def modeler_l2_errors(modeler, sample, sample_divisions=2):
sample_list = divide_sample(sample, sample_divisions)
l2_errors = []
sens_dict = defaultdict(list)
for i in range(len(sample_list)):
training_sample = sample_list[i]
training_in, training_out = sample_to_io(training_sample)
model = modeler(training_in, training_out)
sens_dict_i = model.sensitivity_dict(training_out)
sorted_sens = sorted(sens_dict_i.iteritems(), key=lambda x: x[1]['t'])
total = 0
for syms, d in sorted_sens:
sens_dict[syms].append(d['s'])
# print "Component functions"
# for s, p in model.component_functions(coeff_gate=1.0e-08).items():
# print " ", s, p
for j in range(len(sample_list)):
if i != j:
testing_sample = sample_list[j]
test_in, test_out = sample_to_io(testing_sample)
test_approx = array_map(model, test_in)
l2_errors.append(relative_l2_error(test_out, test_approx))
keys, vals = zip(
*sorted(sens_dict.items(), key=lambda t: sum(t[1]), reverse=True))
bp.plot_boxplots(vals[:15], [str(list(k)) for k in keys][:15])
plt.show()
return l2_errors
def __init__(self, p, space):
self.syms = space.syms
self.p = p
def f(x):
return float(self.p.eval(tuple(x)))
self.sample_values = array_map(f, space.sample)
self.sq_norm = sum(self.sample_values**2)
def __init__(self, p, space):
self.syms = space.syms
self.p = p
def f(x):
return float(self.p.eval(tuple(x)))
self.sample_values = array_map(f, space.sample)
self.sq_norm = sum(self.sample_values**2)
def run_training_test(sample, expander, training_proportion=0.5):
training_data, test_data = divide_sample(sample, training_proportion)
training_in, training_out = sample_to_io(training_data)
test_in, test_out = sample_to_io(test_data)
trained_expansion = expander(training_in, training_out)
test_approx = array_map(trained_expansion, test_in)
training_approx = array_map(trained_expansion, training_in)
print "Training L2 error:", \
relative_l2_error(training_out, training_approx)
print "Training L1 error:", \
average_relative_l1_error(training_out, training_approx)
print "Test L1 error:", \
average_relative_l1_error(test_out, test_approx)
return relative_l2_error(test_out, test_approx)
def run_training_test(sample, expander, training_proportion=0.5):
training_data, test_data = divide_sample(sample, training_proportion)
training_in, training_out = sample_to_io(training_data)
test_in, test_out = sample_to_io(test_data)
trained_expansion = expander(training_in, training_out)
test_approx = array_map(trained_expansion, test_in)
training_approx = array_map(trained_expansion, training_in)
print "Training L2 error:", \
relative_l2_error(training_out, training_approx)
print "Training L1 error:", \
average_relative_l1_error(training_out, training_approx)
print "Test L1 error:", \
average_relative_l1_error(test_out, test_approx)
return relative_l2_error(test_out, test_approx)
def modeler_l2_errors(modeler, sample, sample_divisions=2):
sample_list = divide_sample(sample, sample_divisions)
l2_errors = []
sens_dict = defaultdict(list)
for i in range(len(sample_list)):
training_sample = sample_list[i]
training_in, training_out = sample_to_io(training_sample)
model = modeler(training_in, training_out)
sens_dict_i = model.sensitivity_dict(training_out)
sorted_sens = sorted(
sens_dict_i.iteritems(),
key=lambda x: x[1]['t']
)
total = 0
for syms, d in sorted_sens:
sens_dict[syms].append(d['s'])
# print "Component functions"
# for s, p in model.component_functions(coeff_gate=1.0e-08).items():
# print " ", s, p
for j in range(len(sample_list)):
if i != j:
testing_sample = sample_list[j]
test_in, test_out = sample_to_io(testing_sample)
test_approx = array_map(model, test_in)
l2_errors.append(
relative_l2_error(test_out, test_approx)
)
keys, vals = zip(*sorted(sens_dict.items(), key=lambda t: sum(t[1]), reverse=True))
bp.plot_boxplots(vals[:15], [str(list(k)) for k in keys][:15])
plt.show()
return l2_errors
