def evaluate_pic(genome, debug=False, save=None):
verbose = True
net = NEAT.NeuralNetwork()
genome.BuildPhenotype(net)
if verbose:
print 'dcalc'
genome.CalculateDepth()
if verbose:
print 'dcalc complete'
depth = genome.GetDepth()
error = 0
# do stuff and return the fitness
tot_vox = sz_x * sz_y * sz_z
voxels = numpy.zeros((tot_vox, 4))
print "calling batch...", genome.NumNeurons()
voxels = net.Batch_input(coordinates, depth)
print "complete"
voxels = voxels.reshape((sz_x, sz_y, 4))
voxels = voxels[:, :, :3]
imgs = [voxels]
if verbose:
print 'running image rec'
results, behavior = run_image(imgs)
print "behavior shape: ", behavior.shape
#print results.shape
#print niche_matrix.T.shape
results = np.dot(results, niche_matrix.T)
full_matrix = results.copy()
if debug:
return imgs, results
results = results[0, :] #results.prod(axis=0)
print results.shape
#return float(results[target_class]),results,full_matrix
return behavior, results, full_matrix
def evaluate_pic(genome,debug=False,save=None):
verbose=True
net = NEAT.NeuralNetwork()
genome.BuildPhenotype(net)
if verbose:
print 'dcalc'
genome.CalculateDepth()
if verbose:
print 'dcalc complete'
depth = genome.GetDepth()
error = 0
# do stuff and return the fitness
tot_vox = sz_x*sz_y*sz_z
voxels = numpy.zeros((tot_vox,4))
print "calling batch...", genome.NumNeurons()
voxels = net.Batch_input(coordinates,depth)
print "complete"
voxels = voxels.reshape((sz_x,sz_y,4))
voxels = voxels[:,:,:3]
imgs = [voxels]
if verbose:
print 'running image rec'
results,behavior = run_image(imgs)
print "behavior shape: ", behavior.shape
#print results.shape
#print niche_matrix.T.shape
results=np.dot(results,niche_matrix.T)
full_matrix = results.copy()
if debug:
return imgs,results
results = results[0,:] #results.prod(axis=0)
print results.shape
#return float(results[target_class]),results,full_matrix
return behavior,results,full_matrix
def evaluate(genome, debug=False, save=None, dnn=True):
global lighting
verbose = True
if verbose:
print 'building...'
net = NEAT.NeuralNetwork()
genome.BuildPhenotype(net)
if verbose:
print 'dcalc'
genome.CalculateDepth()
if verbose:
print 'dcalc complete'
depth = genome.GetDepth()
#fixed depth for now...
#depth=6
error = 0
# do stuff and return the fitness
tot_vox = sz_x * sz_y * sz_z
voxels = numpy.zeros((tot_vox, 4))
print "calling batch...", genome.NumNeurons()
voxels = net.Batch_input(coordinates, depth)
print "complete"
voxels = voxels.reshape((sz_x, sz_y, sz_z, 4))
percentile = np.clip(net.neurons[5].bias, 0, 1)
thresh = 0.5
#thresh=np.percentile(voxels[:,:,:,0],100.0*percentile)
voxels[0, :, :, 0] = thresh - 0.01
voxels[-1, :, :, 0] = thresh - 0.01
voxels[:, 0, :, 0] = thresh - 0.01
voxels[:, -1, :, 0] = thresh - 0.01
voxels[:, :, 0, 0] = thresh - 0.01
voxels[:, :, -1, 0] = thresh - 0.01
bg_color = [net.neurons[k].time_const for k in range(3)]
if fixed_bg:
#bg_color=[0.6,0.6,0.6]
bg_color = [0.4078, 0.4575, 0.4811]
oparam = np.clip([net.neurons[k].bias for k in range(4)], 0, 1)
print bg_color
print oparam
if verbose:
print 'rendering images'
theta = 45
jitter = 5
shiny = oparam[0] * 128
spec = oparam[1]
amb = oparam[2]
diff = oparam[3]
img1 = render(voxels,
bg_color,
0,
0,
save=save,
shiny=shiny,
spec=spec,
amb=amb,
lighting=lighting,
diff=diff)
img2 = render(voxels,
bg_color,
theta,
jitter,
shiny=shiny,
spec=spec,
amb=amb,
lighting=lighting,
diff=diff)
img3 = render(voxels,
bg_color,
theta * 2,
0,
shiny=shiny,
spec=spec,
amb=amb,
lighting=lighting,
diff=diff)
img4 = render(voxels,
bg_color,
theta * 3,
jitter,
shiny=shiny,
spec=spec,
amb=amb,
lighting=lighting,
diff=diff)
img5 = render(voxels,
bg_color,
theta * 4,
0,
shiny=shiny,
spec=spec,
amb=amb,
lighting=lighting,
diff=diff)
img6 = render(voxels,
bg_color,
theta * 5,
jitter,
spec=spec,
amb=amb,
lighting=lighting,
diff=diff)
imgs = [img1, img2, img3, img4, img5, img6]
if verbose:
print 'running image rec'
results, beh = None, None
full_matrix = None
if dnn:
print "dnn"
results, beh = run_image(imgs)
#print niche_matrix.T.shape
results = np.dot(results, niche_matrix.T)
full_matrix = results.copy()
else:
print "no dnn"
if debug:
return imgs, results
results = results.prod(axis=0)
print results.shape
return beh, results, full_matrix
def evaluate(genome,debug=False,save=None,dnn=True):
global lighting
verbose=True
if verbose:
print 'building...'
net = NEAT.NeuralNetwork()
genome.BuildPhenotype(net)
if verbose:
print 'dcalc'
genome.CalculateDepth()
if verbose:
print 'dcalc complete'
depth = genome.GetDepth()
#fixed depth for now...
#depth=6
error = 0
# do stuff and return the fitness
tot_vox = sz_x*sz_y*sz_z
voxels = numpy.zeros((tot_vox,4))
print "calling batch...", genome.NumNeurons()
voxels = net.Batch_input(coordinates,depth)
print "complete"
voxels = voxels.reshape((sz_x,sz_y,sz_z,4))
percentile = np.clip(net.neurons[5].bias,0,1)
thresh=0.5
#thresh=np.percentile(voxels[:,:,:,0],100.0*percentile)
voxels[0,:,:,0]=thresh-0.01
voxels[-1,:,:,0]=thresh-0.01
voxels[:,0,:,0]=thresh-0.01
voxels[:,-1,:,0]=thresh-0.01
voxels[:,:,0,0]=thresh-0.01
voxels[:,:,-1,0]=thresh-0.01
bg_color = [net.neurons[k].time_const for k in range(3)]
if fixed_bg:
#bg_color=[0.6,0.6,0.6]
bg_color=[0.4078,0.4575,0.4811]
oparam = np.clip([net.neurons[k].bias for k in range(4)],0,1)
print bg_color
print oparam
if verbose:
print 'rendering images'
theta=45
jitter=5
shiny = oparam[0]*128
spec = oparam[1]
amb = oparam[2]
diff = oparam[3]
img1 = render(voxels,bg_color,0,0,save=save,shiny=shiny,spec=spec,amb=amb,lighting=lighting,diff=diff)
img2 = render(voxels,bg_color,theta,jitter,shiny=shiny,spec=spec,amb=amb,lighting=lighting,diff=diff)
img3 = render(voxels,bg_color,theta*2,0,shiny=shiny,spec=spec,amb=amb,lighting=lighting,diff=diff)
img4 = render(voxels,bg_color,theta*3,jitter,shiny=shiny,spec=spec,amb=amb,lighting=lighting,diff=diff)
img5 = render(voxels,bg_color,theta*4,0,shiny=shiny,spec=spec,amb=amb,lighting=lighting,diff=diff)
img6 = render(voxels,bg_color,theta*5,jitter,spec=spec,amb=amb,lighting=lighting,diff=diff)
imgs = [img1,img2,img3,img4,img5,img6]
if verbose:
print 'running image rec'
results,beh = None,None
full_matrix=None
if dnn:
print "dnn"
results,beh = run_image(imgs)
#print niche_matrix.T.shape
results=np.dot(results,niche_matrix.T)
full_matrix = results.copy()
else:
print "no dnn"
if debug:
return imgs,results
results = results.prod(axis=0)
print results.shape
return beh,results,full_matrix