def plotEvolutionaryOptimizationParallel(directory,axs,plot_all_rollouts=False):
try:
n_parallel = np.loadtxt(directory+"/n_parallel.txt")
except IOError:
n_parallel = 1
#################################
# Load and plot learning curve
(update_at_samples, costs_all, eval_at_samples, costs_eval) = loadLearningCurve(directory)
ax = (None if axs==None else axs[1])
plotLearningCurve(eval_at_samples,costs_eval,ax,costs_all)
#y_limits = [0,1.2*max(learning_curve)];
plotUpdateLines(update_at_samples,ax)
#################################
# Load and plot exploration curve
dim_dir = "";
ax = (None if axs==None else axs[0])
for i_parallel in range(n_parallel):
(covar_at_samples, sqrt_max_eigvals) = loadExplorationCurve(directory,i_parallel)
plotExplorationCurve(covar_at_samples,sqrt_max_eigvals,ax)
plotUpdateLines(update_at_samples,ax)
n_updates = loadNumberOfUpdates(directory)
ax = (None if axs==None else axs[2])
if (ax!=None):
#################################
# Visualize the update in parameter space
for update in range(n_updates):
cur_directory = '%s/update%05d' % (directory, update+1)
plotUpdateSummaryParallelFromDirectory(cur_directory,ax,False)
ax.set_title('Search space')
if ( (axs!=None) and (len(axs)>3) ):
ax = axs[3];
rollouts_python_script = directory+'/plotRollouts.py'
print rollouts_python_script
if (os.path.isfile(rollouts_python_script)):
lib_path = os.path.abspath(directory)
sys.path.append(lib_path)
from plotRollouts import plotRollouts
for update in range(n_updates):
cur_directory = '%s/update%05d' % (directory, update+1)
filename = "/cost_vars_eval.txt"
if plot_all_rollouts:
filename = "/cost_vars.txt"
cost_vars = np.loadtxt(cur_directory+filename)
rollout_lines = plotRollouts(cost_vars,ax)
color_val = (1.0*update/n_updates)
#cur_color = [1.0-0.9*color_val,0.1+0.9*color_val,0.1]
cur_color = [0.0-0.0*color_val, 0.0+1.0*color_val, 0.0-0.0*color_val]
#print str(update)+" "+str(n_updates)+" "+str(cur_color)
plt.setp(rollout_lines,color=cur_color)
if (update==0):
plt.setp(rollout_lines,color='r',linewidth=2)
def plotEvolutionaryOptimization(n_updates,directory,axs=None,plot_all_rollouts=False):
if (n_updates==0):
n_updates += 1;
#################################
# Read data relevant to learning and exploration curve
costs = []
cost_evals = []
n_samples_per_update = [0]
covars_per_update = [];
for update in range(n_updates):
cur_directory = '%s/update%05d' % (directory, update+1)
# Load evaluation cost
try:
cur_cost_eval = np.loadtxt(cur_directory+"/cost_eval.txt")
cost_evals.append(np.atleast_1d(cur_cost_eval)[0])
except IOError:
cur_cost_eval = []
# Load costs
cur_costs = np.loadtxt(cur_directory+"/costs.txt")
costs.extend(cur_costs)
n_samples_per_update.append(len(costs))
# Load covar matrix
covar = np.loadtxt(cur_directory+"/distribution_covar.txt")
covars_per_update.append(covar)
# Load final covar matrix
covar = np.loadtxt(cur_directory+"/distribution_new_covar.txt")
covars_per_update.append(covar)
#################################
# Plot the exploration magnitude
ax = (None if axs==None else axs[0])
max_eigval_per_update = plotExplorationCurve(n_samples_per_update,covars_per_update,ax)
#################################
# Plot the learning curve
ax = (None if axs==None else axs[1])
update_centers, mean_costs = plotLearningCurve(n_samples_per_update,costs,cost_evals,ax)
ax = (None if axs==None else axs[2])
if (ax!=None):
#################################
# Visualize the update in parameter space
for update in range(n_updates):
cur_directory = '%s/update%05d' % (directory, update+1)
plotUpdateSummaryFromDirectory(cur_directory,ax,False)
cur_directory = '%s/update%05d' % (directory, n_updates)
plotUpdateSummaryFromDirectory(cur_directory,ax,True)
ax.set_title('Search space')
if ( (axs!=None) and (len(axs)>3) ):
ax = axs[3];
for update in range(n_updates):
cur_directory = '%s/update%05d' % (directory, update+1)
filename = "/cost_vars_eval.txt"
if plot_all_rollouts:
filename = "/cost_vars.txt"
cost_vars = np.loadtxt(cur_directory+filename)
rollout_lines = plotRollouts(cost_vars,ax)
color_val = (1.0*update/n_updates)
#cur_color = [1.0-0.9*color_val,0.1+0.9*color_val,0.1]
cur_color = [0.0-0.0*color_val, 0.0+1.0*color_val, 0.0-0.0*color_val]
#print str(update)+" "+str(n_updates)+" "+str(cur_color)
plt.setp(rollout_lines,color=cur_color)
if (update==0):
plt.setp(rollout_lines,color='r',linewidth=2)
return (update_centers, mean_costs, n_samples_per_update, max_eigval_per_update)
def plotEvolutionaryOptimization(n_updates,
directory,
axs=None,
plot_all_rollouts=False):
if (n_updates == 0):
n_updates += 1
#################################
# Read data relevant to learning and exploration curve
costs = []
cost_evals = []
n_samples_per_update = [0]
covars_per_update = []
for update in range(n_updates):
cur_directory = '%s/update%05d' % (directory, update + 1)
# Load evaluation cost
try:
cur_cost_eval = np.loadtxt(cur_directory + "/cost_eval.txt")
cost_evals.append(np.atleast_1d(cur_cost_eval)[0])
except IOError:
cur_cost_eval = []
# Load costs
cur_costs = np.loadtxt(cur_directory + "/costs.txt")
costs.extend(cur_costs)
n_samples_per_update.append(len(costs))
# Load covar matrix
covar = np.loadtxt(cur_directory + "/distribution_covar.txt")
covars_per_update.append(covar)
# Load final covar matrix
covar = np.loadtxt(cur_directory + "/distribution_new_covar.txt")
covars_per_update.append(covar)
#################################
# Plot the exploration magnitude
ax = (None if axs == None else axs[0])
max_eigval_per_update = plotExplorationCurve(n_samples_per_update,
covars_per_update, ax)
#################################
# Plot the learning curve
ax = (None if axs == None else axs[1])
update_centers, mean_costs = plotLearningCurve(n_samples_per_update, costs,
cost_evals, ax)
ax = (None if axs == None else axs[2])
if (ax != None):
#################################
# Visualize the update in parameter space
for update in range(n_updates):
cur_directory = '%s/update%05d' % (directory, update + 1)
plotUpdateSummaryFromDirectory(cur_directory, ax, False)
cur_directory = '%s/update%05d' % (directory, n_updates)
plotUpdateSummaryFromDirectory(cur_directory, ax, True)
ax.set_title('Search space')
if ((axs != None) and (len(axs) > 3)):
ax = axs[3]
for update in range(n_updates):
cur_directory = '%s/update%05d' % (directory, update + 1)
filename = "/cost_vars_eval.txt"
if plot_all_rollouts:
filename = "/cost_vars.txt"
cost_vars = np.loadtxt(cur_directory + filename)
rollout_lines = plotRollouts(cost_vars, ax)
color_val = (1.0 * update / n_updates)
#cur_color = [1.0-0.9*color_val,0.1+0.9*color_val,0.1]
cur_color = [
0.0 - 0.0 * color_val, 0.0 + 1.0 * color_val,
0.0 - 0.0 * color_val
]
#print str(update)+" "+str(n_updates)+" "+str(cur_color)
plt.setp(rollout_lines, color=cur_color)
if (update == 0):
plt.setp(rollout_lines, color='r', linewidth=2)
return (update_centers, mean_costs, n_samples_per_update,
max_eigval_per_update)
