-
Notifications
You must be signed in to change notification settings - Fork 0
/
OnlineAlgorithm.py
145 lines (92 loc) · 4.98 KB
/
OnlineAlgorithm.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
import numpy as np
from random import choices
class OnlineAlgorithm:
def __init__(self, num_experts, decimal_places, eta, reward_function):
self._reward_decimal_places = decimal_places
self._eta = eta
self._reward_function=reward_function
self._chosen_expert = None
self._weights = [1 for i in range(num_experts)]
self._weights_as_probabilities = [1 / num_experts for i in range(num_experts)]
@property
def chosen_expert(self):
return self._chosen_expert
@property
def weights_as_probabilities(self):
return self._weights_as_probabilities
def forecaster(self, experts, instance_id):
self._chosen_expert = choices(range(0, len(experts)), self._weights_as_probabilities)[0]
return experts[self._chosen_expert].get_model_prediction(instance_id)
def update(self, experts, current_iteration, instance_id):
pass
def _reward_(self, expert, instance_id):
if str.startswith(self._reward_function, "human"):
reward_value = expert.get_human_DA(instance_id)
if np.isnan(reward_value):
if self._reward_function == "human": #human-zero
reward_value = 0
elif self._reward_function == "human-avg":
reward_value = expert.avg_reward
elif self._reward_function == "human-comet":
reward_value = expert.get_comet_score(instance_id)
else:
reward_value = reward_value * 0.01
elif self._reward_function == "bleu":
reward_value = expert.get_bleu_score(instance_id)
reward_value = reward_value * 0.01
elif self._reward_function == "comet":
reward_value = expert.get_comet_score(instance_id)
else:
return # FIXME throw exception
if not(self._reward_decimal_places == None):
return np.round(reward_value, decimals=self._reward_decimal_places)
else:
return reward_value
################################################################################
class EWAF(OnlineAlgorithm):
def update(self, experts, current_iteration, instance_id):
num_experts = len(experts)
self._weights = []
self._weights_as_probabilities = []
for expert in experts:
print(expert.model_name)
expert_reward = self._reward_(expert, instance_id)
print("Current reward", expert_reward)
expert.cumulative_reward = expert.cumulative_reward + expert_reward
print("Total reward", expert.cumulative_reward)
expert.avg_reward = expert.cumulative_reward / max(1, (current_iteration - 1))
eta = np.sqrt((self._eta * np.log(num_experts)) / current_iteration)
expert.weight = np.exp(eta * expert.cumulative_reward)
self._weights.append(expert.weight)
total_weight = np.sum(self._weights)
for expert in experts:
expert.weight_as_probability = expert.weight / total_weight
self._weights_as_probabilities.append(expert.weight_as_probability)
def __str__(self):
return "EWAF | decimal places=" + str(self._reward_decimal_places) + " | eta=" + str(self._eta) + " | reward=" + self._reward_function
################################################################################
class EXP3(OnlineAlgorithm):
def update(self, arms, current_iteration, instance_id):
num_arms = len(arms)
arm_chosen = arms[self._chosen_expert]
arm_reward = self._reward_(arm_chosen, instance_id)
print("Current reward", arm_reward)
arm_chosen.cumulative_reward = arm_chosen.cumulative_reward + (arm_reward / arm_chosen.weight_as_probability)
arm_chosen.avg_reward = arm_chosen.cumulative_reward / max(1, (current_iteration - 1))
print("Total reward", arm_chosen.cumulative_reward)
eta = np.sqrt((2 * np.log(num_arms)) / (current_iteration * num_arms))
arm_chosen.weight = np.exp(eta * arm_chosen.cumulative_reward)
self._weights[self._chosen_expert] = arm_chosen.weight
total_weight = np.sum(self._weights)
arm_chosen.weight_as_probability = arm_chosen.weight / total_weight
self._weights_as_probabilities = [ w / total_weight for w in self._weights]
def __str__(self):
return "EXP3 | decimal places=" + str(self._reward_decimal_places) + " | eta=" + str(self._eta) + " | reward=" + self._reward_function
################################################################################
def init_online_algorithm(algorithm, num_experts, decimal_places=None, eta_value=8, reward_function="human"):
if algorithm == "EWAF":
return EWAF(num_experts, decimal_places, eta_value, reward_function)
elif algorithm == "EXP3":
return EXP3(num_experts, decimal_places, eta_value, reward_function)
else:
return # FIXME throw exception