feed_dict = {x_placeholder: x_eval,
dropout_placeholder: dropout}
predictions = []
for _ in range(n_passes):
predictions.append(sess.run(prediction_op, feed_dict)[0])
y_eval = np.mean(predictions, axis=0).flatten()
uncertainty_eval = np.var(predictions, axis=0).flatten()
plotting.plot_mean_vs_truth(x, y, x_eval, y_eval, uncertainty_eval, ax)
if __name__ == "__main__":
dropout_values = [0.1, 0.3, 0.5, 0.6]
fig, axs = plt.subplots(len(dropout_values), 1, figsize=(30, 5*len(dropout_values)), sharey=True)
x, y = generate_osband_sin_samples()
#x_train, y_train, x_test,y_test = get_mnist_data()
for dropout, ax in zip(dropout_values, axs):
ax.set_title("%.3f Dropout" % dropout)
dropout_evaluation(x, y, dropout, 1e-3, 20000, 100, ax)
fig.savefig("Dropout_Sinus.pdf")
fig, axs = plt.subplots(len(dropout_values), 1, figsize=(30, 5*len(dropout_values)), sharey=True)
x, y = generate_osband_nonlinear_samples()
for dropout, ax in zip(dropout_values, axs):
ax.set_title("%.3f Dropout" % dropout)
dropout_evaluation(x, y, dropout, 1e-3, 20000, 100, ax)
fig.savefig("Dropout_Nonlinear.pdf")
# ax.fill_between(x_eval.flatten(), 0, epistemic_eval, label="epistemic", color="green", alpha=0.4)
# ax.fill_between(x_eval.flatten(), 0, aleatoric_eval, label="aleatoric", color="orange", alpha=0.4)
ax.legend()
if __name__ == "__main__":
dropout_values = [0.1, 0.2, 0.3, 0.5]
fig, axs = plt.subplots(len(dropout_values),
1,
figsize=(30, 5 * len(dropout_values)),
sharey=True)
axs[0].set_ylim([-1, 3])
fig.suptitle('Combined-Model | Epochs: 15000, Learning Rate: 1e-3',
fontsize=20)
x, y = sample_generators.generate_osband_sin_samples(60)
for dropout, ax in zip(dropout_values, axs):
ax.set_title("%.3f Dropout" % dropout)
combined_evaluation(x, y, dropout, 1e-3, 20000, 500, ax)
fig.savefig("Combined_Sinus.pdf")
fig, axs = plt.subplots(len(dropout_values),
1,
figsize=(30, 5 * len(dropout_values)),
sharey=True)
fig.suptitle('Combined-Model | Epochs: 15000, Learning Rate: 1e-3',
fontsize=20)
x, y = sample_generators.generate_osband_nonlinear_samples()
for dropout, ax in zip(dropout_values, axs):
ax.set_title("%.3f Dropout" % dropout)
combined_evaluation(x, y, dropout, 1e-3, 20000, 500, ax)