def map_lr(self):
lr_values = dict()
with self.lrs_lock:
for lr in self.lrs:
for key, value in lr.items():
add_dict(lr_values, key, [value * (1. - 2. * self.learnrate * self.regulizer) ** self.i])
return lr_values
def stat_dict_to_list(attr_stat_dict):
"""
:param attr_stat_dict: {attribute_value: {count: int, revenue: int, rate: int}}
:return: [attribute_value], {counts=[], revenue=[], rate=[]}
"""
attr_values = sorted(list(attr_stat_dict.keys()))
attr_stat_lists = dict()
for attribute_value in attr_values:
for stat_key in STATS:
add_dict(attr_stat_lists, stat_key, [attr_stat_dict[attribute_value][stat_key]])
return list(attr_values), attr_stat_lists
def stat_dict_to_list(attr_stat_dict):
"""
:param attr_stat_dict: {attribute_value: {count: int, revenue: int, rate: int}}
:return: [attribute_value], {counts=[], revenue=[], rate=[]}
"""
attr_values = sorted(list(attr_stat_dict.keys()))
attr_stat_lists = dict()
for attribute_value in attr_values:
for stat_key in STATS:
add_dict(attr_stat_lists, stat_key,
[attr_stat_dict[attribute_value][stat_key]])
return list(attr_values), attr_stat_lists
def lsr_update(t, y, x, weights, learnrate, default_value=None):
"""
Update values using logistic streaming regression
:param t: target
:param y: output
:param x: input
:param weights:
:return:
"""
if default_value is None:
default_value = lambda: None
error = t - y
add_dict(weights, "intercept", learnrate * error, default_value())
for key in x:
add_dict(weights, key, learnrate * error * x[key], default_value())
return weights
def average_param_reward(files):
average = {}
for file in files:
agent = np.load(os.path.join('agents', file)).item()
log = np.load(os.path.join('log', file)).item()
if 'reward' in log:
if 'thomp(' in file:
learnrate = agent['learnrate']
regulizer = agent['regulizer']
reward = log['reward']
create_dictionary(average, learnrate, {})
add_dict(average[learnrate], regulizer, [reward], [])
elif 'greedy' in file:
create_dictionary(average, 0.0, {})
add_dict(average[0.0], 'greedy', [log['reward']], [])
k2_length = 0
lengths = {}
for k1 in average:
k2_length = max(k2_length, len(average[k1]))
for k2 in average[k1]:
create_dictionary(lengths, k1, {})
lengths[k1][k2] = len(average[k1][k2])
average[k1][k2] = sum(average[k1][k2]) / len(average[k1][k2])
return average, lengths
def average_param_reward(files):
average = {}
for file in files:
agent = np.load(os.path.join('agents', file)).item()
log = np.load(os.path.join('log', file)).item()
if 'reward' in log:
if 'thomp(' in file:
learnrate = agent['learnrate']
regulizer = agent['regulizer']
reward = log['reward']
create_dictionary(average, learnrate, {})
add_dict(average[learnrate], regulizer, [reward], [])
elif 'greedy' in file:
create_dictionary(average, 0.0, {})
add_dict(average[0.0], 'greedy', [log['reward']], [])
k2_length = 0
lengths = {}
for k1 in average:
k2_length = max(k2_length, len(average[k1]))
for k2 in average[k1]:
create_dictionary(lengths, k1, {})
lengths[k1][k2] = len(average[k1][k2])
average[k1][k2] = sum(average[k1][k2]) / len(average[k1][k2])
return average, lengths
def run(self):
for run_id in self.run_idx:
for j in range(self.MAX_I + 1):
context = get_context(run_id, j)
action = self.agent.decide(context)
result = propose_page(run_id, j, **action)
self.agent.feedback(result)
add_dict(self.data, run_id, [{'context': context, 'action': action, 'result': result}], [])
success = "Success!" if result["effect"]["Success"] else ""
print(self.to_string(action, run_id, j, success))
add_dict(self.data, "reward", self.agent.cum_reward / self.agent.i)
add_dict(self.data, "cum_reward", self.agent.cum_reward)
self.save()
def run(self):
for run_id in self.run_idx:
for j in range(self.MAX_I + 1):
context = get_context(run_id, j)
action = self.agent.decide(context)
result = propose_page(run_id, j, **action)
self.agent.feedback(result)
add_dict(self.data, run_id, [{
'context': context,
'action': action,
'result': result
}], [])
success = "Success!" if result["effect"]["Success"] else ""
print(self.to_string(action, run_id, j, success))
add_dict(self.data, "reward", self.agent.cum_reward / self.agent.i)
add_dict(self.data, "cum_reward", self.agent.cum_reward)
self.save()
