def testMultiIOTaskManager(self):
maxlen = 100
padchr = '['
task = code_tasks.make_paper_task(
'print', timestep_limit=maxlen, do_code_simplification=False)
reward_fns = task.rl_batch(1)
r = reward_fns[0]
self.assertClose(
r(pad('++++++++.---.+++++++...', maxlen, padchr)).episode_rewards[-1],
0.2444)
self.assertClose(
r(pad('++++++++.---.+++++++..+++.',
maxlen, padchr)).episode_rewards[-1],
1.0)
task = code_tasks.make_paper_task(
'print', timestep_limit=maxlen, do_code_simplification=True)
reward_fns = task.rl_batch(1)
r = reward_fns[0]
self.assertClose(
r('++++++++.---.+++++++...').episode_rewards[-1],
0.2444)
self.assertClose(
r('++++++++.---.+++++++..+++.').episode_rewards[-1],
0.935)
self.assertClose(
r(pad('++++++++.---.+++++++..+++.',
maxlen, padchr)).episode_rewards[-1],
0.75)
task = code_tasks.make_paper_task(
'reverse', timestep_limit=maxlen, do_code_simplification=False)
reward_fns = task.rl_batch(1)
r = reward_fns[0]
self.assertClose(
r(pad('>,>,>,.<.<.<.', maxlen, padchr)).episode_rewards[-1],
0.1345)
self.assertClose(
r(pad(',[>,]+[,<.]', maxlen, padchr)).episode_rewards[-1],
1.0)
task = code_tasks.make_paper_task(
'reverse', timestep_limit=maxlen, do_code_simplification=True)
reward_fns = task.rl_batch(1)
r = reward_fns[0]
self.assertClose(r('>,>,>,.<.<.<.').episode_rewards[-1], 0.1324)
self.assertClose(r(',[>,]+[,<.]').episode_rewards[-1], 0.9725)
self.assertClose(
r(pad(',[>,]+[,<.]', maxlen, padchr)).episode_rewards[-1],
0.75)
def testKnownCodeBaseTask(self):
maxlen = 100
padchr = '['
task = code_tasks.make_paper_task('shift-left',
timestep_limit=maxlen,
do_code_simplification=False)
reward_fns = task.rl_batch(1)
r = reward_fns[0]
self.assertClose(
r(pad(',>,[.,]<.,.', maxlen, padchr)).episode_rewards[-1], 1.0)
def testKnownCodeBaseTask(self):
maxlen = 100
padchr = '['
task = code_tasks.make_paper_task(
'shift-left', timestep_limit=maxlen, do_code_simplification=False)
reward_fns = task.rl_batch(1)
r = reward_fns[0]
self.assertClose(
r(pad(',>,[.,]<.,.', maxlen, padchr)).episode_rewards[-1],
1.0)