algorithms.AGen package

Submodules

algorithms.AGen.my_gaussian_gru_policy module

class algorithms.AGen.my_gaussian_gru_policy.myGaussianGRUPolicy(name, env_spec, hidden_dim=32, feature_network=None, state_include_action=True, hidden_nonlinearity=<function tanh>, gru_layer_cls=<class 'sandbox.rocky.tf.core.layers.GRULayer'>, learn_std=True, init_std=1.0, output_nonlinearity=None)[source]

Bases: sandbox.rocky.tf.policies.base.StochasticPolicy, sandbox.rocky.tf.core.layers_powered.LayersPowered, rllab.core.serializable.Serializable

dist_info_sym(obs_var, state_info_vars)[source]

Return the symbolic distribution information about the actions. :param obs_var: symbolic variable for observations :param state_info_vars: a dictionary whose values should contain information about the state of the policy at the time it received the observation :return:

property distribution

:rtype Distribution

get_action(observation)[source]
get_actions(observations)[source]
get_actions_with_prev(observations, prev_actions, prev_hiddens)[source]
log_diagnostics(paths)[source]

Log extra information per iteration based on the collected paths

property recurrent

Indicates whether the policy is recurrent. :return:

reset(dones=None)[source]
property state_info_specs

Return keys and shapes for the information related to the policy’s state when taking an action. :return:

property vectorized

Indicates whether the policy is vectorized. If True, it should implement get_actions(), and support resetting with multiple simultaneous states.

algorithms.AGen.rls module

class algorithms.AGen.rls.rls(lbd, theta, nn_dim, output_dim)[source]

Bases: object

docstring for ClassName

initialize()[source]
predict(hidden_vec)[source]
Parameters

hidden_vec – hidden vector

Returns

predicted result using current theta

update(hidden_vec, obs_Y)[source]
Parameters

  • hidden_vec – hidden vector

  • obs_Y – ground truth label

Returns

update the theta

algorithms.AGen.validate_utils module

algorithms.AGen.validate_utils.build_ngsim_env(args, exp_dir='/tmp', alpha=0.001, vectorize=False, render_params=None, videoMaking=False)[source]
algorithms.AGen.validate_utils.build_policy(args, env, latent_sampler=None)[source]
algorithms.AGen.validate_utils.get_ground_truth(ngsim_filename: str, h5_filename: str)[source]

Namespace(batch_size=10000, critic_batch_size=1000, critic_dropout_keep_prob=0.8, critic_grad_rescale=40.0, critic_hidden_layer_dims=(128, 128, 64), critic_learning_rate=0.0004, decay_reward=False, discount=0.95, do_curriculum=False, env_H=200, env_action_repeat=1, env_multiagent=False, env_primesteps=50, env_reward=0, exp_dir=’../../data/experiments’, exp_name=’singleagent_def_3’, expert_filepath=’../../data/trajectories/ngsim.h5’, gradient_penalty=2.0, itrs_per_decay=25, latent_dim=4, load_params_init=’NONE’, max_path_length=1000, n_critic_train_epochs=40, n_envs=1, n_envs_end=50, n_envs_start=10, n_envs_step=10, n_itr=1000, n_recognition_train_epochs=30, ngsim_filename=’trajdata_i101_trajectories-0750am-0805am.txt’, normalize_clip_std_multiple=10.0, params_filepath=’‘, policy_mean_hidden_layer_dims=(128, 128, 64), policy_recurrent=True, policy_std_hidden_layer_dims=(128, 64), recognition_hidden_layer_dims=(128, 64), recognition_learning_rate=0.0005, recurrent_hidden_dim=64, remove_ngsim_veh=False, render_every=25, reward_handler_critic_final_scale=1.0, reward_handler_max_epochs=100, reward_handler_recognition_final_scale=0.2, reward_handler_use_env_rewards=True, scheduler_k=20, trpo_step_size=0.01, use_critic_replay_memory=True, use_infogail=False, validator_render=False, vectorize=True)

algorithms.AGen.validate_utils.get_multiagent_ground_truth(ngsim_filename: str, h5_filename: str)[source]

Namespace(batch_size=10000, critic_batch_size=1000, critic_dropout_keep_prob=0.8, critic_grad_rescale=40.0, critic_hidden_layer_dims=(128, 128, 64), critic_learning_rate=0.0004, decay_reward=False, discount=0.95, do_curriculum=False, env_H=200, env_action_repeat=1, env_multiagent=False, env_primesteps=50, env_reward=0, exp_dir=’../../data/experiments’, exp_name=’singleagent_def_3’, expert_filepath=’../../data/trajectories/ngsim.h5’, gradient_penalty=2.0, itrs_per_decay=25, latent_dim=4, load_params_init=’NONE’, max_path_length=1000, n_critic_train_epochs=40, n_envs=1, n_envs_end=50, n_envs_start=10, n_envs_step=10, n_itr=1000, n_recognition_train_epochs=30, ngsim_filename=’trajdata_i101_trajectories-0750am-0805am.txt’, normalize_clip_std_multiple=10.0, params_filepath=’‘, policy_mean_hidden_layer_dims=(128, 128, 64), policy_recurrent=True, policy_std_hidden_layer_dims=(128, 64), recognition_hidden_layer_dims=(128, 64), recognition_learning_rate=0.0005, recurrent_hidden_dim=64, remove_ngsim_veh=False, render_every=25, reward_handler_critic_final_scale=1.0, reward_handler_max_epochs=100, reward_handler_recognition_final_scale=0.2, reward_handler_use_env_rewards=True, scheduler_k=20, trpo_step_size=0.01, use_critic_replay_memory=True, use_infogail=False, validator_render=False, vectorize=True)

algorithms.AGen.validate_utils.load_validate_data(filepath, act_keys=['accel', 'turn_rate_global'], ngsim_filename='trajdata_holo_trajectories.txt', debug_size=None, min_length=50, normalize_data=True, shuffle=False, act_low=-1, act_high=1, clip_std_multiple=inf)[source]
algorithms.AGen.validate_utils.normalize(x, clip_std_multiple=inf)[source]
algorithms.AGen.validate_utils.normalize_range(x, low, high)[source]

Module contents