def learn_concept(self, word, percept_data):
""" learn a concept (word + percept)
"""
# if word is already know, update perceptual data
if self.lex.check_word(word):
cs_tag = self.cs_lex.get_h_tag(word)
self.cs.get_concept(cs_tag).add_data(percept_data)
# if word is not known, create new concept
else:
self.lex.add_word(word) # create word in lexicon
self.lex_lex.create_node(word) # create lex node in lex_lex
self.lex_lex.add_link(word, word) # add link between lex and lex node
cs_list = [] # create list of percept tags
for i in percept_data:
cs_tag = "no_tag"
while ((cs_tag == "no_tag") or (cs_tag in self.cs.concept_tags)): # make sure the cs_tag is unique
cs_tag = auks.generateRandomTag(6)
cs_list.append(cs_tag)
self.cs.create_pconcept(cs_tag, [i]) # create pconcept in CS
self.cs_cs.create_node(cs_tag) # create cs node in cs_cs
self.cs_cs.add_link(cs_tag, cs_tag) # add link between cs and cs node
self.cs_lex.add_link(cs_tag, word) # add link between lex and cs matrix
self.n_percepts = len(self.cs.concepts) # update numbers
self.n_words = len(self.lex.words)
if len(cs_list) > 1 and cfg.associative_object_learning:
while len(cs_list) > 1:
self.cs_cs.add_link(cs_list[0], cs_list[1])
self.cs_cs.add_link(cs_list[1], cs_list[0])
cs_list = cs_list[1:]
def discrimination_game(agent, context, topic_index):
""" Discrimination game in which an agent has to distinguish the topic
from the context. The game succeeds if the agent has a concept
which uniquely matches the topic and no other stimuli from the context.
If this is not the case, a new concept is build, or the existing concepts
are shifted.
context = sets of data [ [[ "domain", [ [d1, value], [d2, value], ..., [dn, value] ]], ..., ]]
[ [["c", [ ["r",1.0], ["g",0.0], ["b",0.0]]]]]
"""
# if agent knowledge is empty, create new percept and concept
if agent.cs.concepts == []:
cs_tag = auks.generateRandomTag(6)
agent.cs.create_pconcept(cs_tag, context[topic_index])
agent.cs_cs.create_node(cs_tag) # create cs node in cs_cs matrix
agent.cs_cs.add_link(cs_tag, cs_tag) # add link between cs and cs nod
answer = cs_tag
else:
best_matching_percepts = []
for i in context:
best_matching_percepts.append(agent.cs.get_closest_concept_tag(i))
# determine the outcome of the discrimination game
if best_matching_percepts.count(best_matching_percepts[topic_index]) == 1:
success = True
agent.n_success_dg += 1.0
answer = best_matching_percepts[topic_index]
agent.cs.set_concept_disc_use(best_matching_percepts[topic_index], 1)
else:
agent.cs.set_concept_disc_use(best_matching_percepts[topic_index], 0)
# if agent discrimination success is below threshold a new percept is created
if agent.discrimination_success < cfg.adapt_threshold:
cs_tag = auks.generateRandomTag(6)
agent.cs.create_pconcept(cs_tag, context[topic_index])
agent.cs_cs.create_node(cs_tag) # create cs node in cs_cs matrix
agent.cs_cs.add_link(cs_tag, cs_tag) # add link between cs and cs nod
answer = cs_tag
# else, the best matching percept is shifted towards the topic
else:
cs_tag = best_matching_percepts[topic_index]
agent.cs.get_concept(cs_tag).add_data(context[topic_index])
#agent.cs.get_concept(cs_tag).move_to(context[topic_index])
answer = cs_tag
agent.n_discrimination_games += 1
agent.discrimination_success = agent.n_success_dg/agent.n_discrimination_games
return answer
def load_initial_knowledge(self, dat):
""" load of initial knowledge
"""
for i in dat:
cs_tag = auks.generateRandomTag(6)
self.cs.create_pconcept(cs_tag, i[1]) # create pconcept in CS
self.cs_cs.create_node(cs_tag) # create cs node in cs_cs
self.cs_cs.add_link(cs_tag, cs_tag) # add link between cs and cs node
self.lex.add_word(i[0]) # create word in lexicon
self.lex_lex.create_node(i[0]) # create lex node in lex_lex
self.lex_lex.add_link(i[0], i[0]) # add link between lex and lex node
self.cs_lex.add_link(cs_tag, i[0]) # add link between lex and cs matrix
