def process(path):
clicked, lines_evaluated, lines_total = 0, 0, 0
with open(path) as inf:
for line in inf:
lines_total += 1
# Parsing the log line.
logline = line.strip().split()
chosen = int(logline.pop(7))
user_action = int(logline.pop(7))
time = int(logline[0])
user_features = [float(x) for x in logline[1:7]]
articles = [int(x) for x in logline[7:]]
# Getting the recommended article.
calculated = policy.reccomend(time, user_features, articles)
if not calculated in articles:
raise Exception("Article was not in the list.")
# Updating the policy.
if calculated == chosen:
policy.update(user_action)
clicked += user_action
lines_evaluated += 1
else:
policy.update(-1)
print "Evaluated %d/%d lines." % ( lines_evaluated, lines_total)
print "CTR=%f" % (float(clicked) / lines_evaluated)
def _run_once(self, verbose=True): policy = construct_policy(self.simu_params) iter_rewards = construct_iter_rewards(self.simu_params) # Pre-allocate to significantly increase performance rewards = [[]] * self.max_time policy_rewards = [0.0] * self.max_time reward_sums = [0.0] * policy.num_arms arm_choices = [] for t, rewards_t in enumerate(iter_rewards): if t >= self.max_time: break rewards[t] = rewards_t arm = policy.choose_arm() arm_choices.append(arm) reward = rewards_t[arm] policy_rewards[t] = reward policy.update(arm, reward) for i, r in enumerate(rewards_t): reward_sums[i] += r if verbose and t % 1000 == 0: print >>sys.stderr, "%s: distro %s ran %s time steps" % ( self.simu_params['name'], self.simu_params['distro_name'], t) return rewards, reward_sums, policy_rewards, arm_choices
def _run_once(self, verbose=True):
policy = construct_policy(self.simu_params)
iter_rewards = construct_iter_rewards(self.simu_params)
# Pre-allocate to significantly increase performance
rewards = [[]] * self.max_time
policy_rewards = [0.0] * self.max_time
reward_sums = [0.0] * policy.num_arms
arm_choices = []
for t, rewards_t in enumerate(iter_rewards):
if t >= self.max_time:
break
rewards[t] = rewards_t
arm = policy.choose_arm()
arm_choices.append(arm)
reward = rewards_t[arm]
policy_rewards[t] = reward
policy.update(arm, reward)
for i, r in enumerate(rewards_t):
reward_sums[i] += r
if verbose and t % 1000 == 0:
print >> sys.stderr, "%s: distro %s ran %s time steps" % (
self.simu_params['name'], self.simu_params['distro_name'],
t)
return rewards, reward_sums, policy_rewards, arm_choices
def main(args):
if (len(args) != 3):
raise Exception("Usage: ./evaluator.py articles_file log_file")
with file(args[1]) as inf:
articles = []
for line in inf:
features = line.strip().split(" ")
article = [int(features[0])]
article.extend(float(x) for x in features[1:])
articles.append(article)
policy.set_articles(articles)
score = 0
total_evaluated = 0
n_lines = 0
with file(args[2]) as inf:
for line in inf:
n_lines += 1
logline = line.strip().split()
chosen = int(logline.pop(7))
reward = int(logline.pop(7))
time = int(logline[0])
user_features = [float(x) for x in logline[1:7]]
articles = [int(x) for x in logline[7:]]
calculated = policy.reccomend(time, user_features, articles)
if calculated == chosen:
policy.update(reward)
score += reward
total_evaluated += 1
else:
policy.update(-1)
print "Evaluated %d/%d lines." % (total_evaluated, n_lines)
print "CTR=%f" % (float(score) / total_evaluated)
def process(path):
clicked, lines_evaluated, lines_total = 0, 0, 0
with file(sys.argv[2]) as inf:
for line in inf:
lines_total += 1
# Parsing the log line.
logline = line.strip().split()
chosen = int(logline.pop(7))
user_action = int(logline.pop(7))
time = int(logline[0])
user_features = [float(x) for x in logline[1:7]]
articles = [int(x) for x in logline[7:]]
# Getting the recommended article.
calculated = policy.reccomend(time, user_features, articles)
if calculated not in articles:
raise Exception("Article was not in the list.")
# Updating the policy.
if calculated == chosen:
policy.update(user_action)
clicked += user_action
lines_evaluated += 1
else:
policy.update(-1)
if lines_total % 1000 == 0:
print "Evaluated %d/%d lines." % (lines_evaluated, lines_total)
print "CTR = %f" % (float(clicked) / lines_evaluated)
print "Number of articles viewed %d" % len(policy.A.keys())
print "Evaluated %d/%d lines." % (lines_evaluated, lines_total)
print "CTR = %f" % (float(clicked) / lines_evaluated)
score = 0
total_evaluated = 0
n_lines = 0
with file(sys.argv[2]) as inf:
for line in inf:
n_lines += 1
logline = line.strip().split()
chosen = int(logline.pop(7))
reward = int(logline.pop(7))
time = int(logline[0])
user_features = [float(x) for x in logline[1:7]]
articles = [int(x) for x in logline[7:]]
calculated = policy.reccomend(time, user_features, articles)
if calculated == chosen:
policy.update(reward)
score += reward
total_evaluated += 1
else:
policy.update(-1)
print "Evaluated %d/%d lines." % (total_evaluated, n_lines)
print "CTR=%f" % (float(score) / total_evaluated)
pr.disable()
ps = pstats.Stats(pr, stream=open("profile.txt", "w"))
ps.sort_stats("cumtime")
ps.print_stats()
values = [items[1:] for items in articles]
dictionary = dict(zip(keys, values))
policy.set_articles(dictionary)
score = 0
total_evaluated = 0
n_lines = 0
with file(sys.argv[2]) as inf:
for line in inf:
n_lines += 1
logline = line.strip().split()
chosen = int(logline.pop(7))
reward = int(logline.pop(7))
time = int(logline[0])
user_features = [float(x) for x in logline[1:7]]
articles = [int(x) for x in logline[7:]]
calculated = policy.reccomend(time, user_features, articles)
if calculated == chosen:
policy.update(reward)
score += reward
total_evaluated += 1
else:
policy.update(-1)
print "Evaluated %d/%d lines." % (total_evaluated, n_lines)
print "CTR=%f" % (float(score) / total_evaluated)
