def exp_run_industrial_datasets(iteration,
datasets=['iofrol',
'paintcontrol']): #, 'gsdtsr']):
ags = [
#lambda: (
# agents.TableauAgent(histlen=retecs.DEFAULT_HISTORY_LENGTH, learning_rate=retecs.DEFAULT_LEARNING_RATE,
# state_size=retecs.DEFAULT_STATE_SIZE,
# action_size=retecs.DEFAULT_NO_ACTIONS, epsilon=retecs.DEFAULT_EPSILON),
# retecs.preprocess_discrete, reward.timerank),
#lambda: (agents.NetworkAgent(histlen=retecs.DEFAULT_HISTORY_LENGTH, state_size=retecs.DEFAULT_STATE_SIZE,
# action_size=1,
# hidden_size=retecs.DEFAULT_NO_HIDDEN_NODES), retecs.preprocess_continuous,
# reward.tcfail),
lambda: (agents.LSTMAgent(histlen=retecs.DEFAULT_HISTORY_LENGTH,
state_size=retecs.DEFAULT_STATE_SIZE,
action_size=1,
hidden_size=retecs.DEFAULT_NO_HIDDEN_NODES),
retecs.preprocess_continuous, reward.tcfail)
]
reward_funs = {
'failcount': reward.failcount,
'timerank': reward.timerank,
'tcfail': reward.tcfail
}
avg_napfd = []
for i, get_agent in enumerate(ags):
for sc in datasets:
for (reward_name, reward_fun) in reward_funs.items():
agent, preprocessor, _ = get_agent()
file_appendix = 'rq_%s_%s_%s_%d' % (agent.name, sc,
reward_name, iteration)
scenario = get_scenario(sc)
rl_learning = retecs.PrioLearning(
agent=agent,
scenario_provider=scenario,
reward_function=reward_fun,
preprocess_function=preprocessor,
file_prefix=file_appendix,
dump_interval=100, #100
validation_interval=0,
output_dir=DATA_DIR,
output_csv_dir=DATA_DIR_CSV)
res = rl_learning.train(no_scenarios=CI_CYCLES,
print_log=False,
plot_graphs=False,
save_graphs=False,
collect_comparison=False) #(i == 0))
avg_napfd.append(res)
return avg_napfd
def exp_run_industrial_datasets(iteration,
datasets=['paintcontrol', 'iofrol', 'gsdtsr']):
ags = [
lambda: (XCS_ER(2000, list(range(0, 45)), 6), retecs.preprocess_xcs,
reward.tcfail), lambda: (XCS(2000, list(range(
0, 45)), 6), retecs.preprocess_xcs, reward.tcfail),
lambda:
(agents.NetworkAgent(histlen=retecs.DEFAULT_HISTORY_LENGTH,
state_size=retecs.DEFAULT_STATE_SIZE,
action_size=1,
hidden_size=retecs.DEFAULT_NO_HIDDEN_NODES),
retecs.preprocess_continuous, reward.tcfail)
]
reward_funs = {
'failcount': reward.failcount,
'timerank': reward.timerank,
'tcfail': reward.tcfail
}
avg_napfd = []
for i, get_agent in enumerate(ags):
for sc in datasets:
for (reward_name, reward_fun) in reward_funs.items():
agent, preprocessor, _ = get_agent()
file_appendix = 'rq_%s_%s_%s_%d' % (agent.name, sc,
reward_name, iteration)
scenario = get_scenario(sc)
rl_learning = retecs.PrioLearning(
agent=agent,
scenario_provider=scenario,
reward_function=reward_fun,
preprocess_function=preprocessor,
file_prefix=file_appendix,
dump_interval=100,
validation_interval=0,
output_dir=DATA_DIR)
res = rl_learning.train(no_scenarios=CI_CYCLES,
print_log=False,
plot_graphs=False,
save_graphs=False,
collect_comparison=(i == 0))
avg_napfd.append(res)
return avg_napfd