-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathRL_Agents_Trainer.py
204 lines (131 loc) · 6.51 KB
/
RL_Agents_Trainer.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
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
import numpy as np
import copy
import random
from Agents import AgentML, AgentRandom
from My_Model import baseline_model
def test(test_data, true_labels, model, word_vector_size):
envML = AgentML()
for k in range(len(test_data)): # no. of games(sentences) in one epoch
sentence_embedding = test_data[k]
ml_intent_list = []
ml_reward_list = []
for l in range(len(sentence_embedding)): # no. of episodes(words) in one game(sentence)
ml_reward = -1
current_state = sentence_embedding[l].reshape(1, word_vector_size)
ml_dist = envML.act(current_state, model)
if np.argmax(true_labels[k]) == np.argmax(ml_dist):
ml_reward = 1
ml_intent_list.append(ml_dist)
ml_reward_list.append(ml_reward)
envML.dist_holder(ml_intent_list, ml_reward_list)
cum_intent_ml = []
abs_intent_ml = []
cum_test_wins = 0
abs_test_wins = 0
for k in range(len(envML.intent_dist)):
cum_intent_ml.append(np.mean(envML.intent_dist[k], axis=0))
for i in range(len(true_labels)):
if np.argmax(true_labels[i]) == np.argmax(cum_intent_ml[i]):
cum_test_wins += 1
# Here onwards
abs_dist = copy.deepcopy(envML.intent_dist)
# abs_dist = envML.intent_dist.copy()
for k in range(len(abs_dist)):
for l in range(len(abs_dist[k])):
temp = envML.intent_dist[k][l][0]
abs_dist[k][l] = ((temp == temp[np.argmax(temp)]) * 1)
for k in range(len(envML.intent_dist)):
abs_intent_ml.append(np.sum(abs_dist[k], axis=0))
for i in range(len(true_labels)):
if np.argmax(true_labels[i]) in np.argwhere(abs_intent_ml[i] == np.amax(abs_intent_ml[i])).flatten():
abs_test_wins += 1
abs_test_acc = (abs_test_wins/len(test_data)) * 100
cum_test_acc = (cum_test_wins/len(test_data)) * 100
return cum_test_acc, abs_test_acc
def train(train_data, labels, epochs, test_data, true_labels, word_vector_size, num_intents, hidden_size):
model = baseline_model(word_vector_size, num_intents, hidden_size)
inputs = []
targets = []
cum_train_acc_hist = []
cum_test_acc_hist = []
abs_train_acc_hist = []
abs_test_acc_hist = []
random_wins = []
# Overpopulating start
max_intent = max(np.unique(labels, axis=0, return_counts=True)[1])
for i in range(num_intents):
tup = np.unique(labels, axis=0, return_counts=True)
current_intent = tup[0][i]
current_count = tup[1][i]
diff_intent = max_intent - current_count
indices = [k for k, x in enumerate(labels) if sum(x == current_intent) == num_intents]
for j in range(diff_intent):
index = random.choice(indices)
train_data.append(train_data[index])
labels = labels.tolist()
labels.append(labels[index])
labels = np.array(labels)
# Overpopulating end
for e in range(epochs): # epochs
envRM = AgentRandom()
envML = AgentML()
for i in range(len(train_data)): # no. of games(sentences) in one epoch
sentence_embedding = train_data[i]
rm_intent_list = []
ml_intent_list = []
rm_reward_list = []
ml_reward_list = []
for j in range(len(sentence_embedding)): # no. of episodes(words) in one game(sentence)
rm_reward = -1
ml_reward = -1
current_state = sentence_embedding[j].reshape(1, word_vector_size)
rm_dist = envRM.act(num_intents)
ml_dist = envML.act(current_state, model)
if np.argmax(labels[i]) == np.argmax(rm_dist):
rm_reward = 1
if np.argmax(labels[i]) == np.argmax(ml_dist):
ml_reward = 1
rm_intent_list.append(rm_dist)
ml_intent_list.append(ml_dist)
rm_reward_list.append(rm_reward)
ml_reward_list.append(ml_reward)
envRM.dist_holder(rm_intent_list, rm_reward_list)
envML.dist_holder(ml_intent_list, ml_reward_list)
cum_reward_rm = []
cum_reward_ml = []
for i in range(len(envRM.reward_dist)):
cum_reward_rm.append(np.mean(envRM.reward_dist[i]))
for i in range(len(envML.reward_dist)):
cum_reward_ml.append(np.mean(envML.reward_dist[i]))
diff = np.array(cum_reward_rm) - np.array(cum_reward_ml)
rm_wins = (diff > 0)
random_wins.append(sum(rm_wins))
for i in range(len(rm_wins)):
if rm_wins[i] == True: # Agent Random performed better than Agent ML
for j in range(len(train_data[i])):
if envRM.reward_dist[i][j] == 1:
if (train_data[i][j]).tolist() in inputs:
input_index = inputs.index((train_data[i][j]).tolist())
target_index = np.argmax(envRM.intent_dist[i][j])
targets[input_index][target_index] += diff[i]
else:
inputs.append((train_data[i][j]).tolist())
index = np.argmax(envRM.intent_dist[i][j])
target = (np.zeros((num_intents,))).tolist()
target[index] = diff[i]
targets.append(target)
X_generated = copy.deepcopy(inputs)
y_generated = copy.deepcopy(targets)
for i in range(len(y_generated)):
y_generated[i] = [float(j) / max(y_generated[i]) for j in y_generated[i]]
model = baseline_model(word_vector_size, num_intents, hidden_size)
model.fit(np.array(X_generated), np.array(y_generated), epochs=500, batch_size=32)
cum_train_acc, abs_train_acc = test(train_data, labels, model, word_vector_size)
cum_test_acc, abs_test_acc = test(test_data, true_labels, model, word_vector_size)
cum_train_acc_hist.append(cum_train_acc)
abs_train_acc_hist.append(abs_train_acc)
cum_test_acc_hist.append(cum_test_acc)
abs_test_acc_hist.append(abs_test_acc)
print("Epoch " + str(e) + " Train " + str(cum_train_acc) + " , " + str(abs_train_acc))
print("Epoch " + str(e) + " Test " + str(cum_test_acc) + " , " + str(abs_test_acc))
return model, cum_train_acc_hist, abs_train_acc_hist, cum_test_acc_hist, abs_test_acc_hist, random_wins