-
Notifications
You must be signed in to change notification settings - Fork 2
Expand file tree
/
Copy pathMDP.cpp
More file actions
417 lines (384 loc) · 12.9 KB
/
MDP.cpp
File metadata and controls
417 lines (384 loc) · 12.9 KB
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
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
#include <stdio.h>
#include "MDP.h"
/*Compute a Q-value for a given state and its action
Input: state variable s; action (go to East/North/West/South), the corresponding interger value for each action is defined in the "MDP.h" Macros definition on the top
Output: you need to update the corresponding q_values of the input state variable s
*/
void MDP::computeQValue(State &s, const int action)
{
/*store each direction value after computation*/
float q1;
float q2;
float q3;
/*each direction coresponding Q value*/
if (action == ACTION_EAST)
{
/*face the wall situation*/
if (s.location.x == 0 && s.location.y == 1)
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[0][0].state_value);
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[2][0].state_value);
/*compute convergence*/
if (cur_convergence < abs(s.q_values[0] - (q2 + q1 + q3)))
cur_convergence = abs(s.q_values[0] - (q2 + q1 + q3));
s.q_values[0] = q2 + q1 + q3;
}
else
{
if (s.location.y == 2)
{
/*go south*/
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
if ((s.location.y == 0) && (s.location.x == 1))
{
/*face the wall situation*/
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
/*go south*/
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y + 1][s.location.x].state_value);
}
}
if (s.location.x == 3)
{
/*face the wall situation*/
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
if (s.location.y == 1 && s.location.x == 0)
{
/*face the wall situation with EAST action*/
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
/*take EAST action*/
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x + 1].state_value);
}
}
if (s.location.y == 0)
{
/*go north*/
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
if (s.location.y == 2 && s.location.x == 1)
{
/*face the wall situation*/
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
/*go north*/
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y - 1][s.location.x].state_value);
}
}
/*compute convergence*/
if (cur_convergence < abs(s.q_values[0] - (q2 + q1 + q3)))
cur_convergence = abs(s.q_values[0] - (q2 + q1 + q3));
s.q_values[0] = q2 + q1 + q3;
}
}
else if (action == ACTION_WEST)
{
/*face the wall situation*/
if (s.location.x == 2 && s.location.y == 1)
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y - 1][s.location.x].state_value);
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y + 1][s.location.x].state_value);
/*compute convergence*/
if (cur_convergence < abs(s.q_values[2] - (q2 + q1 + q3)))
cur_convergence = abs(s.q_values[2] - (q2 + q1 + q3));
s.q_values[2] = q2 + q1 + q3;
}
else
{
/*go south*/
if (s.location.y == 2)
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
/*face the wall situation*/
if (s.location.y == 0 && s.location.x == 1)
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y + 1][s.location.x].state_value);
}
}
/*go west*/
if (s.location.x == 0)
{
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
if (s.location.y == 1 && s.location.x == 2)
{
/*face the wall situation with take WEST action*/
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x - 1].state_value);
}
}
/*go north*/
if (s.location.y == 0)
{
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
if (s.location.y == 2 && s.location.x == 1)
{
/*face the wall situation*/
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y - 1][s.location.x].state_value);
}
}
/*computer convergence*/
if (cur_convergence < abs(s.q_values[2] - (q2 + q1 + q3)))
cur_convergence = abs(s.q_values[2] - (q2 + q1 + q3));
s.q_values[2] = q2 + q1 + q3;
}
}
else if (action == ACTION_NORTH)
{
/*face the wall situation*/
if (s.location.x == 1 && s.location.y == 2)
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y ][s.location.x- 1].state_value);
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y ][s.location.x+ 1].state_value);
/*computer convergence*/
if (cur_convergence < abs(s.q_values[3] - (q2 + q1 + q3)))
cur_convergence = abs(s.q_values[3] - (q2 + q1 + q3));
s.q_values[3] = q2 + q1 + q3;
}
else
{
/*go west*/
if (s.location.x == 0)
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
/*face the wall situation with take WEST action*/
if (s.location.y == 1 && s.location.x == 2)
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x - 1].state_value);
}
}
/*go north*/
if (s.location.y == 0)
{
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
if (s.location.y == 2 && s.location.x == 1)
{
/*face the wall situation*/
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y - 1][s.location.x].state_value);
}
}
/*face the wall go EAST*/
if (s.location.x == 3)
{
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
if (s.location.y == 1 && s.location.x == 0)
{
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x + 1].state_value);
}
}
/*computer convergence*/
if (cur_convergence < abs(s.q_values[3] - (q2 + q1 + q3)))
cur_convergence = abs(s.q_values[3] - (q2 + q1 + q3));
s.q_values[3] = q2 + q1 + q3;
}
}
else
{
/*go SOUTH with facing wall situation*/
if (s.location.x == 1 && s.location.y == 0)
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x - 1].state_value);
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x + 1].state_value);
/*computer convergence*/
if (cur_convergence < abs(s.q_values[1] - (q2 + q1 + q3)))
cur_convergence = abs(s.q_values[1] - (q2 + q1 + q3));
s.q_values[1] = q2 + q1 + q3;
}
else
{
/*go WEST*/
if (s.location.x == 0)
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
if (s.location.y == 1 && s.location.x == 2)
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
q1 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x - 1].state_value);
}
}
/*go SOUTH*/
if (s.location.y == 2)
{
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
if (s.location.y == 0 && s.location.x == 1)
{
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
q2 = TRANSITION_SUCCEED*(ACTION_REWARD + GAMMA*states[s.location.y + 1][s.location.x].state_value);
}
}
/*go EAST*/
if (s.location.x == 3)
{
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
if (s.location.y == 1 && s.location.x == 0)
{
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x].state_value);
}
else
{
q3 = TRANSITION_FAIL*0.5*(ACTION_REWARD + GAMMA*states[s.location.y][s.location.x + 1].state_value);
}
}
/*computer convergence*/
if (cur_convergence < abs(s.q_values[1] - (q2 + q1 + q3)))
cur_convergence = abs(s.q_values[1] - (q2 + q1 + q3));
s.q_values[1] = q2 + q1 + q3;
}
}
}
/*There is no return value, actually you need to use the computing result to update the state values of all the states defined as data member "State states[3][4]". Of course, you don't need to update the wall state: states[1][1], the diamond state: states[0][3], and pitfall state: states[1][3] */
void MDP::valueIteration()
{
/*first set the variable to be 0*/
cur_convergence = 0;
for (int i = 0; i < 3; i++)
{
for (int j = 0; j < 4; j++)
{
if ((i == 1 && j == 1) || (i == 0 && j == 3) || (i == 1&&j == 3))
continue;
else
{
/*computer every single state with forth actions*/
computeQValue(states[i][j], ACTION_EAST);
computeQValue(states[i][j], ACTION_SOUTH);
computeQValue(states[i][j], ACTION_WEST);
computeQValue(states[i][j], ACTION_NORTH);
}
}
}
/*set value for all the states by choosing MAX(Qvalue)*/
for (int i = 0; i < 3; i++)
{
for (int j = 0; j < 4; j++)
{
if ((i == 1 && j == 1) || (i == 0 && j == 3) || (i == 1 && j == 3))
continue;
else
{
states[i][j].state_value = max(max(max(states[i][j].q_values[0], states[i][j].q_values[1]), states[i][j].q_values[2]), states[i][j].q_values[3]);
}
}
}
}
MDP::MDP()
{
/*Initialize all the state with 0.0 state_value and 0.0 Q_values*/
for(int y = 0; y < 3; y++)
{
for(int x = 0; x < 4; x++)
{
states[y][x].location.x = x; //specify the location for this state
states[y][x].location.y = y;
states[y][x].state_value = 0.0; //define the state value
states[y][x].q_values[0] = 0.0; //define the Q value
states[y][x].q_values[1] = 0.0;
states[y][x].q_values[2] = 0.0;
states[y][x].q_values[3] = 0.0;
}
}
/* Reset the values for the two special states: diamonds (0, 3), pitfall (1, 3). Actually there are no Q-values for these two states as these two states represents the final state of the game. Similarly, for the wall (1, 1), it does not have any state value or Q values. So make sure not to update these three states during your program*/
states[0][3].state_value = 1.0;
states[1][3].state_value = -1.0;
/*Set the current convergence to a big number initially*/
cur_convergence = 100; //the reason this value is set to a big value is to ensure
}
MDP::~MDP()
{
}
/*Reset the current computed state and Q values*/
void MDP::resetData()
{
/*Initialize all the state with 0.0 state_value and 0.0 Q_values*/
for(int y = 0; y < 3; y++)
{
for(int x = 0; x < 4; x++)
{
states[y][x].location.x = x; //specify the location for this state
states[y][x].location.y = y;
states[y][x].state_value = 0.0; //define the state value
states[y][x].q_values[0] = 0.0; //define the Q value
states[y][x].q_values[1] = 0.0;
states[y][x].q_values[2] = 0.0;
states[y][x].q_values[3] = 0.0;
}
}
/* Reset the values for the two special states: diamonds (0, 3), pitfall (1, 3). Actually there are no Q-values for these two states as these two states represents the final state of the game. Similarly, for the wall (1, 1), it does not have any state value or Q values. So make sure not to update these three states during your program*/
states[0][3].state_value = 1.0;
states[1][3].state_value = -1.0;
/*Set the current convergence to a big number initially*/
cur_convergence = 0; //the reason this value is set to a big value is to ensure
}