def onerun(r):
avg_payouts_in_run = []
learned_actions_in_run = []
for prediction_accuracy in np.linspace(linspace_from, linspace_to,
linspace_steps):
problem = Newcomb(predictor_accuracy=prediction_accuracy,
payouts=np.array([[1000000, 0], [1001000,
1000]]))
agent = SARSANewcombAgent(problem,
alpha=0.1,
gamma=0.9,
epsilon=0.9)
log.info('Playing ...')
log.info('%s' % (str(agent)))
log.info('%s' % (str(problem)))
_, payouts = interact_multiple(agent, problem, interactions)
avg_payout = payouts.mean(axis=0)
avg_payouts_in_run.append(avg_payout)
log.info('Average Payout for predicion accuraccy %.3f: %.3f' %
(prediction_accuracy, avg_payout))
learned_actions_in_run.append(agent.get_learned_action())
return (np.array(avg_payouts_in_run), np.array(learned_actions_in_run))
@author: Dominik Meyer <*****@*****.**>
"""
import sys
sys.path.append('..')
sys.path.append('.')
import logging as log
import numpy as np
log.basicConfig(level=log.DEBUG)
from problems import Newcomb
from agents import SARSANewcombAgent
from experiment_helpers import interact_multiple
if __name__ == '__main__':
problem = Newcomb(predictor_accuracy=0.1,
payouts=np.array([[1000000, 0], [1001000, 1000]]))
agent = SARSANewcombAgent(problem, alpha=0.1, gamma=0.9, epsilon=0.9)
interactions = 10000
log.info('Playing ...')
log.info('%s' % (str(agent)))
log.info('%s' % (str(problem)))
_, payouts = interact_multiple(agent, problem, interactions)
log.info('Average Payout: %f, Learned Action: %i' %
(payouts.mean(axis=0), agent.get_learned_action()))
payouts=np.array([[1000000, 0],
[1001000, 1000]]))
problem2 = RandomNewcomb(predictor_accuracy=predictor_accuracy,
payouts=np.array([[1000000, 0],
[1001000, 1000]]))
agent1 = OneBoxNewcombAgent(problem1)
agent2 = TwoBoxNewcombAgent(problem2)
log.info('Playing ...')
log.info('%s' % (str(agent1)))
log.info('%s' % (str(problem1)))
log.info(' VERSUS')
log.info('%s' % (str(agent2)))
log.info('%s' % (str(problem2)))
_, payouts1 = interact_multiple(agent1, problem1, interactions)
_, payouts2 = interact_multiple(agent2, problem2, interactions)
avg_payout1 = payouts1.mean(axis=0)
avg_payout2 = payouts2.mean(axis=0)
avg_payouts1.append(avg_payout1)
avg_payouts2.append(avg_payout2)
log.info('Average Payout: %.3f vs. %.3f' % (avg_payout1, avg_payout2))
avg_payouts1 = np.array(avg_payouts1)
avg_payouts2 = np.array(avg_payouts2)
plot_that_pretty_rldm15([
np.linspace(linspace_from, linspace_to, linspace_steps),
np.linspace(linspace_from, linspace_to, linspace_steps)