Skip to content

Adds multiprocessing support for q_learning experiment. #20

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Adds multiprocessing support for q_learning experiment. #20

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
159 changes: 88 additions & 71 deletions assignment4/experiments/q_learner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,9 @@
import json
import os
import time
import multiprocessing
import numpy as np
from concurrent import futures

from .base import BaseExperiment, OUTPUT_DIRECTORY

Expand All @@ -18,7 +20,9 @@
class QLearnerExperiment(BaseExperiment):
def __init__(self, details, verbose=False):
self.max_episodes = 2000

self.threads = details.threads
if self.threads == -1:
self.threads = multiprocessing.cpu_count()
super(QLearnerExperiment, self).__init__(details, verbose)

def convergence_check_fn(self, solver, step_count):
Expand All @@ -42,79 +46,92 @@ def perform(self):
self.log("Searching Q in {} dimensions".format(dims))

runs = 1
params = []
for alpha in alphas:
for q_init in q_inits:
for epsilon in epsilons:
for epsilon_decay in epsilon_decays:
for discount_factor in discount_factors:
t = time.clock()
self.log("{}/{} Processing Q with alpha {}, q_init {}, epsilon {}, epsilon_decay {},"
" discount_factor {}".format(
runs, dims, alpha, q_init, epsilon, epsilon_decay, discount_factor
))

qs = solvers.QLearningSolver(self._details.env, self.max_episodes,
discount_factor=discount_factor,
alpha=alpha,
epsilon=epsilon, epsilon_decay=epsilon_decay,
q_init=q_init, verbose=self._verbose)

stats = self.run_solver_and_collect(qs, self.convergence_check_fn)

self.log("Took {} episodes".format(len(stats.steps)))
stats.to_csv('{}/Q/{}_{}_{}_{}_{}_{}.csv'.format(OUTPUT_DIRECTORY, self._details.env_name,
alpha, q_init, epsilon, epsilon_decay,
discount_factor))
stats.pickle_results('{}/Q/pkl/{}_{}_{}_{}_{}_{}_{}.pkl'.format(OUTPUT_DIRECTORY,
self._details.env_name,
alpha, q_init, epsilon,
epsilon_decay,
discount_factor,
'{}'), map_desc.shape,
step_size=self.max_episodes/20.0)
stats.plot_policies_on_map('{}/images/Q/{}_{}_{}_{}_{}_{}_{}.png'.format(OUTPUT_DIRECTORY,
self._details.env_name,
alpha, q_init, epsilon,
epsilon_decay,
discount_factor,
'{}_{}'),
map_desc, self._details.env.colors(),
self._details.env.directions(),
'Q-Learner', 'Episode', self._details,
step_size=self.max_episodes / 20.0,
only_last=True)

# We have extra stats about the episode we might want to look at later
episode_stats = qs.get_stats()
episode_stats.to_csv('{}/Q/{}_{}_{}_{}_{}_{}_episode.csv'.format(OUTPUT_DIRECTORY,
self._details.env_name,
alpha, q_init, epsilon,
epsilon_decay,
discount_factor))

optimal_policy_stats = self.run_policy_and_collect(qs, stats.optimal_policy)
self.log('{}'.format(optimal_policy_stats))
optimal_policy_stats.to_csv('{}/Q/{}_{}_{}_{}_{}_{}_optimal.csv'.format(OUTPUT_DIRECTORY,
self._details.env_name,
alpha, q_init, epsilon,
epsilon_decay,
discount_factor))

with open(grid_file_name, 'a') as f:
f.write('"{}",{},{},{},{},{},{},{}\n'.format(
json.dumps({
'alpha': alpha,
'q_init': q_init,
'epsilon': epsilon,
'epsilon_decay': epsilon_decay,
'discount_factor': discount_factor,
}).replace('"', '""'),
time.clock() - t,
len(optimal_policy_stats.rewards),
optimal_policy_stats.reward_mean,
optimal_policy_stats.reward_median,
optimal_policy_stats.reward_min,
optimal_policy_stats.reward_max,
optimal_policy_stats.reward_std,
))
params.append((alpha, q_init, epsilon, epsilon_decay, discount_factor, runs, dims, map_desc))
runs += 1
print("Totals runs {} to be processed on {} threads".format(runs-1, self.threads))

with futures.ProcessPoolExecutor(max_workers = self.threads) as pool:
for res in pool.map(self.run_q, params):
print("Completed run {}".format(res))

def run_q(self, params):
grid_file_name = '{}/Q/{}_grid.csv'.format(OUTPUT_DIRECTORY, self._details.env_name)
alpha, q_init, epsilon, epsilon_decay, discount_factor, runs, dims, map_desc = params
print("Processing run {}".format(runs))
t = time.clock()
self.log("{}/{} Processing Q with alpha {}, q_init {}, epsilon {}, epsilon_decay {},"
" discount_factor {}".format(
runs, dims, alpha, q_init, epsilon, epsilon_decay, discount_factor
))

qs = solvers.QLearningSolver(self._details.env, self.max_episodes,
discount_factor=discount_factor,
alpha=alpha,
epsilon=epsilon, epsilon_decay=epsilon_decay,
q_init=q_init, verbose=self._verbose, theta=0.001)

stats = self.run_solver_and_collect(qs, self.convergence_check_fn)

self.log("Took {} episodes".format(len(stats.steps)))
stats.to_csv('{}/Q/{}_{}_{}_{}_{}_{}.csv'.format(OUTPUT_DIRECTORY, self._details.env_name,
alpha, q_init, epsilon, epsilon_decay,
discount_factor))
stats.pickle_results('{}/Q/pkl/{}_{}_{}_{}_{}_{}_{}.pkl'.format(OUTPUT_DIRECTORY,
self._details.env_name,
alpha, q_init, epsilon,
epsilon_decay,
discount_factor,
'{}'), map_desc.shape,
step_size=self.max_episodes/20.0)
stats.plot_policies_on_map('{}/images/Q/{}_{}_{}_{}_{}_{}_{}.png'.format(OUTPUT_DIRECTORY,
self._details.env_name,
alpha, q_init, epsilon,
epsilon_decay,
discount_factor,
'{}_{}'),
map_desc, self._details.env.colors(),
self._details.env.directions(),
'Q-Learner', 'Episode', self._details,
step_size=self.max_episodes / 20.0,
only_last=True)

# We have extra stats about the episode we might want to look at later
episode_stats = qs.get_stats()
episode_stats.to_csv('{}/Q/{}_{}_{}_{}_{}_{}_episode.csv'.format(OUTPUT_DIRECTORY,
self._details.env_name,
alpha, q_init, epsilon,
epsilon_decay,
discount_factor))

optimal_policy_stats = self.run_policy_and_collect(qs, stats.optimal_policy)
self.log('{}'.format(optimal_policy_stats))
optimal_policy_stats.to_csv('{}/Q/{}_{}_{}_{}_{}_{}_optimal.csv'.format(OUTPUT_DIRECTORY,
self._details.env_name,
alpha, q_init, epsilon,
epsilon_decay,
discount_factor))

with open(grid_file_name, 'a') as f:
f.write('"{}",{},{},{},{},{},{},{}\n'.format(
json.dumps({
'alpha': alpha,
'q_init': q_init,
'epsilon': epsilon,
'epsilon_decay': epsilon_decay,
'discount_factor': discount_factor,
}).replace('"', '""'),
time.clock() - t,
len(optimal_policy_stats.rewards),
optimal_policy_stats.reward_mean,
optimal_policy_stats.reward_median,
optimal_policy_stats.reward_min,
optimal_policy_stats.reward_max,
optimal_policy_stats.reward_std,
))
return runs