def __del__( self ):
# Save best population.
self.experiment.saveBest()
# Cleanup HyperNEAT.
neat.cleanup()
# Exit evolve thread.
try:
self.evolve_thread.terminate()
self.evolve_thread.wait( 5000 )
except NameError:
print "Evolve thread not created yet, thus not destroyed."
def __del__( self ):
# Save best population.
self.experiment.saveBest()
# Cleanup HyperNEAT.
neat.cleanup()
# Exit evolve thread.
try:
self.evolve_thread.terminate()
self.evolve_thread.wait( 5000 )
except NameError:
print "Evolve thread not created yet, thus not destroyed."
def evaluate_network(self, evaluation_function):
print "Evaluating the networks..."
# Reset statistics.
self.evaluation = []
self.max_fitness = []
self.node_count = []
self.link_count = []
# Load population.
population = neat.loadFromPopulation(self.filename)
# Go through each generation, build the best network, and evaluate the completion results.
for generation in xrange(population.getGenerationCount()):
# First individual is always the most fit.
ind = population.getIndividual(0, generation)
self.max_fitness.append(ind.getFitness())
self.node_count.append(ind.getNodesCount())
self.link_count.append(ind.getLinksCount())
# Build network.
net = ind.spawnFastPhenotypeStack()
# Evaluate the network.
self.evaluation.append(evaluation_function(net))
def evaluate_network( self, evaluation_function ):
print "Evaluating the networks..."
# Reset statistics.
self.evaluation = []
self.max_fitness = []
self.node_count = []
self.link_count = []
# Load population.
population = neat.loadFromPopulation( self.filename )
# Go through each generation, build the best network, and evaluate the completion results.
for generation in xrange(population.getGenerationCount()):
# First individual is always the most fit.
ind = population.getIndividual( 0, generation )
self.max_fitness.append( ind.getFitness() )
self.node_count.append( ind.getNodesCount() )
self.link_count.append( ind.getLinksCount() )
# Build network.
net = ind.spawnFastPhenotypeStack()
# Evaluate the network.
self.evaluation.append( evaluation_function(net) )
import PyHyperNEAT as neat
import sys
file = sys.argv[1]
pop = neat.loadFromPopulation( file )
generations = pop.getGenerationCount()
individuals = pop.getIndividualCount(generations-1)
print "Loaded population from '%s'..." % file
print " - Generations: %d" % generations
print " - Ind/gen: %d" % individuals
print "Determining maximum fitness..."
speciesIDs = set()
max_fit = -1
max_fit_id = -1
max_fit_species = -1
for i in xrange(individuals):
ind = pop.getIndividual( i, generations-1 )
fitness = ind.getFitness()
speciesID = ind.getSpeciesID()
if speciesID not in speciesIDs:
print " - Found new species ID: %d" % speciesID
speciesIDs.add( speciesID )
speciesIDs.add(speciesID)
print " - Evaluating individual %d (fitness %f; species %d)..." % (i, fitness, speciesID)
if fitness > max_fit:
print " - Found new fitness leader! %f -> %f" % (max_fit, fitness)
max_fit = fitness
max_fit_id = i
def __init__( self, parent = None ):
super(Window, self).__init__( parent )
self.setWindowTitle( 'Interactive Distortion Evolver' )
### Experiment Configuration
# ATW: TODO: When we start getting reasonable distortions, there
# needs to be a method to reload an old population. For now we'll
# always make a new one.
self.experiment_data_dir = "../external/HyperNEAT/NE/HyperNEAT/out/data"
self.date_specifier = datetime.now().strftime("%y%m%d_%H%M%S")
self.image_storage = []
# Initialize HyperNEAT.
neat.initialize()
# See if there is an output directory.
if not os.path.exists(os.getcwd() + "/output"):
os.mkdir(os.getcwd() + "/output")
# Load the image experiment.
self.experiment = neat.setupExperiment(
"%s/ImageExperiment.dat" % self.experiment_data_dir,
"output/imageExp_out_%s.xml" % self.date_specifier )
# Grab the global parameters.
self.globals = neat.getGlobalParameters()
population_size = int(self.globals.getParameterValue('PopulationSize'))
### GUI Configuration
# Set default window size.
self.setFixedSize( 670, 700 )
# Initialize the list view for the distorted images.
lv = QListView( )
lv.setViewMode( QListView.IconMode )
lv.setUniformItemSizes( True )
lv.setSelectionRectVisible( True )
lv.setMovement( QListView.Static )
lv.setSelectionMode( QListView.MultiSelection )
lv.setEditTriggers( QListView.CurrentChanged )
lv.setResizeMode( QListView.Adjust )
lv.setIconSize( QSize(120, 120) )
lv.setMinimumSize( 500, 385 )
lv.setSpacing( 5 )
lv.setEnabled( False )
self.population_list = lv
# Context menu.
lv.setContextMenuPolicy( Qt.CustomContextMenu )
lv.connect( lv, SIGNAL('customContextMenuRequested (const QPoint&)'), self.handle_context_menu )
# Create the population model.
pm = PopulationModel( population_size )
self.population_model = pm
lv.setModel( pm )
for i in xrange(population_size):
self.population_model.update_item( i, DummyNetwork( i % 4 + 1 ) )
# Monitor population list selection changes.
self.connect(
lv.selectionModel(),
SIGNAL('selectionChanged(const QItemSelection &, const QItemSelection &)'),
self.handle_listview_change )
# Initialize widgets for displaying the graphic and for choosing
# a new base image.
btn_select_image = QPushButton( "Load Image" )
self.connect( btn_select_image, SIGNAL('released()'), self.select_image )
lbl_image = QLabel( "Nothing loaded" )
lbl_image.setFixedSize( 120, 120 )
self.original_image_label = lbl_image
image_layout = QVBoxLayout()
image_layout.addWidget( btn_select_image )
image_layout.addWidget( lbl_image )
# Initialize a horizontal layout for the parameters.
gb = QGroupBox( "Evolution Parameters" )
gb.setSizePolicy( QSizePolicy.Expanding, QSizePolicy.Expanding )
tw = QTableWidget( 1, 2 )
tw.setHorizontalHeaderLabels(QString("Parameter;Value").split(';'))
tw.setColumnWidth( 0, 350 )
tw.horizontalHeader().setStretchLastSection( True )
self.connect( tw, SIGNAL('itemChanged(QTableWidgetItem *)'), self.handle_parameter_change )
self.parameter_table = tw
tw_layout = QVBoxLayout()
tw_layout.addWidget( tw )
gb.setLayout( tw_layout )
param_layout = QHBoxLayout()
param_layout.addLayout( image_layout )
param_layout.addWidget( gb )
# Grab the global parameters.
parameter_count = self.globals.getParameterCount()
tw.setRowCount( parameter_count )
for p in xrange(parameter_count):
parameter_name = self.globals.getParameterName( p )
tw.model().setData( tw.model().index( p, 0 ), parameter_name )
tw.model().setData( tw.model().index( p, 1 ), self.globals.getParameterValue( parameter_name ) )
index = tw.item( p, 0 )
index.setFlags( index.flags() ^ Qt.ItemIsEditable )
# Initialize a horizontal layout for the evolve button.
btn_evolve = QPushButton( "Shuffle" )
self.connect( btn_evolve, SIGNAL('released()'), self.evolve_image )
btn_evolve.setEnabled( False )
self.btn_evolve = btn_evolve
self.btn_evolve.setSizePolicy( QSizePolicy.Expanding, QSizePolicy.Preferred )
self.spin_evolve_iterations = QSpinBox()
self.spin_evolve_iterations.setRange( 1, 100 )
self.spin_evolve_iterations.setValue( 1 )
self.cb_cross_correlate = QCheckBox( "X-Corr" )
self.cb_cross_correlate.setChecked( True )
control_layout = QHBoxLayout()
control_layout.addWidget( btn_evolve )
control_layout.addWidget( self.spin_evolve_iterations )
control_layout.addWidget( self.cb_cross_correlate )
# Initialize vertical central layout.
central_layout = QVBoxLayout()
central_layout.addLayout( param_layout )
central_layout.addWidget( lv )
central_layout.addLayout( control_layout )
central_widget = QFrame()
central_widget.setLayout( central_layout )
self.setCentralWidget( central_widget )
# Create a menu
self.image_menu = QMenu("Menu", self)
self.image_menu_save_image = self.image_menu.addAction("Save Selected Images")
self.image_menu_save_network = self.image_menu.addAction("Save Selected Neural Network")
### Initialization
# Handle command line parameters.
if len(sys.argv) >= 2:
self.select_image(sys.argv[1])
# Create progress dialog.
self.progress_dialog = QProgressDialog( self );
self.progress_dialog.setLabelText( "Evolving population..." )
self.progress_dialog.setWindowModality( Qt.WindowModal );
self.progress_dialog.setVisible( False )
self.progress_dialog.setCancelButton( None )
# Start evolver thread.
self.evolve_thread = EvolveThread( self )
self.evolve_thread.finished_job.connect( self.finish_evolution )
self.evolve_thread.update_progress.connect( self.update_evolution_progress )
self.evolve_thread.start()
# Generate first population.
self.population = self.experiment.pythonEvaluationSet()
self.evolve_image()
def __init__( self, parent = None ):
super(Window, self).__init__( parent )
self.setWindowTitle( 'Interactive Distortion Evolver' )
### Experiment Configuration
# ATW: TODO: When we start getting reasonable distortions, there
# needs to be a method to reload an old population. For now we'll
# always make a new one.
self.experiment_data_dir = "../external/HyperNEAT/NE/HyperNEAT/out/data"
self.date_specifier = datetime.now().strftime("%y%m%d_%H%M%S")
self.image_storage = []
# Initialize HyperNEAT.
neat.initialize()
# See if there is an output directory.
if not os.path.exists(os.getcwd() + "/output"):
os.mkdir(os.getcwd() + "/output")
# Load the image experiment.
self.experiment = neat.setupExperiment(
"%s/ImageExperiment.dat" % self.experiment_data_dir,
"output/imageExp_out_%s.xml" % self.date_specifier )
# Grab the global parameters.
self.globals = neat.getGlobalParameters()
population_size = int(self.globals.getParameterValue('PopulationSize'))
### GUI Configuration
# Set default window size.
self.setFixedSize( 670, 700 )
# Initialize the list view for the distorted images.
lv = QListView( )
lv.setViewMode( QListView.IconMode )
lv.setUniformItemSizes( True )
lv.setSelectionRectVisible( True )
lv.setMovement( QListView.Static )
lv.setSelectionMode( QListView.MultiSelection )
lv.setEditTriggers( QListView.CurrentChanged )
lv.setResizeMode( QListView.Adjust )
lv.setIconSize( QSize(120, 120) )
lv.setMinimumSize( 500, 385 )
lv.setSpacing( 5 )
lv.setEnabled( False )
self.population_list = lv
# Context menu.
lv.setContextMenuPolicy( Qt.CustomContextMenu )
lv.customContextMenuRequested.connect(self.handle_context_menu)
# Create the population model.
pm = PopulationModel( population_size )
self.population_model = pm
lv.setModel( pm )
for i in xrange(population_size):
self.population_model.update_item( i, DummyNetwork( i % 4 + 1 ) )
# Monitor population list selection changes.
lv.selectionModel().selectionChanged.connect(self.handle_listview_change )
# Initialize widgets for displaying the graphic and for choosing
# a new base image.
btn_select_image = QPushButton( "Load Image" )
btn_select_image.released.connect(self.select_image)
lbl_image = QLabel( "Nothing loaded" )
lbl_image.setFixedSize( 120, 120 )
self.original_image_label = lbl_image
image_layout = QVBoxLayout()
image_layout.addWidget( btn_select_image )
image_layout.addWidget( lbl_image )
# Initialize a horizontal layout for the parameters.
gb = QGroupBox( "Evolution Parameters" )
gb.setSizePolicy( QSizePolicy.Expanding, QSizePolicy.Expanding )
tw = QTableWidget( 1, 2 )
tw.setHorizontalHeaderLabels("Parameter;Value".split(';'))
tw.setColumnWidth( 0, 350 )
tw.horizontalHeader().setStretchLastSection( True )
tw.itemChanged.connect(self.handle_parameter_change)
self.parameter_table = tw
tw_layout = QVBoxLayout()
tw_layout.addWidget( tw )
gb.setLayout( tw_layout )
param_layout = QHBoxLayout()
param_layout.addLayout( image_layout )
param_layout.addWidget( gb )
# Grab the global parameters.
parameter_count = self.globals.getParameterCount()
tw.setRowCount( parameter_count )
for p in xrange(parameter_count):
parameter_name = self.globals.getParameterName( p )
tw.model().setData( tw.model().index( p, 0 ), parameter_name )
tw.model().setData( tw.model().index( p, 1 ), self.globals.getParameterValue( parameter_name ) )
index = tw.item( p, 0 )
index.setFlags( index.flags() ^ Qt.ItemIsEditable )
# Initialize a horizontal layout for the evolve button.
btn_evolve = QPushButton( "Shuffle" )
btn_evolve.released.connect(self.evolve_image)
btn_evolve.setEnabled( False )
self.btn_evolve = btn_evolve
self.btn_evolve.setSizePolicy( QSizePolicy.Expanding, QSizePolicy.Preferred )
self.spin_evolve_iterations = QSpinBox()
self.spin_evolve_iterations.setRange( 1, 100 )
self.spin_evolve_iterations.setValue( 1 )
self.cb_cross_correlate = QCheckBox( "X-Corr" )
self.cb_cross_correlate.setChecked( True )
control_layout = QHBoxLayout()
control_layout.addWidget( btn_evolve )
control_layout.addWidget( self.spin_evolve_iterations )
control_layout.addWidget( self.cb_cross_correlate )
# Initialize vertical central layout.
central_layout = QVBoxLayout()
central_layout.addLayout( param_layout )
central_layout.addWidget( lv )
central_layout.addLayout( control_layout )
central_widget = QFrame()
central_widget.setLayout( central_layout )
self.setCentralWidget( central_widget )
# Create a menu
self.image_menu = QMenu("Menu", self)
self.image_menu_save_image = self.image_menu.addAction("Save Selected Images")
self.image_menu_save_network = self.image_menu.addAction("Save Selected Neural Network")
### Initialization
# Handle command line parameters.
if len(sys.argv) >= 2:
self.select_image(sys.argv[1])
# Create progress dialog.
self.progress_dialog = QProgressDialog( self );
self.progress_dialog.setLabelText( "Evolving population..." )
self.progress_dialog.setWindowModality( Qt.WindowModal );
self.progress_dialog.setVisible( False )
self.progress_dialog.setCancelButton( None )
# Start evolver thread.
self.evolve_thread = EvolveThread( self )
self.evolve_thread.finished_job.connect( self.finish_evolution )
self.evolve_thread.update_progress.connect( self.update_evolution_progress )
self.evolve_thread.start()
# Generate first population.
self.population = self.experiment.pythonEvaluationSet()
self.evolve_image()
print "Proper usage: ./run distort.py <network> <input_image> <output_image>"
exit(0)
# Imports.
import PyHyperNEAT as neat
from PyQt5.QtGui import QApplication, QImage, QPixmap
import gui.PopulationModel
print "-------------------------------------------------------------------"
print "DISTORTING IMAGE"
# Load the file and generate the network.
print "-------------------------------------------------------------------"
print "Loading network..."
neat.initialize()
population = neat.load(argv[1])
individual = population.getIndividual(0, population.getGenerationCount() - 1)
network = individual.spawnFastPhenotypeStack()
# Open the input image.
print "-------------------------------------------------------------------"
print "Distorting image..."
distortion = gui.PopulationModel.PopulationItem()
application = QApplication(argv)
image_map = QPixmap(argv[2])
distorted_image_map = distortion.distort(image_map, network)
distorted_image = QImage(distorted_image_map)
distorted_image.save(argv[3])
# Specified experiment in command line arguments?
if len(sys.argv) >= 2:
experiment_name = sys.argv[1]
if len(sys.argv) >= 3:
experiment_number = int(sys.argv[2])
# Experiment prefix.
prefix = datetime.now().strftime("%y%m%d_%H%M%S")
# Move to experiment directory.
print "Moving to experiment run directory '%s'..." % experiment_run_dir
os.chdir( experiment_run_dir )
# Initialize HyperNEAT.
print "Initializing HyperNEAT..."
neat.initializeHyperNEAT()
# Load experiment.
print "Running HyperNEAT experiment '%s'..." % experiment_name
neat.setupExperiment( "%s/%sExperiment.dat" % (experiment_data_dir, experiment_name), "%s.xml" % output_name )
# Cleanup.
print "Cleaning up..."
neat.cleanupHyperNEAT()
# Move files over.
os.chdir( start_dir )
call( "mv %s output/%s_%02d_%s.xml.gz" % ("%s/%s_best.xml.gz" % (experiment_run_dir, output_name), experiment_name, experiment_number, prefix), shell=True )
# Done.
print "===== Done."
print "Proper usage: ./run distort.py <network> <input_image> <output_image>"
exit( 0 )
# Imports.
import PyHyperNEAT as neat
from PyQt5.QtGui import QApplication, QImage, QPixmap
import gui.PopulationModel
print "-------------------------------------------------------------------"
print "DISTORTING IMAGE"
# Load the file and generate the network.
print "-------------------------------------------------------------------"
print "Loading network..."
neat.initialize()
population = neat.load( argv[1] )
individual = population.getIndividual( 0, population.getGenerationCount() - 1 )
network = individual.spawnFastPhenotypeStack()
# Open the input image.
print "-------------------------------------------------------------------"
print "Distorting image..."
distortion = gui.PopulationModel.PopulationItem()
application = QApplication( argv )
image_map = QPixmap( argv[2] )
distorted_image_map = distortion.distort( image_map, network )
distorted_image = QImage( distorted_image_map )
distorted_image.save( argv[3] )
experiment_name = sys.argv[1]
if len(sys.argv) >= 3:
experiment_number = int(sys.argv[2])
print "%s - %d" % (experiment_name, experiment_number)
# Experiment prefix.
prefix = datetime.now().strftime("%y%m%d_%H%M%S")
# Move to experiment directory.
print "Moving to experiment run directory '%s'..." % experiment_run_dir
os.chdir(experiment_run_dir)
# Initialize HyperNEAT.
print "Initializing HyperNEAT..."
neat.initializeHyperNEAT()
# Load experiment.
print "Running HyperNEAT experiment '%s'..." % experiment_name
experimentRun = neat.setupExperiment(
"%s/%sExperiment.dat" % (experiment_data_dir, experiment_name),
"%s.xml" % output_name)
maxgen = neat.getMaximumGenerations()
for x in range(maxgen):
if (x > 0):
experimentRun.produceNextGeneration()
#population pre-process, in experiment running
experiment_name = sys.argv[1]
if len(sys.argv) >= 3:
experiment_number = int(sys.argv[2])
print "%s - %d" % (experiment_name, experiment_number)
# Experiment prefix.
prefix = datetime.now().strftime("%y%m%d_%H%M%S")
# Move to experiment directory.
print "Moving to experiment run directory '%s'..." % experiment_run_dir
os.chdir( experiment_run_dir )
# Initialize HyperNEAT.
print "Initializing HyperNEAT..."
neat.initializeHyperNEAT()
# Load experiment.
print "Running HyperNEAT experiment '%s'..." % experiment_name
experimentRun = neat.setupExperiment( "%s/%sExperiment.dat" % (experiment_data_dir, experiment_name), "%s.xml" % output_name )
maxgen = neat.getMaximumGenerations()
for x in range(maxgen):
if(x > 0):
experimentRun.produceNextGeneration()
#population pre-process, in experiment running
experimentRun.preprocessPopulation()
# Specified experiment in command line arguments?
if len(sys.argv) >= 2:
experiment_name = sys.argv[1]
if len(sys.argv) >= 3:
experiment_number = int(sys.argv[2])
# Experiment prefix.
prefix = datetime.now().strftime("%y%m%d_%H%M%S")
# Move to experiment directory.
print "Moving to experiment run directory '%s'..." % experiment_run_dir
os.chdir(experiment_run_dir)
# Initialize HyperNEAT.
print "Initializing HyperNEAT..."
neat.initializeHyperNEAT()
# Load experiment.
print "Running HyperNEAT experiment '%s'..." % experiment_name
neat.setupExperiment(
"%s/%sExperiment.dat" % (experiment_data_dir, experiment_name),
"%s.xml" % output_name)
# Cleanup.
print "Cleaning up..."
neat.cleanupHyperNEAT()
# Move files over.
os.chdir(start_dir)
call("mv %s output/%s_%02d_%s.xml.gz" %
("%s/%s_best.xml.gz" % (experiment_run_dir, output_name), experiment_name,
