def run_silent_save(params):
"""Run the agent for a finite number of trials."""
print 'run_silent({})'.format(params)
#...run it 10 times to verify... (just for now)
repeat = 1
score = 0.0
scores = []
penalties = []
for _ in range(repeat):
# Set up environment and agent
e = Environment(params) # create environment (also adds some dummy traffic)
a = e.create_agent(LearningAgent) # create agent
e.set_primary_agent(a, enforce_deadline=True) # set agent to track
# Now simulate it
sim = Simulator(e, update_delay=0.0,silent=True) # reduce update_delay to speed up simulation
score += sim.run(n_trials=params['max_epoch']) # press Esc or close pygame window to quit
scores += [sim.getScores()]
penalties += [sim.getPenalties()]
print 'state span size : {}'.format(len(a.qtable))
losses = sim.getLosses() #cannot average over repeated runs
a.print_policy() #[debug]
scores = np.average(scores,0)
penalties = np.average(penalties,0)
if params['save']:
fscores = open('score.csv','w+')
fpenalties = open('penalties.csv','w+')
flosses = open('losses.csv','w+')
for s in scores:
fscores.write(str(s) + '\n')
fscores.flush()
fscores.close()
for p in penalties:
fpenalties.write(str(p) + '\n')
fpenalties.flush()
fpenalties.close()
for l in losses:
flosses.write(str(l) + '\n')
flosses.flush()
flosses.close()
score /= float(repeat)
return params, score
