def train_model(pathID, model_fit, numbvaccines, steps, efficacy, iterations,
open_port):
save_path = master_path + '/user_data/' + str(pathID)
#Cleaning the data
names, pars, total_pops, i_states = clean_data(model_fit)
print("Cleaned Data")
#Start training
env = CustomEnv(pars, i_states, total_pops, numbvaccines, steps, names,
efficacy)
train_agent(env, 0.0001, iterations, save_path, open_port)
print("\n Training Complete \n")
#Evaluate the model
policy_data = rewards_from_policy(env, save_path + "/best_model")
print("All Rewards Data")
np.save(save_path + "/policy_data.npy", policy_data)
model = PPO2(MlpPolicy,
env,
policy_kwargs=policy_kwargs,
tensorboard_log=tensorboard_path,
gamma=1.0,
nminibatches=2)
cmd_str = "tensorboard --logdir " + tensorboard_path + " --port=" + str(
port)
print(cmd_str)
p = subprocess.Popen(cmd_str.split(" "))
print(total_episodes * env.max_steps)
model.learn(total_timesteps=total_episodes * env.max_steps, callback=cb)
model.save(path + "/ppo2_vaccine")
p.kill()
if __name__ == "__main__": #Just to test
place_names, params, total_populations, initial_states = clean_data(
np.load(
"/home/steelshot/nwHacksVaccDistr/model_fitting/data/tests.npy",
allow_pickle=True))
env = CustomEnv(params, initial_states, total_populations, 500000, 180,
place_names, 0.95)
train_agent(env, 0.0003, 10000,
"/home/steelshot/nwHacksVaccDistr/optimization/models")
def run(countries,
states,
steps,
numbvaccines,
efficacy,
pathID,
iterations=100):
"""
INPUT PARAMETERS:
----------------
countries : array-like
The countries that the user is interested in.
states : array-like (or None)
The states/provinces that the user would like to consider if they are keeping the country fixed. Default is None.
steps : int
The number of days that the user wants to run the simulation.
numbvaccines : int
The number of vaccines that the user wants to allocate per day.
efficacy : float
The percentage effectiveness of the vaccine that they have chosen
pathID : str
The ID of the path for the user
iterations : int (default to 100)
The number of iterations for the training to run
"""
#First getting the model fitting params
fitting = covsir_models(countries, states)
results = fitting.calling()
print("Fitted data")
print(results)
#Cleaning the data
names, pars, total_pops, i_states = clean_data(results)
print("Cleaned Data")
#Start training
env = CustomEnv(pars, i_states, total_pops, numbvaccines, steps, names,
efficacy)
train_agent(env, 0.0003, iterations,
master_path + "/user_data/" + str(pathID))
print("\n Training Complete \n")
model_path = master_path + "/user_data/" + str(
pathID) + "/best_model" #Defining the model path
#Evaluate the model
all_rewards = rewards_from_policy(env, numbvaccines, steps, efficacy,
model_path)
print("All Rewards Data")
#Get data from policies
data = data_from_policy(env, model_path)
np.save(master_path + "/user_data/" + str(pathID) + "/policy_data.npy",
data)
print("Data from policy")
plt.plot(time, policy_results["learned_policy"]["deaths"])
plt.plot(time, policy_results["no_vaccine"]["deaths"])
plt.subplot(2, 2, 3)
plt.ylabel("Succeptible")
plt.plot(time, policy_results["infected_ratio"]["susceptible"])
plt.plot(time, policy_results["population_ratio"]["susceptible"])
plt.plot(time, policy_results["learned_policy"]["susceptible"])
plt.plot(time, policy_results["no_vaccine"]["susceptible"])
plt.subplot(2, 2, 4)
plt.ylabel("Recovered")
plt.plot(time, policy_results["infected_ratio"]["recovered"])
plt.plot(time, policy_results["population_ratio"]["recovered"])
plt.plot(time, policy_results["learned_policy"]["recovered"])
plt.plot(time, policy_results["no_vaccine"]["recovered"])
plt.show()
if __name__ == "__main__": #Just testing to see if it works, this won't run otherwise
place_names, params, total_populations, initial_states = clean_data(
np.load(
"/home/steelshot/nwHacksVaccDistr/user_data/6/fitted_model.npy",
allow_pickle=True))
env = CustomEnv(params, initial_states, total_populations, 100000, 30,
place_names, 0.95)
policy_results = rewards_from_policy(
env, 180, 0.95,
"/home/steelshot/nwHacksVaccDistr/user_data/6/best_model")
print(policy_results["policy_comparision"]) #Just call this
visualize_results(policy_results)