decay_param=gamma_lambda, hidden_activation_deriv=self.hidden_activation_deriv, output_activation_deriv=self.output_activation_deriv ) if __name__ == '__main__': this_dnn_obj = DNNSpec( neurons=[2], hidden_activation=DNNSpec.relu, hidden_activation_deriv=DNNSpec.relu_deriv, output_activation=DNNSpec.identity, output_activation_deriv=DNNSpec.identity_deriv ) nn = DNN( feature_funcs=FuncApproxBase.get_identity_feature_funcs(3), dnn_obj=this_dnn_obj, reglr_coeff=0., learning_rate=1., adam=True, adam_decay1=0.9, adam_decay2=0.999 ) init_eval = nn.get_func_eval((2.0, 3.0, -4.0)) print(init_eval) x_pts = np.arange(-10.0, 10.0, 0.5) y_pts = np.arange(-10.0, 10.0, 0.5) z_pts = np.arange(-10.0, 10.0, 0.5) pts = [(x, y, z) for x in x_pts for y in y_pts for z in z_pts]
mdp_ref_obj = ic.get_mdp_refined() this_tolerance = 1e-3 this_first_visit_mc = True num_samples = 30 this_softmax = True this_epsilon = 0.05 this_epsilon_half_life = 30 this_learning_rate = 0.1 this_learning_rate_decay = 1e6 this_lambd = 0.8 this_num_episodes = 3000 this_max_steps = 1000 this_tdl_fa_offline = True this_fa_spec = FuncApproxSpec( state_feature_funcs=FuncApproxBase.get_identity_feature_funcs( ic.lead_time + 1 ), action_feature_funcs=[lambda x: x], dnn_spec=DNNSpec( neurons=[2, 4], hidden_activation=DNNSpec.relu, hidden_activation_deriv=DNNSpec.relu_deriv ) ) raa = RunAllAlgorithms( mdp_refined=mdp_ref_obj, tolerance=this_tolerance, first_visit_mc=this_first_visit_mc, num_samples=num_samples, softmax=this_softmax,