def test_get_mean_and_standard_deviation_difference_results():
"""Tests that get_mean_and_standard_deviation_difference_results method produces correct output"""
results = [[1.0, 2.0, 3.0], [5.0, -33.0, 55.0], [2.5, 2.5, 2.5]]
mean_results = [
np.mean([1.0, 5.0, 2.5]),
np.mean([2.0, -33.0, 2.5]),
np.mean([3.0, 55.0, 2.5])
]
std_results = [
np.std([1.0, 5.0, 2.5]),
np.std([2.0, -33.0, 2.5]),
np.std([3.0, 55.0, 2.5])
]
mean_minus_1_std = [
mean - std_val for mean, std_val in zip(mean_results, std_results)
]
mean_plus_1_std = [
mean + std_val for mean, std_val in zip(mean_results, std_results)
]
config = Config()
config.standard_deviation_results = 1.0
trainer = Trainer(config, [])
mean_minus_x_std_guess, mean_results_guess, mean_plus_x_std_guess = trainer.get_mean_and_standard_deviation_difference_results(
results)
assert mean_results == mean_results_guess
assert mean_minus_1_std == mean_minus_x_std_guess
assert mean_plus_1_std == mean_plus_x_std_guess
config.standard_deviation_results = 3.0
trainer = Trainer(config, [])
mean_minus_x_std_guess, mean_results_guess, mean_plus_x_std_guess = trainer.get_mean_and_standard_deviation_difference_results(
results)
mean_plus_3_std = [
mean + 3.0 * std_val
for mean, std_val in zip(mean_results, std_results)
]
mean_minus_3_std = [
mean - 3.0 * std_val
for mean, std_val in zip(mean_results, std_results)
]
assert mean_results == mean_results_guess
assert mean_minus_3_std == mean_minus_x_std_guess
assert mean_plus_3_std == mean_plus_x_std_guess
from environments.Bit_Flipping_Environment import Bit_Flipping_Environment
from drl.agents.policy_gradient_agents.PPO import PPO
from environments.Four_Rooms_Environment import Four_Rooms_Environment
from drl.agents.hierarchical_agents.SNN_HRL import SNN_HRL
from drl.agents.actor_critic_agents.TD3 import TD3
from drl.agents.Trainer import Trainer
from drl.utilities.data_structures.Config import Config
from drl.agents.DQN_agents.DQN import DQN
import numpy as np
import torch
random.seed(1)
np.random.seed(1)
torch.manual_seed(1)
config = Config()
config.seed = 1
config.environment = Bit_Flipping_Environment(4)
config.num_episodes_to_run = 2000
config.file_to_save_data_results = None
config.file_to_save_results_graph = None
config.visualise_individual_results = False
config.visualise_overall_agent_results = False
config.randomise_random_seed = False
config.runs_per_agent = 1
config.use_GPU = False
config.hyperparameters = {
"DQN_Agents": {
"learning_rate": 0.005,
"batch_size": 64,
"buffer_size": 40000,
import gym
from drl.agents.hierarchical_agents.HRL.HRL import HRL
from drl.agents.Trainer import Trainer
from drl.utilities.data_structures.Config import Config
config = Config()
config.environment = gym.make("Taxi-v2")
config.seed = 1
config.env_parameters = {}
config.num_episodes_to_run = 2000
config.file_to_save_data_results = None
config.file_to_save_results_graph = None
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 3
config.use_GPU = False
config.overwrite_existing_results_file = False
config.randomise_random_seed = True
config.save_model = False
linear_hidden_units = [32, 32]
learning_rate = 0.01
buffer_size = 100000
batch_size = 256
batch_norm = False
embedding_dimensionality = 10
gradient_clipping_norm = 5
update_every_n_steps = 1
import gym
import pytest
from drl.utilities.Utility_Functions import flatten_action_id_to_actions
from drl.utilities.data_structures.Config import Config
config = Config()
config.seed = 1
config.environment = gym.make("Taxi-v2")
config.env_parameters = {}
config.num_episodes_to_run = 1000
config.file_to_save_data_results = None
config.file_to_save_results_graph = None
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 3
config.use_GPU = False
config.overwrite_existing_results_file = False
config.randomise_random_seed = True
config.save_model = False
linear_hidden_units = [10, 5]
learning_rate = 0.01
buffer_size = 40000
batch_size = 256
batch_norm = False
embedding_dimensionality = 15
gradient_clipping_norm = 5
update_every_n_steps = 1
import gym
from environments.Atari_Environment import make_atari_game
from drl.agents.DQN_agents.DDQN import DDQN
from drl.agents.hierarchical_agents.HRL.HRL import HRL
from drl.agents.hierarchical_agents.HRL.Model_HRL import Model_HRL
from drl.agents.Trainer import Trainer
from drl.utilities.data_structures.Config import Config
config = Config()
config.seed = 1
config.environment = make_atari_game("SpaceInvaders-v0")
config.env_parameters = {}
config.num_episodes_to_run = 500
config.file_to_save_data_results = "data_and_graphs/hrl_experiments/Space_Invaders_Data.pkl"
config.file_to_save_results_graph = "data_and_graphs/hrl_experiments/Space_Invaders.png"
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 10
config.use_GPU = False
config.overwrite_existing_results_file = False
config.randomise_random_seed = True
config.save_model = False
# Loss is not drawing a random sample! otherwise wouldnt jump around that much!!
linear_hidden_units = [32, 32]
learning_rate = 0.005 # 0.001 taxi
buffer_size = 1000000
from drl.agents.DQN_agents.DDQN import DDQN
from drl.agents.DQN_agents.DDQN_With_Prioritised_Experience_Replay import DDQN_With_Prioritised_Experience_Replay
from drl.agents.DQN_agents.DQN_With_Fixed_Q_Targets import DQN_With_Fixed_Q_Targets
from environments.Bit_Flipping_Environment import Bit_Flipping_Environment
from drl.agents.policy_gradient_agents.PPO import PPO
from drl.agents.Trainer import Trainer
from drl.utilities.data_structures.Config import Config
from drl.agents.DQN_agents.DQN import DQN
import numpy as np
import torch
random.seed(1)
np.random.seed(1)
torch.manual_seed(1)
config = Config()
config.seed = 1
config.environment = Bit_Flipping_Environment(4)
config.num_episodes_to_run = 1
config.file_to_save_data_results = None
config.file_to_save_results_graph = None
config.visualise_individual_results = False
config.visualise_overall_agent_results = False
config.randomise_random_seed = False
config.runs_per_agent = 1
config.use_GPU = False
config.hyperparameters = {
"DQN_Agents": {
"learning_rate": 0.005,
import gym
from drl.agents.actor_critic_agents.A2C import A2C
from drl.agents.DQN_agents.Dueling_DDQN import Dueling_DDQN
from drl.agents.actor_critic_agents.SAC_Discrete import SAC_Discrete
from drl.agents.actor_critic_agents.A3C import A3C
from drl.agents.policy_gradient_agents.PPO import PPO
from drl.agents.Trainer import Trainer
from drl.utilities.data_structures.Config import Config
from drl.agents.DQN_agents.DDQN import DDQN
from drl.agents.DQN_agents.DDQN_With_Prioritised_Experience_Replay import DDQN_With_Prioritised_Experience_Replay
from drl.agents.DQN_agents.DQN import DQN
from drl.agents.DQN_agents.DQN_With_Fixed_Q_Targets import DQN_With_Fixed_Q_Targets
config = Config()
config.seed = 1
config.environment = gym.make("CartPole-v0")
config.num_episodes_to_run = 450
config.file_to_save_data_results = "results/data_and_graphs/Cart_Pole_Results_Data.pkl"
config.file_to_save_results_graph = "results/data_and_graphs/Cart_Pole_Results_Graph.png"
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 1
config.use_GPU = True
config.overwrite_existing_results_file = False
config.randomise_random_seed = True
config.save_model = False
import pickle
from drl.utilities.data_structures.Config import Config
from Trainer import Trainer
trainer = Trainer(config=Config(), agents=None)
#
# trainer.visualise_set_of_preexisting_results(save_image_path="Four_Rooms_and_Long_Corridor.png", results_data_paths=["Long_Corridor_Results_Data.pkl", "Four_Rooms.pkl"],
# plot_titles=["Long Corridor", "Four Rooms"], y_limits=[(0.0, 0.25), (-90.0, 100.25)])
trainer.visualise_preexisting_results(
save_image_path="hrl_experiments/Taxi_graph_comparison.png",
data_path="hrl_experiments/Taxi_data.pkl",
title="Taxi v2",
y_limits=(-800.0, 0.0))
# trainer.visualise_preexisting_results(save_image_path="Long_Corridor_Graph.png", data_path="Long_Corridor_Results_Data.pkl",
# title="Long Corridor", y_limits=(0.0, 0.25))
# trainer.visualise_preexisting_results(save_image_path="Hopper_Results_Graph_Both_Agents.png", data_path="Hopper_Results_Data.pkl",
# title="Hopper") #, y_limits=(0.0, 0.25))
# trainer.visualise_set_of_preexisting_results(results_data_paths=["Cart_Pole_Results_Data.pkl",
# "Mountain_Car_Results_Data.pkl"],
# plot_titles=["Cart Pole (Discrete Actions)", "Mountain Car (Continuous Actions)"],
# save_image_path="CartPole_and_MountainCar_Graph.png")
# trainer.visualise_set_of_preexisting_results(results_data_paths=["Data_and_Graphs/Bit_Flipping_Results_Data.pkl",
# "Data_and_Graphs/Fetch_Reach_Results_Data.pkl"],
# plot_titles=["Bit Flipping", "Fetch Reach"],
import gym
from drl.agents.policy_gradient_agents.PPO import PPO
from drl.agents.actor_critic_agents.DDPG import DDPG
from drl.agents.actor_critic_agents.SAC import SAC
from drl.agents.actor_critic_agents.TD3 import TD3
from drl.agents.Trainer import Trainer
from drl.agents.hierarchical_agents.DIAYN import DIAYN
from drl.utilities.data_structures.Config import Config
config = Config()
config.seed = 1
config.environment = gym.make("Walker2d-v2")
config.num_episodes_to_run = 400
config.file_to_save_data_results = "data_and_graphs/Walker_Results_Data.pkl"
config.file_to_save_results_graph = "data_and_graphs/Walker_Results_Graph.png"
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 3
config.use_GPU = False
config.overwrite_existing_results_file = False
config.randomise_random_seed = True
config.save_model = False
actor_critic_agent_hyperparameters = {
"Actor": {
"learning_rate": 0.0003,
"linear_hidden_units": [64, 64],
"final_layer_activation": None,
"batch_norm": False,
import os
import sys
from os.path import dirname, abspath
sys.path.append(dirname(dirname(abspath(__file__))))
import gym
from drl.agents.Trainer import Trainer
from drl.agents.actor_critic_agents.DDPG import DDPG
from drl.agents.hierarchical_agents.HIRO import HIRO
from drl.utilities.data_structures.Config import Config
config = Config()
config.seed = 1
config.environment = gym.make("Reacher-v2") # Reacher-v2 "InvertedPendulum-v2") #Pendulum-v0
config.num_episodes_to_run = 1500
config.file_to_save_data_results = None
config.file_to_save_results_graph = None
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 1
config.use_GPU = False
config.overwrite_existing_results_file = False
config.randomise_random_seed = True
config.save_model = False
config.hyperparameters = {
"HIRO": {
import gym
from drl.agents.actor_critic_agents.DDPG import DDPG
from drl.agents.actor_critic_agents.DDPG_HER import DDPG_HER
from drl.utilities.data_structures.Config import Config
from drl.agents.Trainer import Trainer
config = Config()
config.seed = 1
config.environment = gym.make("FetchReach-v1")
config.num_episodes_to_run = 1000
config.file_to_save_data_results = None
config.file_to_save_results_graph = None
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 3
config.use_GPU = False
config.overwrite_existing_results_file = False
config.randomise_random_seed = True
config.save_model = False
config.hyperparameters = {
"Actor_Critic_Agents": {
"Actor": {
"learning_rate": 0.001,
"linear_hidden_units": [50, 50],
"final_layer_activation": "TANH",
"batch_norm": False,
"tau": 0.01,
from gym.wrappers import FlattenDictWrapper
from drl.agents.DQN_agents.DQN_HER import DQN_HER
from environments.Bit_Flipping_Environment import Bit_Flipping_Environment
from drl.agents.Trainer import Trainer
from drl.utilities.data_structures.Config import Config
from drl.agents.DQN_agents.DQN import DQN
config = Config()
config.seed = 1
config.environment = Bit_Flipping_Environment(14)
config.num_episodes_to_run = 4500
config.file_to_save_data_results = None #"Data_and_Graphs/Bit_Flipping_Results_Data.pkl"
config.file_to_save_results_graph = None #"Data_and_Graphs/Bit_Flipping_Results_Graph.png"
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 3
config.use_GPU = False
config.overwrite_existing_results_file = False
config.randomise_random_seed = True
config.save_model = False
config.hyperparameters = {
"DQN_Agents": {
"learning_rate": 0.001,
"batch_size": 128,
"buffer_size": 100000,
"epsilon_decay_rate_denominator": 150,
"discount_rate": 0.999,
"incremental_td_error": 1e-8,
from drl.agents.hierarchical_agents.SNN_HRL import SNN_HRL
from drl.agents.Trainer import Trainer
from drl.utilities.data_structures.Config import Config
from drl.agents.DQN_agents.DQN import DQN
from drl.agents.hierarchical_agents.h_DQN import h_DQN
from environments.Long_Corridor_Environment import Long_Corridor_Environment
config = Config()
config.seed = 1
config.env_parameters = {"stochasticity_of_action_right": 0.5}
config.environment = Long_Corridor_Environment(stochasticity_of_action_right=config.env_parameters["stochasticity_of_action_right"])
config.num_episodes_to_run = 10000
config.file_to_save_data_results = "Data_and_Graphs/Long_Corridor_Results_Data.pkl"
config.file_to_save_results_graph = "Data_and_Graphs/Long_Corridor_Results_Graph.png"
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 3
config.use_GPU = False
config.overwrite_existing_results_file = False
config.randomise_random_seed = True
config.save_model = False
config.hyperparameters = {
"h_DQN": {
"CONTROLLER": {
"batch_size": 256,
"learning_rate": 0.01,
"buffer_size": 40000,
from drl.agents.DQN_agents.DDQN import DDQN
from environments.Four_Rooms_Environment import Four_Rooms_Environment
from drl.agents.Trainer import Trainer
from drl.utilities.data_structures.Config import Config
config = Config()
config.seed = 1
height = 15
width = 15
random_goal_place = False
num_possible_states = (height * width)**(1 + 1 * random_goal_place)
embedding_dimensions = [[num_possible_states, 20]]
print("Num possible states ", num_possible_states)
config.environment = Four_Rooms_Environment(
height,
width,
stochastic_actions_probability=0.0,
random_start_user_place=True,
random_goal_place=random_goal_place)
config.num_episodes_to_run = 1000
config.file_to_save_data_results = "Data_and_Graphs/Four_Rooms.pkl"
config.file_to_save_results_graph = "Data_and_Graphs/Four_Rooms.png"
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 3
config.use_GPU = False
import gym
from drl.agents.policy_gradient_agents.PPO import PPO
from drl.agents.actor_critic_agents.DDPG import DDPG
from drl.agents.actor_critic_agents.SAC import SAC
from drl.agents.actor_critic_agents.TD3 import TD3
from drl.agents.Trainer import Trainer
from drl.utilities.data_structures.Config import Config
config = Config()
config.seed = 1
config.environment = gym.make("MountainCarContinuous-v0")
config.num_episodes_to_run = 450
config.file_to_save_data_results = None
config.file_to_save_results_graph = None
config.show_solution_score = False
config.visualise_individual_results = False
config.visualise_overall_agent_results = True
config.standard_deviation_results = 1.0
config.runs_per_agent = 3
config.use_GPU = False
config.overwrite_existing_results_file = False
config.randomise_random_seed = True
config.save_model = False
config.hyperparameters = {
"Policy_Gradient_Agents": {
"learning_rate": 0.05,
"linear_hidden_units": [30, 15],
"final_layer_activation": "TANH",
"learning_iterations_per_round": 10,