def plot_prediction_1D(self, x0, bounds, idx_output=0, idx_input=0, resolution_prediction=1000):
"""
Plot the model sliced (along each dimension) around a pivot point x0
:param x0: pivot point
:param bounds: bounds for the predictions
:param idx_input:
:param resolution_prediction:
:param interactive:
:return:
"""
import matplotlib.pyplot as plt
import scipyplot as spp
# TODO:assert bounds ot type utils.bounds
# TODO: THIS IS not goirg to work FOR MULTI_DIMENSIONAL MODELS!!!!
idx = idx_input
if bounds.get_n_dim() == 1:
X = np.matrix(np.linspace(bounds.get_min(idx).flatten(), bounds.get_max(idx).flatten(), num=resolution_prediction))
else:
X = np.matrix(np.tile(x0, (resolution_prediction, 1)))
X[:, idx] = np.linspace(bounds.get_min(idx).flatten(), bounds.get_max(idx).flatten(), num=resolution_prediction)
prediction = self.predict(X)
if self.isprobabilistic():
h = spp.gauss_1D(x=X[idx, :], y=prediction[0][:, idx_output], variance=prediction[1][:, idx_output],
color='b') # X[:, idx]
else:
h = plt.plot(X[idx, :], prediction[:, idx_output], color='b', linestyle='-', linewidth=2,
label='Prediction')
def plotComparison(idx):
handle = [None, None]
handle[0] = plt.plot(dataset.output[idx, :].T, color='g', linestyle='-', linewidth=2, marker='o', label='Groundtruth')
if self.isprobabilistic():
handle[1] = spp.gauss_1D(y=prediction[0][:, idx], variance=prediction[1][:, idx], color='b')
# handle[1] = plt.plot(prediction[0][:, idx], color='b', linestyle='-', linewidth=2, label='Prediction')
else:
handle[1] = plt.plot(prediction[:, idx], color='b', linestyle='-', linewidth=2, label='Prediction')
plt.legend()
plt.title(dataset.name)
plt.ylabel(dataset.get_label_input(idx))
def plot_optimization_curve(self, scale='log', plotDelta=True):
import scipyplot as spp
logs = self.get_logs()
plt.figure()
# logs.plot_optimization_curve()
if (self.task.opt_obj is None) and (plotDelta is True):
plt.plot(logs.get_objectives().T, c='red', linewidth=2)
plt.ylabel('Obj.Func.')
n_evals = logs.data.m.shape[0]
x = np.arange(start=logs.get_n_evals() - n_evals,
stop=logs.get_n_evals())
spp.gauss_1D(y=logs.data.m,
variance=logs.data.v,
x=x,
color='blue')
if self.log_best_mean:
spp.gauss_1D(y=logs.data.best_m,
variance=logs.data.best_v,
x=x,
color='green')
else:
plt.plot(logs.get_objectives().T - self.task.opt_obj,
c='red',
linewidth=2)
plt.ylabel('Optimality gap')
n_evals = logs.data.m.shape[0]
x = np.arange(start=logs.get_n_evals() - n_evals,
stop=logs.get_n_evals())
spp.gauss_1D(y=logs.data.m - self.task.opt_obj,
variance=logs.data.v,
x=x,
color='blue')
if self.log_best_mean:
spp.gauss_1D(y=logs.data.best_m - self.task.opt_obj,
variance=logs.data.best_v,
x=x,
color='green')
plt.xlabel('N. Evaluations')
if scale == 'log':
ax = plt.gca()
ax.set_yscale('log')
# TODO: best performance expected
# if self.log_best_mean:
# plt.legend(['Performance evaluated', 'performance expected', 'Best performance expected'])
# else:
# plt.legend(['Performance evaluated', 'performance expected'])
plt.show()
def plotComparison(idx):
if self.n_inputs > 1:
X = np.tile(x0, (resolution_prediction, 1))
X[:, idx] = np.linspace(bounds.get_min(idx), bounds.get_max(idx), num=resolution_prediction)
prediction = self.predict(X)
if self.isprobabilistic():
h = spp.gauss_1D(x=X[:, idx], y=prediction[0][:, idx_output], variance=prediction[1][:, idx_output],
color='b')
else:
h = plt.plot(X[:, idx], prediction[:, idx_output], color='b', linestyle='-', linewidth=2,
label='Prediction')
plt.axvline(x=x0[idx], linestyle='--', linewidth=2, color='r')
else:
X = np.linspace(bounds.get_min(), bounds.get_max(), num=resolution_prediction).T
prediction = self.predict(X)
if self.isprobabilistic():
h = spp.gauss_1D(x=X, y=prediction[0], variance=prediction[1], color='b')
else:
h = plt.plot(X[:, idx], prediction, color='b', linestyle='-', linewidth=2, label='Prediction')
# plt.legend()
# plt.title(dataset.name)
plt.title('%d' % (idx_output))
plt.xlabel('%d' % (idx))
def test_gauss_1D_vector(self):
y = np.random.rand(100)
variance = np.random.rand(100)
h = spp.gauss_1D(y=y, variance=variance)