dagger_lr = LinearRegression(fit_intercept=False)#KernelRidge(kernel='rbf')# dagger_learner = SKLearner(dagger_lr) dagger_trajs, dagger_traj_controls, dagger_costs, dagger_avg_costs, dagger_avg_loss, dagger_accs = trials.dagger_trial(dagger_learner, robot, sys) dagger_final_trajs, dagger_final_controls, dagger_final_costs, dagger_final_avg_costs, dagger_final_avg_loss = trials.dagger_final(dagger_learner, robot, sys) dagger_data[t, :] = dagger_avg_costs dagger_loss_data[t, :] = dagger_avg_loss dagger_final_data[t, :] = dagger_final_avg_costs dagger_acc_data[t, :] = dagger_accs # print im_lr.coef_ # print dagger_lr.coef_ # print robot.lqr.K1 vis = Visualizer() vis.show_trajs(sup_trajs, x_f, "sup_trajs", data_directory) vis.show_trajs(im_trajs, x_f, "sl_trajs", data_directory) vis.show_trajs(dagger_trajs, x_f, "dagger_trajs", data_directory) vis.show_trajs(dagger_final_trajs, x_f, "dagger_final_trajs", data_directory) print im_accs print dagger_accs print "\n\n\n" for state, control in im_learner.data: print control print im_learner.predict(state) print dagger_learner.predict(state) state = state.reshape((xdims, 1))
print "\nLearner: " print learner.estimator.coef_ print "\nLQR: " print robot.lqr.K trajs4 = [] for i in range(10): sys.reset_robot() states, controls, costs = robot.rollout_learner(learner, verbose=False) trajs4.append(states) vis = Visualizer() vis.show_trajs(trajs1, x_f, "Trajs 1") vis.show_trajs(trajs2, x_f, "Trajs 2") vis.show_trajs(trajs3, x_f, "Trajs 3") vis.show_trajs(trajs4, x_f, "Trajs 4") """vis = Visualizer() vis.set_recording(states) vis.set_target(x_f) vis.show() print "Learner cost: " + str(sum(costs)) """ #print np.dot(learner.estimator.coef_, init_state) #print learner.predict(init_state)