def network_atoms_to_graphviz(network_atoms):
"""Create a pygraphviz.AGraph object from the given network_atoms."""
atoms = (atoms_split(a) for a in network_atoms.split('.') if a != '')
graph = defaultdict(str) # ID: type
edges = [] # list of 2-tuple
for name, args in atoms:
if name == 'output' and len(args) == 1: # output/1
# If already in graph, then idn is an output
# node already set as neuron.
graph[args] += '\n(output)'
elif name == 'neuron' and len(args) == 2: # neuron/2
idn, ntype = args
graph[idn] = NeuronType(ntype).name + graph.get(idn, '')
elif name == 'edge' and len(args) == 2: # edge/2
ida, idb = args
edges.append((ida, idb))
else:
# some received atoms are not handled…
# see logs for details
LOGGER.error('converter.network_atoms_to_graphviz:'
' UNHANDLED ATOMS: ' + str((name, args)) + '.')
# Create and return the dot from graph and edges
dot = pgv.AGraph(strict=False, directed=True)
[dot.add_node(idn, label=str(idn) + '\n' + label) for idn, label in graph.items()]
[dot.add_edge(a, b) for a, b in edges]
return dot
def __init__(self, config):
super().__init__(config=config, config_fields=[
'memory_min_size', 'memory_max_size',
'neuron_inter_mincount', 'neuron_inter_maxcount',
'neuron_edges_mincount', 'neuron_edges_maxcount',
'mutator',
])
self.neuron_types = NeuronType.xano()
def __init__(self, config):
super().__init__(config=config,
config_fields=[
'memory_min_size',
'memory_max_size',
'neuron_inter_mincount',
'neuron_inter_maxcount',
'neuron_edges_mincount',
'neuron_edges_maxcount',
'mutator',
])
self.neuron_types = NeuronType.xano()
def mutate(self, nb_intermediate_neuron:int, nb_total_neuron:int,
neuron_types:iter, edges:iter):
"""Return the data received in input, modified according to
mutation settings.
nb_intermediate_neuron: integer equal to number of intermediate neuron.
nb_total_neuron: integer equal to total number of neuron.
neuron_types: iterable of NeuronType.
edges: iterable of 2-tuple describing links between neurons.
"""
MUTATION_RATE = self.mutation_rate
mutation = lambda: random() <= MUTATION_RATE
mutate_nb_neuron = mutation(), mutation()
mutate_neuron_type = mutation(), mutation()
mutate_edges = mutation(), mutation()
del mutation
if any(mutate_nb_neuron):
add, rmv = mutate_nb_neuron
neuron_types = list(neuron_types) # allow modifications
if add:
nb_total_neuron += 1
nb_intermediate_neuron += 1
neuron_types.append(random_choice(NeuronType.xano()))
if commons.log_level() >= logging.INFO:
LOGGER.info('Mutator ' + str(self)
+ ' add new neuron of type '
+ neuron_types[-1].name + '.')
if rmv and nb_intermediate_neuron > 0:
nb_total_neuron -= 1
nb_intermediate_neuron -= 1
target_idx = randrange(0, len(neuron_types))
if commons.log_level() >= logging.INFO:
LOGGER.info('Mutator ' + str(self)
+ ' remove neuron ' + str(target_idx) + ' ('
+ neuron_types[target_idx].name + ').')
del neuron_types[target_idx]
if any(mutate_neuron_type):
modify, swap = mutate_neuron_type
neuron_types = list(neuron_types) # allow modifications
if modify: # modify just one type
target_idx = randrange(0, len(neuron_types))
new_type = random_choice(NeuronType.xano())
if commons.log_level() >= logging.INFO:
LOGGER.info('Mutator ' + str(self) + ' modify type of '
+ str(target_idx) + ' from '
+ neuron_types[target_idx].name + ' to '
+ new_type.name + '.'
)
neuron_types[target_idx] = new_type
if swap: # swap two types in the list
target1_idx = randrange(0, len(neuron_types))
target2_idx = randrange(0, len(neuron_types))
neuron_types[target1_idx], neuron_types[target2_idx] = (
neuron_types[target2_idx], neuron_types[target1_idx]
)
if commons.log_level() >= logging.INFO:
LOGGER.info('Mutator ' + str(self) + ' swaps '
+ str(target1_idx) + ' ('
+ neuron_types[target2_idx].name + ') and '
+ str(target1_idx) + ' ('
+ neuron_types[target2_idx].name + ').'
)
if any(mutate_edges):
add, rmv = mutate_edges
edges = list(edges) # allow modifications
if add: # add one new edge
target1_idx = randrange(1, nb_total_neuron + 1)
target2_idx = randrange(1, nb_total_neuron + 1)
edges.append((target1_idx, target2_idx))
if commons.log_level() >= logging.INFO:
LOGGER.info('Mutator ' + str(self) + ' get an edge '
+ str((target1_idx, target2_idx)) + '.')
if rmv: # remove one existing edge
idx = randrange(0, len(edges))
if commons.log_level() >= logging.INFO:
LOGGER.info('Mutator ' + str(self) + ' lose its edge '
+ str(edges[idx]) + '.')
neuron_types = tuple(neuron_types)
edges = tuple(edges)
return nb_intermediate_neuron, neuron_types, edges