def jobs_parallel_learning_concurrent(context, data_central, id_agent, id_robot, episodes,
n, intermediate_reports=True, final_report=True):
"""
In this way, there are "n" instances of each agent.
Each instance sees all logs, but is given a hint using the
parallell_hint(i, n) function.
The instances are merged using the merge() function.
Needs:
RM_EPISODE_READY, id_robot, id_episode
"""
agents = []
contexts = ['learn%02dof%02d' % (i + 1, n) for i in range(n)]
for i, (c, progress) in enumerate(iterate_context_names(context, contexts)):
for id_episode in episodes:
c.needs(RM_EPISODE_READY, id_robot=id_robot, id_episode=id_episode)
agent_i = c.subtask(LearnLogNoSaveHint,
boot_root=data_central.get_boot_root(),
agent=id_agent, robot=id_robot,
episodes=episodes,
parallel_hint=[i, n],
add_job_prefix='')
if intermediate_reports:
report = c.comp(get_agentstate_report, agent_i, progress, job_id='report')
c.add_report(report, 'agent_report_partial',
id_agent=id_agent, id_robot=id_robot,
progress=progress)
agents.append(agent_i)
agent_state = jobs_merging_recursive(context, agents)
if final_report:
report = context.comp(get_agentstate_report, agent_state, 'all', job_id='report')
context.add_report(report, 'agent_report', id_agent=id_agent, id_robot=id_robot)
save = context.comp(save_state, data_central,
id_agent, id_robot, agent_state)
return save
def jobs_parallel_learning_concurrent_reps(context, data_central, id_agent, id_robot, episodes,
n, max_reps, intermediate_reports=True,
final_report=True):
"""
This one cycles multiple times in the data.
"""
agents = []
contexts = ['learn%02dof%02d' % (i + 1, n) for i in range(n)]
for i, (c, progress) in enumerate(iterate_context_names(context, contexts)):
extra_dep = []
for id_episode in episodes:
er = c.get_resource(RM_EPISODE_READY, id_robot=id_robot, id_episode=id_episode)
extra_dep.append(er)
agent_i = c.subtask(LearnLogNoSaveHintRepeated,
boot_root=data_central.get_boot_root(),
agent=id_agent, robot=id_robot,
episodes=episodes,
parallel_hint=[i, n],
max_reps=max_reps,
intermediate_reports=True,
add_job_prefix='',
extra_dep=extra_dep)
if intermediate_reports:
report = c.comp(get_agentstate_report, agent_i, progress, job_id='report')
c.add_report(report, 'agent_report_partial',
id_agent=id_agent, id_robot=id_robot,
progress=progress)
agents.append(agent_i)
agent_state = jobs_merging_recursive(context, agents)
if final_report:
report = context.comp(get_agentstate_report, agent_state, 'all', job_id='report')
context.add_report(report, 'agent_report', id_agent=id_agent, id_robot=id_robot)
save = context.comp(save_state, data_central,
id_agent, id_robot, agent_state)
return save
