env1-dqn-CR.cfg 4.49 KB
Newer Older
Carel van Niekerk's avatar
Carel van Niekerk committed
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
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
# Error model: 0% error rate, addditive confscorer, uniform nbestgenerator
# User model: standard sampled params, sampled patience
# Masks: on

###### General parameters ######
[GENERAL]
domains = CamRestaurants
singledomain = True 
tracedialog = 0
seed = 07051991

[exec_config]
configdir = _benchmarkpolicies
logfiledir = _benchmarklogs
numtrainbatches = 4
traindialogsperbatch = 1000
numbatchtestdialogs =  500
trainsourceiteration = 0
numtestdialogs =  500
trainerrorrate = 0
testerrorrate  = 0
testeverybatch = True
#deleteprevpolicy = True

[logging]
usecolor = False
screen_level = results
file_level = results
file = auto

###### Environment parameters ######

[agent]
maxturns = 25

[usermodel]
usenewgoalscenarios = True
oldstylepatience = False
patience = 4,6
configfile = config/sampledUM.cfg

[errormodel]
nbestsize = 1
confusionmodel = RandomConfusions
nbestgeneratormodel = SampledNBestGenerator
confscorer = additive


[summaryacts]
maxinformslots = 5
informmask = True
requestmask = True
informcountaccepted = 4
byemask = True

###### Dialogue Manager parameters ######

## Comment the following lines if using any other policy (this uses handcrafted policy)##
# [policy]
# policydir = _benchmarkpolicies
# belieftype = focus
# useconfreq = False
# learning = True
# policytype = hdc
# startwithhello = False
# inpolicyfile = auto
# outpolicyfile = auto

## Uncomment for GP policy ##
#[policy]
#policydir = _benchmarkpolicies
#belieftype = focus
#useconfreq = False
#learning = True
#policytype = gp
#startwithhello = False
#inpolicyfile = auto
#outpolicyfile = auto
#
#[gppolicy]
#kernel = polysort
#
#[gpsarsa]
#random = False
#scale = 3

## Uncomment for DQN policy ##
[policy]
policydir = _benchmarkpolicies
belieftype = focus
useconfreq = False
learning = True
policytype = dqn
startwithhello = False
inpolicyfile = auto
outpolicyfile = auto

[dqnpolicy]
maxiter = 4000
gamma = 0.99
learning_rate = 0.001
tau = 0.02
replay_type = vanilla
minibatch_size = 64
capacity = 6000
exploration_type = e-greedy
episodeNum= 0.0
epsilon_start = 0.3
epsilon_end = 0.0
n_in = 268
features = ["discourseAct", "method", "requested", "full", "lastActionInformNone", "offerHappened", "inform_info"]
max_k = 5
learning_algorithm = dqn
architecture = vanilla
h1_size = 300
h2_size = 100
training_frequency = 2
n_samples = 1
stddev_var_mu = 0.01
stddev_var_logsigma = 0.01
mean_log_sigma = 0.000001
sigma_prior = 1.5
alpha =0.85
alpha_divergence =False
sigma_eps = 0.01
delta = 1.0
beta = 0.95
is_threshold = 5.0
train_iters_per_episode = 1

## Uncomment for A2C policy ##
#[policy]
#policydir = _benchmarkpolicies
#belieftype = focus
#useconfreq = False
#learning = True
#policytype = a2c
#startwithhello = False
#inpolicyfile = auto
#outpolicyfile = auto

#[dqnpolicy]
#maxiter = 4000
#gamma = 0.99
#learning_rate = 0.001
#tau = 0.02
#replay_type = vanilla
#minibatch_size = 64
#capacity = 1000
#exploration_type = e-greedy
#episodeNum= 0.0
#epsilon_start = 0.5
#epsilon_end = 0.0
#n_in = 268
#features = ["discourseAct", "method", "requested", "full", "lastActionInformNone", "offerHappened", "inform_info"]
#max_k = 5
#learning_algorithm = dqn
#architecture = vanilla
#h1_size = 200
#h2_size = 75
#training_frequency = 2
#n_samples = 1
#stddev_var_mu = 0.01
#stddev_var_logsigma = 0.01
#mean_log_sigma = 0.000001
#sigma_prior = 1.5
#alpha =0.85
#alpha_divergence =False
#sigma_eps = 0.01
#delta = 1.0
#beta = 0.95
#is_threshold = 5.0
#train_iters_per_episode = 1

## Uncomment for eNAC policy ##
#[policy]
#policydir = _benchmarkpolicies
#belieftype = focus
#useconfreq = False
#learning = True
#policytype = enac
#startwithhello = False
#inpolicyfile = auto
#outpolicyfile = auto

#[dqnpolicy]
#maxiter = 4000
#gamma = 0.99
#learning_rate = 0.001
#tau = 0.02
#replay_type = vanilla
#minibatch_size = 64
#capacity = 1000
#exploration_type = e-greedy
#episodeNum= 0.0
#epsilon_start = 0.3
#epsilon_end = 0.0
#n_in = 268
#features = ["discourseAct", "method", "requested", "full", "lastActionInformNone", "offerHappened", "inform_info"]
#max_k = 5
#learning_algorithm = dqn
#architecture = vanilla
#h1_size = 130
#h2_size = 50
#training_frequency = 2
#n_samples = 1
#stddev_var_mu = 0.01
#stddev_var_logsigma = 0.01
#mean_log_sigma = 0.000001
#sigma_prior = 1.5
#alpha =0.85
#alpha_divergence =False
#sigma_eps = 0.01
#delta = 1.0
#beta = 0.95
#is_threshold = 5.0
#train_iters_per_episode = 1

###### Evaluation parameters ######

[eval]
rewardvenuerecommended=0
penaliseallturns = True
wrongvenuepenalty = 0
notmentionedvaluepenalty = 0      
successmeasure = objective 
successreward = 20