def passive_label_query(number_points, scaler):
params = ap.set_params()
dim = params["dim"]
time_bound = params["time_bound"]
# Pool of number_points input points, randomly selected
x_samples = np.zeros((number_points, dim))
for i in range(number_points):
x_samples[i, :] = ap.rand_state()
x_samples_scaled = scaler.transform(x_samples)
start = time.time()
# CALLING MATLAB FUNCTION
eng = matlab.engine.start_matlab()
x_add, y_add = eng.gen_labels(matlab.double((x_samples.T).tolist()),
matlab.double([time_bound]),
nargout=2)
eng.quit()
#x_add = np.array(x_add).T
y_add = np.array(y_add[0])
print("PASSIVE STAGING: uncertain points labeled in {} seconds".format(
time.time() - start))
return x_samples_scaled, y_add
def active_label_query(sess,
type_rej_rule,
number_points,
scaler,
bnn,
rej_rule,
n_emp_samples,
decision_threshold=None):
params = ap.set_params()
dim = params["dim"]
time_bound = params["time_bound"]
# Pool of number_points input points, randomly selected
x_samples = np.zeros((number_points, dim))
for i in range(number_points):
x_samples[i, :] = ap.rand_state()
x_samples_scaled = scaler.transform(x_samples)
samples_avg_probs, samples_std_probs, samples_avg_pred_class = compute_avg_std_pred_probs(
sess, x_samples_scaled, n_emp_samples, bnn)
samples_unc = np.vstack((samples_avg_probs, samples_std_probs))
if type_rej_rule == "SVC":
acc_rej_samples = apply_svc_rejection_rule(rej_rule, samples_unc)
elif type_rej_rule == "GPC":
acc_rej_samples = apply_gpc_rejection_rule(rej_rule, samples_unc,
decision_threshold)
else:
print("Unknown type of rejection rule!")
#x_rej = x_samples[(1-acc_rej_samples).astype(bool)]
#x_rej_scaled = x_samples_scaled[(1-acc_rej_samples).astype(bool)]
x_rej = x_samples[(acc_rej_samples < 0)]
x_rej_scaled = x_samples_scaled[(acc_rej_samples < 0)]
print('Number of points to add: {}\n'.format(x_rej_scaled.shape[0]))
start = time.time()
#rej_samples_avg_pred_class = samples_avg_pred_class[(1-acc_rej_samples).astype(bool)]
#rej_samples_avg_probs = samples_avg_probs[(1-acc_rej_samples).astype(bool)]
#rej_samples_std_probs = samples_std_probs[(1-acc_rej_samples).astype(bool)]
rej_samples_avg_pred_class = samples_avg_pred_class[(acc_rej_samples < 0)]
rej_samples_avg_probs = samples_avg_probs[(acc_rej_samples < 0)]
rej_samples_std_probs = samples_std_probs[(acc_rej_samples < 0)]
rej_samples_unc = np.vstack((rej_samples_avg_probs, rej_samples_std_probs))
# CALLING MATLAB FUNCTION
eng = matlab.engine.start_matlab()
x_add, y_add = eng.gen_labels(matlab.double((x_rej.T).tolist()),
matlab.double([time_bound]),
nargout=2)
eng.quit()
#x_add = np.array(x_add).T
y_add = np.array(y_add[0])
print(
"STAGING: uncertain points labeled in {} seconds".format(time.time() -
start))
return x_rej_scaled, y_add, rej_samples_unc, rej_samples_avg_pred_class
def active_label_query(number_points,
scaler,
local_guide,
rej_rule,
n_emp_samples,
decision_threshold=None):
start = time.time()
params = ap.set_params()
dim = params["dim"]
time_bound = params["time_bound"]
# Pool of number_points input points, randomly selected
x_samples = np.zeros((number_points, dim))
for i in range(number_points):
x_samples[i, :] = ap.rand_state()
x_samples_scaled = scaler.transform(x_samples)
x_samples_scaled_t = torch.FloatTensor(x_samples_scaled)
samples_avg_pred_classes, samples_avg_probs, samples_std_probs = predictions_on_set(
x_samples_scaled_t, number_points, n_emp_samples, local_guide)
samples_unc = np.vstack((samples_avg_probs, samples_std_probs))
acc_rej_samples = apply_gpc_rejection_rule(rej_rule, samples_unc,
decision_threshold)
x_rej = x_samples[(acc_rej_samples < 0)]
x_rej_scaled = x_samples_scaled[(acc_rej_samples < 0)]
eng = matlab.engine.start_matlab()
x_add, y_add = eng.gen_labels(matlab.double((x_rej.T).tolist()),
matlab.double([time_bound]),
nargout=2)
eng.quit()
#x_add = np.array(x_add).T
y_add = np.array(y_add[0])
print(
"STAGING: uncertain points labeled in {} seconds".format(time.time() -
start))
return x_rej_scaled, y_add
def active_label_query(number_points, scaler, alphas, nsc, rej_rule, phase):
params = ap.set_params()
dim = params["dim"]
time_bound = params["time_bound"]
# Pool of number_points input points, randomly selected
x_samples = np.zeros((number_points, dim))
for i in range(number_points):
x_samples[i, :] = ap.rand_state()
x_samples_scaled = scaler.transform(x_samples)
# NSC prob prediction on these samples
samples_unc, samples_pred_probs = compute_test_uncertainties(
nsc, alphas, x_samples_scaled)
acc_rej_samples = apply_svc_rejection_rule(rej_rule, samples_unc)
x_rej = x_samples[(
1 - acc_rej_samples).astype(bool)] #.reshape((number_points,))
x_rej_scaled = x_samples_scaled[(1 - acc_rej_samples).astype(bool)]
print('Number of points to add: {}\n'.format(x_rej.shape[0]))
start = time.time()
# CALLING MATLAB FUNCTION
eng = matlab.engine.start_matlab()
x_add, y_add = eng.gen_labels(matlab.double((x_rej.T).tolist()),
matlab.double([time_bound]),
nargout=2)
eng.quit()
#x_add = np.array(x_add).T
y_add = np.array(y_add[0])
print(
"STAGING: uncertain points labeled in {} seconds".format(time.time() -
start))
return x_rej_scaled, y_add #scaler.transform(x_add)