def predict(self, data):
o = Output(data)
x = self.transform.transform(data.x)
y = x.dot(self.w) + self.b
o.fu = y
o.y = y
return o
def predict(self, data):
o = Output(data)
x = self.transform.transform(data.x)
y = x.dot(self.w) + self.b
o.fu = y
o.y = y
return o
def predict(self, data):
o = Output(data)
#W = pairwise.rbf_kernel(data.x,self.x,self.sigma)
x = data.x
if self.use_stacking:
x = self.get_stacking_x(data)
x = self.transform.transform(x)
o.y = x.dot(self.w) + self.b
o.fu = o.y
o.w = self.w
o.true_w = self.true_w
return o
def predict(self, data):
o = Output(data)
x = data.x
if self.transform is not None:
x = self.transform.transform(x)
y = x.dot(self.w) + self.b
# y = np.round(y)
# y[y >= .5] = 1
# y[y < .5] = 0
y = np.sign(y)
o.y = y
o.fu = y
if self.label_transform is not None:
o.true_y = self.label_transform.transform(o.true_y)
if not self.running_cv:
is_correct = o.y == o.true_y
mean_train = is_correct[o.is_train].mean()
mean_test = is_correct[o.is_test].mean()
mean_train_labeled = is_correct[data.is_train & data.is_labeled].mean()
pass
return o
def predict(self, data):
o = Output(data)
x = data.x
if self.transform is not None:
x = self.transform.transform(x)
y = x.dot(self.w) + self.b
#y = np.round(y)
#y[y >= .5] = 1
#y[y < .5] = 0
y = np.sign(y)
o.y = y
o.fu = y
if self.label_transform is not None:
o.true_y = self.label_transform.transform(o.true_y)
if not self.running_cv:
is_correct = (o.y == o.true_y)
mean_train = is_correct[o.is_train].mean()
mean_test = is_correct[o.is_test].mean()
mean_train_labeled = is_correct[data.is_train
& data.is_labeled].mean()
pass
return o
def predict(self, data):
o = Output(data)
#W = pairwise.rbf_kernel(data.x,self.x,self.sigma)
x = data.x
if self.use_stacking:
x = self.get_stacking_x(data)
x = self.transform.transform(x)
if self.method == MixedFeatureGuidanceMethod.METHOD_LASSO:
o.y = self.learner_lasso.predict(x)
o.fu = o.y
else:
o.y = x.dot(self.w) + self.b
o.fu = o.y
o.w = self.w
o.true_w = self.true_w
return o
def predict(self, data):
I = self.selected_data.is_labeled
#self.f_x = self.target_learner.predict(data).y
self.f_x = data.true_y
self.p_x = density.tune_and_predict_density(self.full_data.x, data.x, bandwidths)
self.f_s = self.subset_learner.predict(data).y
self.p_s = density.tune_and_predict_density(self.selected_data.x[I], data.x, bandwidths)
self.var_x = np.abs(data.true_y - self.target_learner.predict(data).y)
self.var_s = np.abs(data.true_y - self.f_s)
res_p = np.abs(self.p_x - self.p_s) / np.linalg.norm(self.p_x)
'''
if self.use_var:
res_var = np.abs(self.var_x - self.var_s) / np.linalg.norm(self.var_x)
res_total = res_p + res_var
else:
res_f = np.abs(self.f_x - self.f_s) / np.linalg.norm(self.f_x)
res_total = res_f + res_p
'''
res_f = np.abs(self.f_x - self.f_s) / np.linalg.norm(self.f_x)
res_total = res_f + res_p
self.res_total = res_total
o = Output(data)
o.res_total = res_total
o.true_p = self.p_x
o.p = self.p_s
o.true_y = self.f_x
o.y = self.f_s
o.var_x = self.var_x
o.var_s = self.var_s
o.optimization_value = self.optimization_value/data.n
o.is_noisy = self.is_noisy
o.is_selected = self.learned_distribution > 0
o.y_orig = self.y_orig
o.x_orig = getattr(data, 'x_orig', None)
self.output = o
return o
