def analyze():
exp_params = ExperimentParameters()
model_output_dir = "../../../models/sarsa/reward/%i" % exp_params.seed
exp_params.model_output_dir = model_output_dir
image_output_dir = "../../../images/sarsa/reward/%i" % exp_params.seed
exp_params.image_output_dir = image_output_dir
analyze_models(exp_params)
def analyze(): exp_params = ExperimentParameters() model_output_dir = "../../../models/qlearning/state/%i" % exp_params.seed exp_params.model_output_dir = model_output_dir image_output_dir = "../../../images/qlearning/state/%i" % exp_params.seed exp_params.image_output_dir = image_output_dir analyze_models(exp_params)
def analyze():
exp_params = ExperimentParameters()
exp_params.seed = get_reward_seeds()[2]
model_output_dir = "../../../models/qlearning/reward/%i" % exp_params.seed
exp_params.model_output_dir = model_output_dir
image_output_dir = "../../../images/qlearning/reward/%i" % exp_params.seed
exp_params.image_output_dir = image_output_dir
analyze_models(exp_params)
def analyze(): exp_params = ExperimentParameters() exp_params.seed = get_parameters_seed() model_output_dir = "../../../models/expected_sarsa/params/%i" % exp_params.seed exp_params.model_output_dir = model_output_dir image_output_dir = "../../../images/expected_sarsa/params/%i" % exp_params.seed exp_params.image_output_dir = image_output_dir analyze_models(exp_params)
def analyze_aggregated():
exp_params = ExperimentParameters()
exp_params.seed = get_reward_seeds()[0]
image_output_dir = "../../../images/expected_sarsa/reward/%i" % exp_params.seed
exp_params.image_output_dir = image_output_dir
aggregated_models = get_aggregated_models("expected_sarsa", "reward", exp_params, get_reward_seeds())
filenames = list(aggregated_models.keys())
models = list(aggregated_models.values())
analyze_aggregated_models(filenames, models, exp_params)
def analyze_aggregated():
exp_params = ExperimentParameters()
exp_params.seed = get_state_seeds()[0]
image_output_dir = "../../../images/qlearning/state/%i" % exp_params.seed
exp_params.image_output_dir = image_output_dir
aggregated_models = get_aggregated_models("qlearning", "state", exp_params, get_state_seeds())
filenames = list(aggregated_models.keys())
models = list(aggregated_models.values())
analyze_aggregated_models(filenames, models, exp_params)
def analyze_aggregated_reward_food_count_correlations(algorithm):
def get_corr_coefs(models):
corr_coefs = []
for model in models:
average_reward_over_time = compute_mean_over_time(
model.rewards_per_episode)
average_food_count_over_time = compute_mean_over_time(
model.food_count_per_episode)
if np.all(average_reward_over_time == 0):
continue
corr_coef = np.corrcoef(average_reward_over_time,
average_food_count_over_time)
corr_coefs.append(corr_coef)
return corr_coefs
exp_params = ExperimentParameters()
exp_params.seed = get_reward_seeds()[0]
aggregated_models = get_aggregated_models(algorithm, "reward", exp_params,
get_reward_seeds())
states = [filename.split("_")[0] for filename in aggregated_models.keys()]
models = list(aggregated_models.values())
current_models, _ = \
filter_models_by_state(models, states,
lambda state: state.startswith("Board"))
corr_coefs = get_corr_coefs(current_models)
print("Board State")
print("Average correlation:", np.mean(corr_coefs))
print("Average absolute correlation:", np.mean(np.abs(corr_coefs)))
current_models, _ = \
filter_models_by_state(models, states,
lambda state: state.startswith("DirectionalDistance"))
corr_coefs = get_corr_coefs(current_models)
print("Directional Distance State")
print("Average correlation:", np.mean(corr_coefs))
print("Average absolute correlation:", np.mean(np.abs(corr_coefs)))
def train():
for params in get_snake_parameters():
env = SnakeEnvironment(params)
params.policy = EpsilonGreedyPolicy(env, params.epsilon)
exp_params = ExperimentParameters()
exp_params.env = env
exp_params.model_class = Sarsa
exp_params.model_params = params
exp_params.seed = get_parameters_seed()
output_dir = "../../../models/sarsa/params/%i" % exp_params.seed
exp_params.model_output_dir = output_dir
train_and_store_model(exp_params)
def analyze_test():
params = SnakeParameters()
params.discount_factor = StaticDiscountFactor(0.95)
params.learning_rate = StaticLearningRate(0.15)
env = SnakeEnvironment(params)
params.policy = GreedyPolicy(env)
exp_params = ExperimentParameters()
exp_params.env = env
exp_params.model_class = Sarsa
exp_params.model_params = params
model_output_dir = "../../../models/sarsa/reward/%i" % exp_params.seed
exp_params.model_output_dir = model_output_dir
image_output_dir = "../../../images/sarsa/reward/%i" % exp_params.seed
exp_params.image_output_dir = image_output_dir
analyze_models_test(exp_params)
def train():
params = SnakeParameters()
params.discount_factor = StaticDiscountFactor(0.95)
params.learning_rate = StaticLearningRate(0.15)
params.reward = get_state_reward()
env = SnakeEnvironment(params)
params.policy = EpsilonGreedyPolicy(env, params.epsilon)
exp_params = ExperimentParameters()
exp_params.env = env
exp_params.model_class = Sarsa
exp_params.model_params = params
for seed in get_state_seeds():
exp_params.seed = seed
output_dir = "../../../models/sarsa/state/%i" % exp_params.seed
exp_params.model_output_dir = output_dir
train_models(exp_params)
def train():
params = SnakeParameters()
params.discount_factor = StaticDiscountFactor(0.85)
params.learning_rate = StaticLearningRate(0.85)
env = SnakeEnvironment(params)
params.policy = EpsilonGreedyPolicy(env, params.epsilon)
exp_params = ExperimentParameters()
exp_params.env = env
exp_params.model_class = QLearning
exp_params.model_params = params
for seed in get_reward_seeds():
exp_params.seed = seed
output_dir = "../../../models/qlearning/reward/%i" % exp_params.seed
exp_params.model_output_dir = output_dir
for state in get_reward_states():
exp_params.model_params.state = state
train_models(exp_params)