atom([term_x_0], 'side'),
Atom([term_x_0], 'front'),
Atom([term_x_0], 'back')
]
p_a = []
target = Atom([term_x_0], 'up')
# target_rule = (Rule_Template(0, False), Rule_Template(1, True))
target_rule = (Rule_Template(0, False), None)
rules = {target: target_rule}
constants = [str(i) for i in range(0, 400)]
langage_frame = Language_Frame(target, p_e, constants)
program_template = Program_Template(p_a, rules, 10)
dilp = DILP(langage_frame, B, P, N, program_template)
dilp.train(steps=250)
def action_switch_2():
num = 30
inx = random.sample(range(0, 100), num)
neg_inx = list(set(range(100)) - set(inx))
B = [Atom([Term(False, str(id))], 'busy') for id in inx
] + [Atom([Term(False, str(id))], 'NObusy') for id in neg_inx] + [
Atom([Term(False, str(id))], 'front')
for id in neg_inx if id % 2 == 0
] + [
Atom([Term(False, str(id))], 'back')
for id in neg_inx if id % 2 != 0
]
P = [Atom([Term(False, str(id))], 'switch') for id in neg_inx]
N = [Atom([Term(False, str(id))], 'switch') for id in inx]
term_x_0 = Term(True, 'X')
term_x_1 = Term(True, 'X1')
p_e = [
Atom([term_x_0], 'busy'),
Atom([term_x_0], 'NObusy'),
Atom([term_x_0], 'side'),
Atom([term_x_0], 'front'),
Atom([term_x_0], 'back')
]
p_a = []
target = Atom([term_x_0], 'switch')
# target_rule = (Rule_Template(0, False), Rule_Template(1, True))
target_rule = (Rule_Template(0, False), None)
rules = {target: target_rule}
constants = [str(i) for i in range(0, 100)]
langage_frame = Language_Frame(target, p_e, constants)
program_template = Program_Template(p_a, rules, 10)
dilp = DILP(langage_frame, B, P, N, program_template)
dilp.train(steps=252)
def action_sw(bg_all):
#num = 70
#inx=random.sample(range(0,100),num)
#neg_inx = list(set(range(100))-set(inx))
B_0, P_up, P_sw, P_stay, N_up, N_sw, N_stay = get_B_N_P(bg_all)
B = [Atom([Term(False, str(x[1]))], x[0]) for x in B_0]
P = [Atom([Term(False, str(id))], 'sw') for id in P_sw]
N = [Atom([Term(False, str(id))], 'sw') for id in N_sw]
term_x_0 = Term(True, 'X')
term_x_1 = Term(True, 'X1')
p_e = [
Atom([term_x_0], 'busy'),
Atom([term_x_0], 'NObusy'),
Atom([term_x_0], 'side'),
Atom([term_x_0], 'front'),
Atom([term_x_0], 'back')
]
p_a = []
target = Atom([term_x_0], 'sw')
# target_rule = (Rule_Template(0, False), Rule_Template(1, True))
target_rule = (Rule_Template(0, False), None)
rules = {target: target_rule}
constants = [str(i) for i in range(0, 400)]
langage_frame = Language_Frame(target, p_e, constants)
program_template = Program_Template(p_a, rules, 10)
dilp = DILP(langage_frame, B, P, N, program_template)
loss, cl_set = dilp.train(steps=51)
return loss, cl_set
def is_cyclic():
'''
Learn the target predicate no_negative_cycle(X). The predicate checks if
a given node in a graph is part of a negative cycle (cycle with at least one
negative edge).
'''
constants = ['a', 'b', 'c', 'd', 'e', 'f']
### (B, P, N) Number 1
B_atom = [
Atom([Term(False, 'a'), Term(False, 'b')], 'edge'),
Atom([Term(False, 'b'), Term(False, 'c')], 'edge'),
Atom([Term(False, 'c'), Term(False, 'a')], 'edge'),
Atom([Term(False, 'b'), Term(False, 'd')], 'edge'),
Atom([Term(False, 'd'), Term(False, 'e')], 'edge'),
Atom([Term(False, 'd'), Term(False, 'f')], 'edge'),
Atom([Term(False, 'f'), Term(False, 'e')], 'edge'),
Atom([Term(False, 'e'), Term(False, 'f')], 'edge')
]
B = [Literal(atom, False) for atom in B_atom]
P_atom = [
Atom([Term(False, 'a')], 'target'),
Atom([Term(False, 'b')], 'target'),
Atom([Term(False, 'c')], 'target'),
Atom([Term(False, 'e')], 'target'),
Atom([Term(False, 'f')], 'target')
]
P = [Literal(atom, False) for atom in P_atom]
N_atom = [Atom([Term(False, 'd')], 'target')]
N = [Literal(atom, False) for atom in N_atom]
term_x_0 = Term(True, 'X_0')
term_x_1 = Term(True, 'X_1')
term_x_2 = Term(True, 'X_2')
p_e = [Literal(Atom([term_x_0, term_x_1], 'edge'), False)]
p_a = [Literal(Atom([term_x_0, term_x_2], 'pred'), False)]
target = Literal(Atom([term_x_0], 'target'), False)
# Define rules for intensional predicates
p_a_rule = [(Rule_Template_Negation(0, False, False),
Rule_Template_Negation(1, True, False))]
target_rule = (Rule_Template_Negation(0, True, False), None)
rules = {p_a[0]: p_a_rule[0], target: target_rule}
language_frame = Language_Frame(target, p_e, constants)
program_template = Program_Template(p_a, rules, 10)
# program_template = Program_Template(p_a, rules, 300)
dilp = DILP(language_frame, B, P, N, program_template)
return dilp.train(steps=300)
def action_switch():
#num = 70
#inx=random.sample(range(0,100),num)
#neg_inx = list(set(range(100))-set(inx))
B_0, P_up, P_sw, P_stay, N_up, N_sw, N_stay = get_B_N_P()
B = [Atom([Term(False, str(x[1]))], x[0]) for x in B_0]
# B = [Atom([Term(False, str(id))], 'busy')
# for id in inx]+ [Atom([Term(False, str(id))], 'NObusy')
# for id in neg_inx]
# print("B..........")
# print(B)
P = [Atom([Term(False, str(id))], 'sw') for id in P_sw]
N = [] #[Atom([Term(False, str(id))], 'sw')
#for id in N_sw]
# print("P..........")
# print(P)
# print("N..........")
# print(N)
term_x_0 = Term(True, 'X')
term_x_1 = Term(True, 'X1')
p_e = [
Atom([term_x_0], 'busy'),
Atom([term_x_0], 'NObusy'),
Atom([term_x_0], 'side'),
Atom([term_x_0], 'front'),
Atom([term_x_0], 'back')
]
#p_a = []
p_a = [Atom([term_x_0], 'pred')]
target = Atom([term_x_0], 'sw')
# target_rule = (Rule_Template(0, False), Rule_Template(1, True))
target_rule = (Rule_Template(0, True), Rule_Template(0, False))
p_a_rule = (Rule_Template(0, False), None)
rules = {p_a[0]: p_a_rule, target: target_rule}
constants = [str(i) for i in range(0, 400)]
langage_frame = Language_Frame(target, p_e, constants)
program_template = Program_Template(p_a, rules, 10)
dilp = DILP(langage_frame, B, P, N, program_template)
dilp.train(steps=252)
def even_numbers_negation_test():
B_atom = [Atom([Term(False, '0')], 'zero')] + \
[Atom([Term(False, str(i)), Term(False, str(i + 1))], 'succ')
for i in range(0, 20)]
B = [Literal(atom, False) for atom in B_atom]
P_atom = [Atom([Term(False, str(i))], 'target') for i in range(0, 21, 2)]
P = [Literal(atom, False) for atom in P_atom]
N_atom = [Atom([Term(False, str(i))], 'target') for i in range(1, 21, 2)]
N = [Literal(atom, False) for atom in N_atom]
term_x_0 = Term(True, 'X_0')
term_x_1 = Term(True, 'X_1')
p_e = [
Literal(Atom([term_x_0], 'zero'), False),
Literal(Atom([term_x_0, term_x_1], 'succ'), False)
]
p_a = [Literal(Atom([term_x_0, term_x_1], 'pred'), False)]
target = Literal(Atom([term_x_0], 'target'), False)
constants = [str(i) for i in range(0, 21)]
# Define rules for intensional predicates
p_a_rule = (Rule_Template_Negation(1, False, False), None)
target_rule = (Rule_Template_Negation(0, False, True),
Rule_Template_Negation(1, True, False))
rules = {p_a[0]: p_a_rule, target: target_rule}
langage_frame = Language_Frame(target, p_e, constants)
program_template = Program_Template(p_a, rules, 10)
#program_template = Program_Template(p_a, rules, 300)
dilp = DILP(langage_frame, B, P, N, program_template)
loss = dilp.train(steps=300)
dilp.show_definition()
return loss
def generate_stratified_dilps(self):
'''
Construct n DILP objects. All objects have clauses seperated into
strata to ensure stratification.
:return:
'''
rules = self.program_template.rules
n_auxiliary_predicates = len(
rules) - 1 # Subtract 1 for target predicate
# If only target predicate then no risk for stratification
if n_auxiliary_predicates <= 0:
return [
DILP(self.language_frame, self.background, self.positive,
self.negative, self.program_template)
]
target = self.language_frame.target
rules = rules.copy()
rules.pop(target)
keys = list(rules.keys())
if n_auxiliary_predicates == 1:
strata1 = {keys[0]: 1, target: 2}
dilp1 = self.create_DILP(strata1)
# Removed all predicates in same strata. No need
return [dilp1]
if n_auxiliary_predicates == 2:
strata1 = {keys[0]: 2, keys[1]: 1, target: 2}
strata2 = {keys[0]: 1, keys[1]: 2, target: 2}
strata3 = {keys[0]: 1, keys[1]: 1, target: 2}
strata4 = {keys[0]: 1, keys[1]: 2, target: 3}
strata5 = {keys[0]: 2, keys[1]: 1, target: 3}
dilp1 = self.create_DILP(strata1)
dilp2 = self.create_DILP(strata2)
dilp3 = self.create_DILP(strata3)
dilp4 = self.create_DILP(strata4)
dilp5 = self.create_DILP(strata5)
return [dilp1, dilp2, dilp3, dilp4, dilp5]
B, pred_f, constants_f = process_file('%s/facts.dilp' % sys.argv[1])
P, target_p, constants_p = process_file('%s/positive.dilp' % sys.argv[1])
N, target_n, constants_n = process_file('%s/negative.dilp' % sys.argv[1])
if not (target_p == target_n):
raise Exception('Positive and Negative files have different targets')
elif not len(target_p) == 1:
raise Exception('Can learn only one predicate at a time')
elif not constants_n.issubset(constants_f) or not constants_p.issubset(
constants_f):
raise Exception(
'Constants not in fact file exists in positive/negative file')
term_x_0 = Term(True, 'X_0')
term_x_1 = Term(True, 'X_1')
p_e = list(pred_f)
p_a = [Atom([term_x_0, term_x_1], 'pred')]
target = list(target_p)[0]
constants = constants_f
p_a_rule = (Rule_Template(1, False), None)
target_rule = (Rule_Template(0, False), Rule_Template(1, True))
rules = {p_a[0]: p_a_rule, target: target_rule}
langage_frame = Language_Frame(target, p_e, constants)
program_template = Program_Template(p_a, rules, 10)
dilp = DILP(langage_frame, B, P, N, program_template)
dilp.train()
def create_DILP(self, strata):
clause_parameters = self.create_clause_parameters(strata)
return DILP(self.language_frame, self.background, self.positive,
self.negative, self.program_template, clause_parameters)
def no_ancestor():
'''
Learn the target predicate of ancestor(X)
which is true if X has no ancestor
'''
constants = [
'Paul', 'Randy', 'Rachel', 'Bob', 'Alice', 'Stan', 'Kyle', 'Peter',
'Tony', 'Monica', 'Jessica', 'Duncan'
]
B_atom = [
Atom([Term(False, 'Bob'), Term(False, 'Paul')], 'parent'),
Atom([Term(False, 'Bob'), Term(False, 'Alice')], 'parent'),
Atom([Term(False, 'Randy'), Term(False, 'Tony')], 'parent'),
Atom([Term(False, 'Randy'), Term(False, 'Stan')], 'parent'),
Atom([Term(False, 'Peter'), Term(False, 'Kyle')], 'parent'),
Atom([Term(False, 'Alice'), Term(False, 'Kyle')], 'parent'),
Atom([Term(False, 'Jessica'),
Term(False, 'Bob')], 'parent'),
Atom([Term(False, 'Jessica'),
Term(False, 'Randy')], 'parent'),
Atom([Term(False, 'Matthew'),
Term(False, 'Jessica')], 'parent')
]
B = [Literal(atom, False) for atom in B_atom]
P_atom = [
Atom([Term(False, 'Peter')], 'target'),
Atom([Term(False, 'Monica')], 'target'),
Atom([Term(False, 'Rachel')], 'target'),
Atom([Term(False, 'Jessica')], 'target'),
Atom([Term(False, 'Duncan')], 'target'),
Atom([Term(False, 'Matthew')], 'target')
]
positive_literal_P = [Literal(atom, False) for atom in P_atom]
N_atom = [
Atom([Term(False, 'Randy')], 'target'),
Atom([Term(False, 'Bob')], 'target'),
Atom([Term(False, 'Alice')], 'target'),
Atom([Term(False, 'Kyle')], 'target'),
Atom([Term(False, 'Tony')], 'target'),
Atom([Term(False, 'Stan')], 'target'),
Atom([Term(False, 'Paul')], 'target')
]
positive_literal_N = [Literal(atom, False) for atom in N_atom]
negative_literal_P = [Literal(atom, True) for atom in N_atom]
negative_literal_N = [Literal(atom, True) for atom in P_atom]
P = positive_literal_P + negative_literal_P
N = positive_literal_N + negative_literal_N
term_x_0 = Term(True, 'X_0')
term_x_1 = Term(True, 'X_1')
p_e = [Atom([term_x_0, term_x_1], 'parent')]
p_a = []
target = Atom([term_x_0], 'target')
# Define rules for intensional predicates
p_a_rule = (None, None)
target_rule = (Rule_Template_Negation(1, False, True), None)
rules = {target: target_rule}
langage_frame = Language_Frame(target, p_e, constants)
program_template = Program_Template(p_a, rules, 10)
#program_template = Program_Template(p_a, rules, 300)
dilp = DILP(langage_frame, B, P, N, program_template)
dilp.train(steps=200)