Revision: miles@gnu.org--gnu-2005/emacs--unicode--0--patch-68
[bpt/emacs.git] / lisp / play / landmark.el
1 ;;; landmark.el --- neural-network robot that learns landmarks
2
3 ;; Copyright (c) 1996, 1997, 2000, 2003, 2005 Free Software Foundation, Inc.
4
5 ;; Author: Terrence Brannon (was: <brannon@rana.usc.edu>)
6 ;; Created: December 16, 1996 - first release to usenet
7 ;; Keywords: gomoku, neural network, adaptive search, chemotaxis
8
9 ;;;_* Usage
10 ;;; Just type
11 ;;; M-x eval-current-buffer
12 ;;; M-x lm-test-run
13
14
15 ;; This file is part of GNU Emacs.
16
17 ;; GNU Emacs is free software; you can redistribute it and/or modify
18 ;; it under the terms of the GNU General Public License as published by
19 ;; the Free Software Foundation; either version 2, or (at your option)
20 ;; any later version.
21
22 ;; GNU Emacs is distributed in the hope that it will be useful,
23 ;; but WITHOUT ANY WARRANTY; without even the implied warranty of
24 ;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
25 ;; GNU General Public License for more details.
26
27 ;; You should have received a copy of the GNU General Public License
28 ;; along with GNU Emacs; see the file COPYING. If not, write to the
29 ;; Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
30 ;; Boston, MA 02110-1301, USA.
31
32
33 ;;; Commentary:
34 ;;; Lm is a relatively non-participatory game in which a robot
35 ;;; attempts to maneuver towards a tree at the center of the window
36 ;;; based on unique olfactory cues from each of the 4 directions. If
37 ;;; the smell of the tree increases, then the weights in the robot's
38 ;;; brain are adjusted to encourage this odor-driven behavior in the
39 ;;; future. If the smell of the tree decreases, the robots weights are
40 ;;; adjusted to discourage a correct move.
41
42 ;;; In laymen's terms, the search space is initially flat. The point
43 ;;; of training is to "turn up the edges of the search space" so that
44 ;;; the robot rolls toward the center.
45
46 ;;; Further, do not become alarmed if the robot appears to oscillate
47 ;;; back and forth between two or a few positions. This simply means
48 ;;; it is currently caught in a local minimum and is doing its best to
49 ;;; work its way out.
50
51 ;;; The version of this program as described has a small problem. a
52 ;;; move in a net direction can produce gross credit assignment. for
53 ;;; example, if moving south will produce positive payoff, then, if in
54 ;;; a single move, one moves east,west and south, then both east and
55 ;;; west will be improved when they shouldn't
56
57 ;;; Many thanks to Yuri Pryadkin (yuri@rana.usc.edu) for this
58 ;;; concise problem description.
59
60 ;;;_* Require
61 (eval-when-compile (require 'cl))
62
63 ;;;_* From Gomoku
64
65 ;;; Code:
66
67 (defgroup lm nil
68 "Neural-network robot that learns landmarks."
69 :prefix "lm-"
70 :group 'games)
71
72 ;;;_ + THE BOARD.
73
74 ;; The board is a rectangular grid. We code empty squares with 0, X's with 1
75 ;; and O's with 6. The rectangle is recorded in a one dimensional vector
76 ;; containing padding squares (coded with -1). These squares allow us to
77 ;; detect when we are trying to move out of the board. We denote a square by
78 ;; its (X,Y) coords, or by the INDEX corresponding to them in the vector. The
79 ;; leftmost topmost square has coords (1,1) and index lm-board-width + 2.
80 ;; Similarly, vectors between squares may be given by two DX, DY coords or by
81 ;; one DEPL (the difference between indexes).
82
83 (defvar lm-board-width nil
84 "Number of columns on the Lm board.")
85 (defvar lm-board-height nil
86 "Number of lines on the Lm board.")
87
88 (defvar lm-board nil
89 "Vector recording the actual state of the Lm board.")
90
91 (defvar lm-vector-length nil
92 "Length of lm-board vector.")
93
94 (defvar lm-draw-limit nil
95 ;; This is usually set to 70% of the number of squares.
96 "After how many moves will Emacs offer a draw?")
97
98 (defvar lm-cx 0
99 "This is the x coordinate of the center of the board.")
100
101 (defvar lm-cy 0
102 "This is the y coordinate of the center of the board.")
103
104 (defvar lm-m 0
105 "This is the x dimension of the playing board.")
106
107 (defvar lm-n 0
108 "This is the y dimension of the playing board.")
109
110
111 (defun lm-xy-to-index (x y)
112 "Translate X, Y cartesian coords into the corresponding board index."
113 (+ (* y lm-board-width) x y))
114
115 (defun lm-index-to-x (index)
116 "Return corresponding x-coord of board INDEX."
117 (% index (1+ lm-board-width)))
118
119 (defun lm-index-to-y (index)
120 "Return corresponding y-coord of board INDEX."
121 (/ index (1+ lm-board-width)))
122
123 (defun lm-init-board ()
124 "Create the lm-board vector and fill it with initial values."
125 (setq lm-board (make-vector lm-vector-length 0))
126 ;; Every square is 0 (i.e. empty) except padding squares:
127 (let ((i 0) (ii (1- lm-vector-length)))
128 (while (<= i lm-board-width) ; The squares in [0..width] and in
129 (aset lm-board i -1) ; [length - width - 1..length - 1]
130 (aset lm-board ii -1) ; are padding squares.
131 (setq i (1+ i)
132 ii (1- ii))))
133 (let ((i 0))
134 (while (< i lm-vector-length)
135 (aset lm-board i -1) ; and also all k*(width+1)
136 (setq i (+ i lm-board-width 1)))))
137
138 ;;;_ + DISPLAYING THE BOARD.
139
140 ;; You may change these values if you have a small screen or if the squares
141 ;; look rectangular, but spacings SHOULD be at least 2 (MUST BE at least 1).
142
143 (defconst lm-square-width 2
144 "*Horizontal spacing between squares on the Lm board.")
145
146 (defconst lm-square-height 1
147 "*Vertical spacing between squares on the Lm board.")
148
149 (defconst lm-x-offset 3
150 "*Number of columns between the Lm board and the side of the window.")
151
152 (defconst lm-y-offset 1
153 "*Number of lines between the Lm board and the top of the window.")
154
155
156 ;;;_ + LM MODE AND KEYMAP.
157
158 (defcustom lm-mode-hook nil
159 "If non-nil, its value is called on entry to Lm mode."
160 :type 'hook
161 :group 'lm)
162
163 (defvar lm-mode-map nil
164 "Local keymap to use in Lm mode.")
165
166 (if lm-mode-map nil
167 (setq lm-mode-map (make-sparse-keymap))
168
169 ;; Key bindings for cursor motion.
170 (define-key lm-mode-map "y" 'lm-move-nw) ; y
171 (define-key lm-mode-map "u" 'lm-move-ne) ; u
172 (define-key lm-mode-map "b" 'lm-move-sw) ; b
173 (define-key lm-mode-map "n" 'lm-move-se) ; n
174 (define-key lm-mode-map "h" 'backward-char) ; h
175 (define-key lm-mode-map "l" 'forward-char) ; l
176 (define-key lm-mode-map "j" 'lm-move-down) ; j
177 (define-key lm-mode-map "k" 'lm-move-up) ; k
178
179 (define-key lm-mode-map [kp-7] 'lm-move-nw)
180 (define-key lm-mode-map [kp-9] 'lm-move-ne)
181 (define-key lm-mode-map [kp-1] 'lm-move-sw)
182 (define-key lm-mode-map [kp-3] 'lm-move-se)
183 (define-key lm-mode-map [kp-4] 'backward-char)
184 (define-key lm-mode-map [kp-6] 'forward-char)
185 (define-key lm-mode-map [kp-2] 'lm-move-down)
186 (define-key lm-mode-map [kp-8] 'lm-move-up)
187
188 (define-key lm-mode-map "\C-n" 'lm-move-down) ; C-n
189 (define-key lm-mode-map "\C-p" 'lm-move-up) ; C-p
190
191 ;; Key bindings for entering Human moves.
192 (define-key lm-mode-map "X" 'lm-human-plays) ; X
193 (define-key lm-mode-map "x" 'lm-human-plays) ; x
194
195 (define-key lm-mode-map " " 'lm-start-robot) ; SPC
196 (define-key lm-mode-map [down-mouse-1] 'lm-start-robot)
197 (define-key lm-mode-map [drag-mouse-1] 'lm-click)
198 (define-key lm-mode-map [mouse-1] 'lm-click)
199 (define-key lm-mode-map [down-mouse-2] 'lm-click)
200 (define-key lm-mode-map [mouse-2] 'lm-mouse-play)
201 (define-key lm-mode-map [drag-mouse-2] 'lm-mouse-play)
202
203 (define-key lm-mode-map [remap previous-line] 'lm-move-up)
204 (define-key lm-mode-map [remap next-line] 'lm-move-down)
205 (define-key lm-mode-map [remap beginning-of-line] 'lm-beginning-of-line)
206 (define-key lm-mode-map [remap end-of-line] 'lm-end-of-line)
207 (define-key lm-mode-map [remap undo] 'lm-human-takes-back)
208 (define-key lm-mode-map [remap advertised-undo] 'lm-human-takes-back))
209
210 (defvar lm-emacs-won ()
211 "*For making font-lock use the winner's face for the line.")
212
213 (defvar lm-font-lock-face-O
214 (if (display-color-p)
215 (list (facemenu-get-face 'fg:red) 'bold))
216 "*Face to use for Emacs' O.")
217
218 (defvar lm-font-lock-face-X
219 (if (display-color-p)
220 (list (facemenu-get-face 'fg:green) 'bold))
221 "*Face to use for your X.")
222
223 (defvar lm-font-lock-keywords
224 '(("O" . lm-font-lock-face-O)
225 ("X" . lm-font-lock-face-X)
226 ("[-|/\\]" 0 (if lm-emacs-won
227 lm-font-lock-face-O
228 lm-font-lock-face-X)))
229 "*Font lock rules for Lm.")
230
231 (put 'lm-mode 'front-sticky
232 (put 'lm-mode 'rear-nonsticky '(intangible)))
233 (put 'lm-mode 'intangible 1)
234 ;; This one is for when they set view-read-only to t: Landmark cannot
235 ;; allow View Mode to be activated in its buffer.
236 (put 'lm-mode 'mode-class 'special)
237
238 (defun lm-mode ()
239 "Major mode for playing Lm against Emacs.
240 You and Emacs play in turn by marking a free square. You mark it with X
241 and Emacs marks it with O. The winner is the first to get five contiguous
242 marks horizontally, vertically or in diagonal.
243
244 You play by moving the cursor over the square you choose and hitting \\[lm-human-plays].
245
246 Other useful commands:
247 \\{lm-mode-map}
248 Entry to this mode calls the value of `lm-mode-hook' if that value
249 is non-nil. One interesting value is `turn-on-font-lock'."
250 (interactive)
251 (kill-all-local-variables)
252 (setq major-mode 'lm-mode
253 mode-name "Lm")
254 (lm-display-statistics)
255 (use-local-map lm-mode-map)
256 (make-local-variable 'font-lock-defaults)
257 (setq font-lock-defaults '(lm-font-lock-keywords t))
258 (toggle-read-only t)
259 (run-mode-hooks 'lm-mode-hook))
260
261
262 ;;;_ + THE SCORE TABLE.
263
264
265 ;; Every (free) square has a score associated to it, recorded in the
266 ;; LM-SCORE-TABLE vector. The program always plays in the square having
267 ;; the highest score.
268
269 (defvar lm-score-table nil
270 "Vector recording the actual score of the free squares.")
271
272
273 ;; The key point point about the algorithm is that, rather than considering
274 ;; the board as just a set of squares, we prefer to see it as a "space" of
275 ;; internested 5-tuples of contiguous squares (called qtuples).
276 ;;
277 ;; The aim of the program is to fill one qtuple with its O's while preventing
278 ;; you from filling another one with your X's. To that effect, it computes a
279 ;; score for every qtuple, with better qtuples having better scores. Of
280 ;; course, the score of a qtuple (taken in isolation) is just determined by
281 ;; its contents as a set, i.e. not considering the order of its elements. The
282 ;; highest score is given to the "OOOO" qtuples because playing in such a
283 ;; qtuple is winning the game. Just after this comes the "XXXX" qtuple because
284 ;; not playing in it is just loosing the game, and so on. Note that a
285 ;; "polluted" qtuple, i.e. one containing at least one X and at least one O,
286 ;; has score zero because there is no more any point in playing in it, from
287 ;; both an attacking and a defending point of view.
288 ;;
289 ;; Given the score of every qtuple, the score of a given free square on the
290 ;; board is just the sum of the scores of all the qtuples to which it belongs,
291 ;; because playing in that square is playing in all its containing qtuples at
292 ;; once. And it is that function which takes into account the internesting of
293 ;; the qtuples.
294 ;;
295 ;; This algorithm is rather simple but anyway it gives a not so dumb level of
296 ;; play. It easily extends to "n-dimensional Lm", where a win should not
297 ;; be obtained with as few as 5 contiguous marks: 6 or 7 (depending on n !)
298 ;; should be preferred.
299
300
301 ;; Here are the scores of the nine "non-polluted" configurations. Tuning
302 ;; these values will change (hopefully improve) the strength of the program
303 ;; and may change its style (rather aggressive here).
304
305 (defconst nil-score 7 "Score of an empty qtuple.")
306 (defconst Xscore 15 "Score of a qtuple containing one X.")
307 (defconst XXscore 400 "Score of a qtuple containing two X's.")
308 (defconst XXXscore 1800 "Score of a qtuple containing three X's.")
309 (defconst XXXXscore 100000 "Score of a qtuple containing four X's.")
310 (defconst Oscore 35 "Score of a qtuple containing one O.")
311 (defconst OOscore 800 "Score of a qtuple containing two O's.")
312 (defconst OOOscore 15000 "Score of a qtuple containing three O's.")
313 (defconst OOOOscore 800000 "Score of a qtuple containing four O's.")
314
315 ;; These values are not just random: if, given the following situation:
316 ;;
317 ;; . . . . . . . O .
318 ;; . X X a . . . X .
319 ;; . . . X . . . X .
320 ;; . . . X . . . X .
321 ;; . . . . . . . b .
322 ;;
323 ;; you want Emacs to play in "a" and not in "b", then the parameters must
324 ;; satisfy the inequality:
325 ;;
326 ;; 6 * XXscore > XXXscore + XXscore
327 ;;
328 ;; because "a" mainly belongs to six "XX" qtuples (the others are less
329 ;; important) while "b" belongs to one "XXX" and one "XX" qtuples. Other
330 ;; conditions are required to obtain sensible moves, but the previous example
331 ;; should illustrate the point. If you manage to improve on these values,
332 ;; please send me a note. Thanks.
333
334
335 ;; As we chose values 0, 1 and 6 to denote empty, X and O squares, the
336 ;; contents of a qtuple are uniquely determined by the sum of its elements and
337 ;; we just have to set up a translation table.
338
339 (defconst lm-score-trans-table
340 (vector nil-score Xscore XXscore XXXscore XXXXscore 0
341 Oscore 0 0 0 0 0
342 OOscore 0 0 0 0 0
343 OOOscore 0 0 0 0 0
344 OOOOscore 0 0 0 0 0
345 0)
346 "Vector associating qtuple contents to their score.")
347
348
349 ;; If you do not modify drastically the previous constants, the only way for a
350 ;; square to have a score higher than OOOOscore is to belong to a "OOOO"
351 ;; qtuple, thus to be a winning move. Similarly, the only way for a square to
352 ;; have a score between XXXXscore and OOOOscore is to belong to a "XXXX"
353 ;; qtuple. We may use these considerations to detect when a given move is
354 ;; winning or loosing.
355
356 (defconst lm-winning-threshold OOOOscore
357 "Threshold score beyond which an Emacs move is winning.")
358
359 (defconst lm-loosing-threshold XXXXscore
360 "Threshold score beyond which a human move is winning.")
361
362
363 (defun lm-strongest-square ()
364 "Compute index of free square with highest score, or nil if none."
365 ;; We just have to loop other all squares. However there are two problems:
366 ;; 1/ The SCORE-TABLE only gives correct scores to free squares. To speed
367 ;; up future searches, we set the score of padding or occupied squares
368 ;; to -1 whenever we meet them.
369 ;; 2/ We want to choose randomly between equally good moves.
370 (let ((score-max 0)
371 (count 0) ; Number of equally good moves
372 (square (lm-xy-to-index 1 1)) ; First square
373 (end (lm-xy-to-index lm-board-width lm-board-height))
374 best-square score)
375 (while (<= square end)
376 (cond
377 ;; If score is lower (i.e. most of the time), skip to next:
378 ((< (aref lm-score-table square) score-max))
379 ;; If score is better, beware of non free squares:
380 ((> (setq score (aref lm-score-table square)) score-max)
381 (if (zerop (aref lm-board square)) ; is it free ?
382 (setq count 1 ; yes: take it !
383 best-square square
384 score-max score)
385 (aset lm-score-table square -1))) ; no: kill it !
386 ;; If score is equally good, choose randomly. But first check freeness:
387 ((not (zerop (aref lm-board square)))
388 (aset lm-score-table square -1))
389 ((zerop (random (setq count (1+ count))))
390 (setq best-square square
391 score-max score)))
392 (setq square (1+ square))) ; try next square
393 best-square))
394
395 ;;;_ - INITIALIZING THE SCORE TABLE.
396
397 ;; At initialization the board is empty so that every qtuple amounts for
398 ;; nil-score. Therefore, the score of any square is nil-score times the number
399 ;; of qtuples that pass through it. This number is 3 in a corner and 20 if you
400 ;; are sufficiently far from the sides. As computing the number is time
401 ;; consuming, we initialize every square with 20*nil-score and then only
402 ;; consider squares at less than 5 squares from one side. We speed this up by
403 ;; taking symmetry into account.
404 ;; Also, as it is likely that successive games will be played on a board with
405 ;; same size, it is a good idea to save the initial SCORE-TABLE configuration.
406
407 (defvar lm-saved-score-table nil
408 "Recorded initial value of previous score table.")
409
410 (defvar lm-saved-board-width nil
411 "Recorded value of previous board width.")
412
413 (defvar lm-saved-board-height nil
414 "Recorded value of previous board height.")
415
416
417 (defun lm-init-score-table ()
418 "Create the score table vector and fill it with initial values."
419 (if (and lm-saved-score-table ; Has it been stored last time ?
420 (= lm-board-width lm-saved-board-width)
421 (= lm-board-height lm-saved-board-height))
422 (setq lm-score-table (copy-sequence lm-saved-score-table))
423 ;; No, compute it:
424 (setq lm-score-table
425 (make-vector lm-vector-length (* 20 nil-score)))
426 (let (i j maxi maxj maxi2 maxj2)
427 (setq maxi (/ (1+ lm-board-width) 2)
428 maxj (/ (1+ lm-board-height) 2)
429 maxi2 (min 4 maxi)
430 maxj2 (min 4 maxj))
431 ;; We took symmetry into account and could use it more if the board
432 ;; would have been square and not rectangular !
433 ;; In our case we deal with all (i,j) in the set [1..maxi2]*[1..maxj] U
434 ;; [maxi2+1..maxi]*[1..maxj2]. Maxi2 and maxj2 are used because the
435 ;; board may well be less than 8 by 8 !
436 (setq i 1)
437 (while (<= i maxi2)
438 (setq j 1)
439 (while (<= j maxj)
440 (lm-init-square-score i j)
441 (setq j (1+ j)))
442 (setq i (1+ i)))
443 (while (<= i maxi)
444 (setq j 1)
445 (while (<= j maxj2)
446 (lm-init-square-score i j)
447 (setq j (1+ j)))
448 (setq i (1+ i))))
449 (setq lm-saved-score-table (copy-sequence lm-score-table)
450 lm-saved-board-width lm-board-width
451 lm-saved-board-height lm-board-height)))
452
453 (defun lm-nb-qtuples (i j)
454 "Return the number of qtuples containing square I,J."
455 ;; This function is complicated because we have to deal
456 ;; with ugly cases like 3 by 6 boards, but it works.
457 ;; If you have a simpler (and correct) solution, send it to me. Thanks !
458 (let ((left (min 4 (1- i)))
459 (right (min 4 (- lm-board-width i)))
460 (up (min 4 (1- j)))
461 (down (min 4 (- lm-board-height j))))
462 (+ -12
463 (min (max (+ left right) 3) 8)
464 (min (max (+ up down) 3) 8)
465 (min (max (+ (min left up) (min right down)) 3) 8)
466 (min (max (+ (min right up) (min left down)) 3) 8))))
467
468 (defun lm-init-square-score (i j)
469 "Give initial score to square I,J and to its mirror images."
470 (let ((ii (1+ (- lm-board-width i)))
471 (jj (1+ (- lm-board-height j)))
472 (sc (* (lm-nb-qtuples i j) (aref lm-score-trans-table 0))))
473 (aset lm-score-table (lm-xy-to-index i j) sc)
474 (aset lm-score-table (lm-xy-to-index ii j) sc)
475 (aset lm-score-table (lm-xy-to-index i jj) sc)
476 (aset lm-score-table (lm-xy-to-index ii jj) sc)))
477 ;;;_ - MAINTAINING THE SCORE TABLE.
478
479
480 ;; We do not provide functions for computing the SCORE-TABLE given the
481 ;; contents of the BOARD. This would involve heavy nested loops, with time
482 ;; proportional to the size of the board. It is better to update the
483 ;; SCORE-TABLE after each move. Updating needs not modify more than 36
484 ;; squares: it is done in constant time.
485
486 (defun lm-update-score-table (square dval)
487 "Update score table after SQUARE received a DVAL increment."
488 ;; The board has already been updated when this function is called.
489 ;; Updating scores is done by looking for qtuples boundaries in all four
490 ;; directions and then calling update-score-in-direction.
491 ;; Finally all squares received the right increment, and then are up to
492 ;; date, except possibly for SQUARE itself if we are taking a move back for
493 ;; its score had been set to -1 at the time.
494 (let* ((x (lm-index-to-x square))
495 (y (lm-index-to-y square))
496 (imin (max -4 (- 1 x)))
497 (jmin (max -4 (- 1 y)))
498 (imax (min 0 (- lm-board-width x 4)))
499 (jmax (min 0 (- lm-board-height y 4))))
500 (lm-update-score-in-direction imin imax
501 square 1 0 dval)
502 (lm-update-score-in-direction jmin jmax
503 square 0 1 dval)
504 (lm-update-score-in-direction (max imin jmin) (min imax jmax)
505 square 1 1 dval)
506 (lm-update-score-in-direction (max (- 1 y) -4
507 (- x lm-board-width))
508 (min 0 (- x 5)
509 (- lm-board-height y 4))
510 square -1 1 dval)))
511
512 (defun lm-update-score-in-direction (left right square dx dy dval)
513 "Update scores for all squares in the qtuples in range.
514 That is, those between the LEFTth square and the RIGHTth after SQUARE,
515 along the DX, DY direction, considering that DVAL has been added on SQUARE."
516 ;; We always have LEFT <= 0, RIGHT <= 0 and DEPL > 0 but we may very well
517 ;; have LEFT > RIGHT, indicating that no qtuple contains SQUARE along that
518 ;; DX,DY direction.
519 (cond
520 ((> left right)) ; Quit
521 (t ; Else ..
522 (let (depl square0 square1 square2 count delta)
523 (setq depl (lm-xy-to-index dx dy)
524 square0 (+ square (* left depl))
525 square1 (+ square (* right depl))
526 square2 (+ square0 (* 4 depl)))
527 ;; Compute the contents of the first qtuple:
528 (setq square square0
529 count 0)
530 (while (<= square square2)
531 (setq count (+ count (aref lm-board square))
532 square (+ square depl)))
533 (while (<= square0 square1)
534 ;; Update the squares of the qtuple beginning in SQUARE0 and ending
535 ;; in SQUARE2.
536 (setq delta (- (aref lm-score-trans-table count)
537 (aref lm-score-trans-table (- count dval))))
538 (cond ((not (zerop delta)) ; or else nothing to update
539 (setq square square0)
540 (while (<= square square2)
541 (if (zerop (aref lm-board square)) ; only for free squares
542 (aset lm-score-table square
543 (+ (aref lm-score-table square) delta)))
544 (setq square (+ square depl)))))
545 ;; Then shift the qtuple one square along DEPL, this only requires
546 ;; modifying SQUARE0 and SQUARE2.
547 (setq square2 (+ square2 depl)
548 count (+ count (- (aref lm-board square0))
549 (aref lm-board square2))
550 square0 (+ square0 depl)))))))
551
552 ;;;
553 ;;; GAME CONTROL.
554 ;;;
555
556 ;; Several variables are used to monitor a game, including a GAME-HISTORY (the
557 ;; list of all (SQUARE . PREVSCORE) played) that allows to take moves back
558 ;; (anti-updating the score table) and to compute the table from scratch in
559 ;; case of an interruption.
560
561 (defvar lm-game-in-progress nil
562 "Non-nil if a game is in progress.")
563
564 (defvar lm-game-history nil
565 "A record of all moves that have been played during current game.")
566
567 (defvar lm-number-of-moves nil
568 "Number of moves already played in current game.")
569
570 (defvar lm-number-of-human-moves nil
571 "Number of moves already played by human in current game.")
572
573 (defvar lm-emacs-played-first nil
574 "Non-nil if Emacs played first.")
575
576 (defvar lm-human-took-back nil
577 "Non-nil if Human took back a move during the game.")
578
579 (defvar lm-human-refused-draw nil
580 "Non-nil if Human refused Emacs offer of a draw.")
581
582 (defvar lm-emacs-is-computing nil
583 ;; This is used to detect interruptions. Hopefully, it should not be needed.
584 "Non-nil if Emacs is in the middle of a computation.")
585
586
587 (defun lm-start-game (n m)
588 "Initialize a new game on an N by M board."
589 (setq lm-emacs-is-computing t) ; Raise flag
590 (setq lm-game-in-progress t)
591 (setq lm-board-width n
592 lm-board-height m
593 lm-vector-length (1+ (* (+ m 2) (1+ n)))
594 lm-draw-limit (/ (* 7 n m) 10))
595 (setq lm-emacs-won nil
596 lm-game-history nil
597 lm-number-of-moves 0
598 lm-number-of-human-moves 0
599 lm-emacs-played-first nil
600 lm-human-took-back nil
601 lm-human-refused-draw nil)
602 (lm-init-display n m) ; Display first: the rest takes time
603 (lm-init-score-table) ; INIT-BOARD requires that the score
604 (lm-init-board) ; table be already created.
605 (setq lm-emacs-is-computing nil))
606
607 (defun lm-play-move (square val &optional dont-update-score)
608 "Go to SQUARE, play VAL and update everything."
609 (setq lm-emacs-is-computing t) ; Raise flag
610 (cond ((= 1 val) ; a Human move
611 (setq lm-number-of-human-moves (1+ lm-number-of-human-moves)))
612 ((zerop lm-number-of-moves) ; an Emacs move. Is it first ?
613 (setq lm-emacs-played-first t)))
614 (setq lm-game-history
615 (cons (cons square (aref lm-score-table square))
616 lm-game-history)
617 lm-number-of-moves (1+ lm-number-of-moves))
618 (lm-plot-square square val)
619 (aset lm-board square val) ; *BEFORE* UPDATE-SCORE !
620 (if dont-update-score nil
621 (lm-update-score-table square val) ; previous val was 0: dval = val
622 (aset lm-score-table square -1))
623 (setq lm-emacs-is-computing nil))
624
625 (defun lm-take-back ()
626 "Take back last move and update everything."
627 (setq lm-emacs-is-computing t)
628 (let* ((last-move (car lm-game-history))
629 (square (car last-move))
630 (oldval (aref lm-board square)))
631 (if (= 1 oldval)
632 (setq lm-number-of-human-moves (1- lm-number-of-human-moves)))
633 (setq lm-game-history (cdr lm-game-history)
634 lm-number-of-moves (1- lm-number-of-moves))
635 (lm-plot-square square 0)
636 (aset lm-board square 0) ; *BEFORE* UPDATE-SCORE !
637 (lm-update-score-table square (- oldval))
638 (aset lm-score-table square (cdr last-move)))
639 (setq lm-emacs-is-computing nil))
640
641
642 ;;;_ + SESSION CONTROL.
643
644 (defvar lm-number-of-trials 0
645 "The number of times that landmark has been run.")
646
647 (defvar lm-sum-of-moves 0
648 "The total number of moves made in all games.")
649
650 (defvar lm-number-of-emacs-wins 0
651 "Number of games Emacs won in this session.")
652
653 (defvar lm-number-of-human-wins 0
654 "Number of games you won in this session.")
655
656 (defvar lm-number-of-draws 0
657 "Number of games already drawn in this session.")
658
659
660 (defun lm-terminate-game (result)
661 "Terminate the current game with RESULT."
662 (setq lm-number-of-trials (1+ lm-number-of-trials))
663 (setq lm-sum-of-moves (+ lm-sum-of-moves lm-number-of-moves))
664 (if (eq result 'crash-game)
665 (message
666 "Sorry, I have been interrupted and cannot resume that game..."))
667 (lm-display-statistics)
668 ;;(ding)
669 (setq lm-game-in-progress nil))
670
671 (defun lm-crash-game ()
672 "What to do when Emacs detects it has been interrupted."
673 (setq lm-emacs-is-computing nil)
674 (lm-terminate-game 'crash-game)
675 (sit-for 4) ; Let's see the message
676 (lm-prompt-for-other-game))
677
678
679 ;;;_ + INTERACTIVE COMMANDS.
680
681 (defun lm-emacs-plays ()
682 "Compute Emacs next move and play it."
683 (interactive)
684 (lm-switch-to-window)
685 (cond
686 (lm-emacs-is-computing
687 (lm-crash-game))
688 ((not lm-game-in-progress)
689 (lm-prompt-for-other-game))
690 (t
691 (message "Let me think...")
692 (let (square score)
693 (setq square (lm-strongest-square))
694 (cond ((null square)
695 (lm-terminate-game 'nobody-won))
696 (t
697 (setq score (aref lm-score-table square))
698 (lm-play-move square 6)
699 (cond ((>= score lm-winning-threshold)
700 (setq lm-emacs-won t) ; for font-lock
701 (lm-find-filled-qtuple square 6)
702 (lm-terminate-game 'emacs-won))
703 ((zerop score)
704 (lm-terminate-game 'nobody-won))
705 ((and (> lm-number-of-moves lm-draw-limit)
706 (not lm-human-refused-draw)
707 (lm-offer-a-draw))
708 (lm-terminate-game 'draw-agreed))
709 (t
710 (lm-prompt-for-move)))))))))
711
712 ;; For small square dimensions this is approximate, since though measured in
713 ;; pixels, event's (X . Y) is a character's top-left corner.
714 (defun lm-click (click)
715 "Position at the square where you click."
716 (interactive "e")
717 (and (windowp (posn-window (setq click (event-end click))))
718 (numberp (posn-point click))
719 (select-window (posn-window click))
720 (setq click (posn-col-row click))
721 (lm-goto-xy
722 (min (max (/ (+ (- (car click)
723 lm-x-offset
724 1)
725 (window-hscroll)
726 lm-square-width
727 (% lm-square-width 2)
728 (/ lm-square-width 2))
729 lm-square-width)
730 1)
731 lm-board-width)
732 (min (max (/ (+ (- (cdr click)
733 lm-y-offset
734 1)
735 (let ((inhibit-point-motion-hooks t))
736 (count-lines 1 (window-start)))
737 lm-square-height
738 (% lm-square-height 2)
739 (/ lm-square-height 2))
740 lm-square-height)
741 1)
742 lm-board-height))))
743
744 (defun lm-mouse-play (click)
745 "Play at the square where you click."
746 (interactive "e")
747 (if (lm-click click)
748 (lm-human-plays)))
749
750 (defun lm-human-plays ()
751 "Signal to the Lm program that you have played.
752 You must have put the cursor on the square where you want to play.
753 If the game is finished, this command requests for another game."
754 (interactive)
755 (lm-switch-to-window)
756 (cond
757 (lm-emacs-is-computing
758 (lm-crash-game))
759 ((not lm-game-in-progress)
760 (lm-prompt-for-other-game))
761 (t
762 (let (square score)
763 (setq square (lm-point-square))
764 (cond ((null square)
765 (error "Your point is not on a square. Retry !"))
766 ((not (zerop (aref lm-board square)))
767 (error "Your point is not on a free square. Retry !"))
768 (t
769 (setq score (aref lm-score-table square))
770 (lm-play-move square 1)
771 (cond ((and (>= score lm-loosing-threshold)
772 ;; Just testing SCORE > THRESHOLD is not enough for
773 ;; detecting wins, it just gives an indication that
774 ;; we confirm with LM-FIND-FILLED-QTUPLE.
775 (lm-find-filled-qtuple square 1))
776 (lm-terminate-game 'human-won))
777 (t
778 (lm-emacs-plays)))))))))
779
780 (defun lm-human-takes-back ()
781 "Signal to the Lm program that you wish to take back your last move."
782 (interactive)
783 (lm-switch-to-window)
784 (cond
785 (lm-emacs-is-computing
786 (lm-crash-game))
787 ((not lm-game-in-progress)
788 (message "Too late for taking back...")
789 (sit-for 4)
790 (lm-prompt-for-other-game))
791 ((zerop lm-number-of-human-moves)
792 (message "You have not played yet... Your move ?"))
793 (t
794 (message "One moment, please...")
795 ;; It is possible for the user to let Emacs play several consecutive
796 ;; moves, so that the best way to know when to stop taking back moves is
797 ;; to count the number of human moves:
798 (setq lm-human-took-back t)
799 (let ((number lm-number-of-human-moves))
800 (while (= number lm-number-of-human-moves)
801 (lm-take-back)))
802 (lm-prompt-for-move))))
803
804 (defun lm-human-resigns ()
805 "Signal to the Lm program that you may want to resign."
806 (interactive)
807 (lm-switch-to-window)
808 (cond
809 (lm-emacs-is-computing
810 (lm-crash-game))
811 ((not lm-game-in-progress)
812 (message "There is no game in progress"))
813 ((y-or-n-p "You mean, you resign ")
814 (lm-terminate-game 'human-resigned))
815 ((y-or-n-p "You mean, we continue ")
816 (lm-prompt-for-move))
817 (t
818 (lm-terminate-game 'human-resigned)))) ; OK. Accept it
819
820 ;;;_ + PROMPTING THE HUMAN PLAYER.
821
822 (defun lm-prompt-for-move ()
823 "Display a message asking for Human's move."
824 (message (if (zerop lm-number-of-human-moves)
825 "Your move ? (move to a free square and hit X, RET ...)"
826 "Your move ?"))
827 ;; This may seem silly, but if one omits the following line (or a similar
828 ;; one), the cursor may very well go to some place where POINT is not.
829 (save-excursion (set-buffer (other-buffer))))
830
831 (defun lm-prompt-for-other-game ()
832 "Ask for another game, and start it."
833 (if (y-or-n-p "Another game ")
834 (if (y-or-n-p "Retain learned weights ")
835 (lm 2)
836 (lm 1))
837 (message "Chicken !")))
838
839 (defun lm-offer-a-draw ()
840 "Offer a draw and return t if Human accepted it."
841 (or (y-or-n-p "I offer you a draw. Do you accept it ")
842 (not (setq lm-human-refused-draw t))))
843
844
845 (defun lm-max-width ()
846 "Largest possible board width for the current window."
847 (1+ (/ (- (window-width (selected-window))
848 lm-x-offset lm-x-offset 1)
849 lm-square-width)))
850
851 (defun lm-max-height ()
852 "Largest possible board height for the current window."
853 (1+ (/ (- (window-height (selected-window))
854 lm-y-offset lm-y-offset 2)
855 ;; 2 instead of 1 because WINDOW-HEIGHT includes the mode line !
856 lm-square-height)))
857
858 (defun lm-point-y ()
859 "Return the board row where point is."
860 (let ((inhibit-point-motion-hooks t))
861 (1+ (/ (- (count-lines 1 (point)) lm-y-offset (if (bolp) 0 1))
862 lm-square-height))))
863
864 (defun lm-point-square ()
865 "Return the index of the square point is on."
866 (let ((inhibit-point-motion-hooks t))
867 (lm-xy-to-index (1+ (/ (- (current-column) lm-x-offset)
868 lm-square-width))
869 (lm-point-y))))
870
871 (defun lm-goto-square (index)
872 "Move point to square number INDEX."
873 (lm-goto-xy (lm-index-to-x index) (lm-index-to-y index)))
874
875 (defun lm-goto-xy (x y)
876 "Move point to square at X, Y coords."
877 (let ((inhibit-point-motion-hooks t))
878 (goto-line (+ 1 lm-y-offset (* lm-square-height (1- y)))))
879 (move-to-column (+ lm-x-offset (* lm-square-width (1- x)))))
880
881 (defun lm-plot-square (square value)
882 "Draw 'X', 'O' or '.' on SQUARE depending on VALUE, leave point there."
883 (or (= value 1)
884 (lm-goto-square square))
885 (let ((inhibit-read-only t)
886 (inhibit-point-motion-hooks t))
887 (insert-and-inherit (cond ((= value 1) ?.)
888 ((= value 2) ?N)
889 ((= value 3) ?S)
890 ((= value 4) ?E)
891 ((= value 5) ?W)
892 ((= value 6) ?^)))
893
894 (and (zerop value)
895 (add-text-properties (1- (point)) (point)
896 '(mouse-face highlight
897 help-echo "\
898 mouse-1: get robot moving, mouse-2: play on this square")))
899 (delete-char 1)
900 (backward-char 1))
901 (sit-for 0)) ; Display NOW
902
903 (defun lm-init-display (n m)
904 "Display an N by M Lm board."
905 (buffer-disable-undo (current-buffer))
906 (let ((inhibit-read-only t)
907 (point 1) opoint
908 (intangible t)
909 (i m) j x)
910 ;; Try to minimize number of chars (because of text properties)
911 (setq tab-width
912 (if (zerop (% lm-x-offset lm-square-width))
913 lm-square-width
914 (max (/ (+ (% lm-x-offset lm-square-width)
915 lm-square-width 1) 2) 2)))
916 (erase-buffer)
917 (newline lm-y-offset)
918 (while (progn
919 (setq j n
920 x (- lm-x-offset lm-square-width))
921 (while (>= (setq j (1- j)) 0)
922 (insert-char ?\t (/ (- (setq x (+ x lm-square-width))
923 (current-column))
924 tab-width))
925 (insert-char ? (- x (current-column)))
926 (if (setq intangible (not intangible))
927 (put-text-property point (point) 'intangible 2))
928 (and (zerop j)
929 (= i (- m 2))
930 (progn
931 (while (>= i 3)
932 (append-to-buffer (current-buffer) opoint (point))
933 (setq i (- i 2)))
934 (goto-char (point-max))))
935 (setq point (point))
936 (insert ?=)
937 (add-text-properties point (point)
938 '(mouse-face highlight help-echo "\
939 mouse-1: get robot moving, mouse-2: play on this square")))
940 (> (setq i (1- i)) 0))
941 (if (= i (1- m))
942 (setq opoint point))
943 (insert-char ?\n lm-square-height))
944 (or (eq (char-after 1) ?.)
945 (put-text-property 1 2 'point-entered
946 (lambda (x y) (if (bobp) (forward-char)))))
947 (or intangible
948 (put-text-property point (point) 'intangible 2))
949 (put-text-property point (point) 'point-entered
950 (lambda (x y) (if (eobp) (backward-char))))
951 (put-text-property (point-min) (point) 'category 'lm-mode))
952 (lm-goto-xy (/ (1+ n) 2) (/ (1+ m) 2)) ; center of the board
953 (sit-for 0)) ; Display NOW
954
955 (defun lm-display-statistics ()
956 "Obnoxiously display some statistics about previous games in mode line."
957 ;; We store this string in the mode-line-process local variable.
958 ;; This is certainly not the cleanest way out ...
959 (setq mode-line-process
960 (format ": Trials: %d, Avg#Moves: %d"
961 lm-number-of-trials
962 (if (zerop lm-number-of-trials)
963 0
964 (/ lm-sum-of-moves lm-number-of-trials))))
965 (force-mode-line-update))
966
967 (defun lm-switch-to-window ()
968 "Find or create the Lm buffer, and display it."
969 (interactive)
970 (let ((buff (get-buffer "*Lm*")))
971 (if buff ; Buffer exists:
972 (switch-to-buffer buff) ; no problem.
973 (if lm-game-in-progress
974 (lm-crash-game)) ; buffer has been killed or something
975 (switch-to-buffer "*Lm*") ; Anyway, start anew.
976 (lm-mode))))
977
978
979 ;;;_ + CROSSING WINNING QTUPLES.
980
981 ;; When someone succeeds in filling a qtuple, we draw a line over the five
982 ;; corresponding squares. One problem is that the program does not know which
983 ;; squares ! It only knows the square where the last move has been played and
984 ;; who won. The solution is to scan the board along all four directions.
985
986 (defun lm-find-filled-qtuple (square value)
987 "Return t if SQUARE belongs to a qtuple filled with VALUEs."
988 (or (lm-check-filled-qtuple square value 1 0)
989 (lm-check-filled-qtuple square value 0 1)
990 (lm-check-filled-qtuple square value 1 1)
991 (lm-check-filled-qtuple square value -1 1)))
992
993 (defun lm-check-filled-qtuple (square value dx dy)
994 "Return t if SQUARE belongs to a qtuple filled with VALUEs along DX, DY."
995 (let ((a 0) (b 0)
996 (left square) (right square)
997 (depl (lm-xy-to-index dx dy)))
998 (while (and (> a -4) ; stretch tuple left
999 (= value (aref lm-board (setq left (- left depl)))))
1000 (setq a (1- a)))
1001 (while (and (< b (+ a 4)) ; stretch tuple right
1002 (= value (aref lm-board (setq right (+ right depl)))))
1003 (setq b (1+ b)))
1004 (cond ((= b (+ a 4)) ; tuple length = 5 ?
1005 (lm-cross-qtuple (+ square (* a depl)) (+ square (* b depl))
1006 dx dy)
1007 t))))
1008
1009 (defun lm-cross-qtuple (square1 square2 dx dy)
1010 "Cross every square between SQUARE1 and SQUARE2 in the DX, DY direction."
1011 (save-excursion ; Not moving point from last square
1012 (let ((depl (lm-xy-to-index dx dy))
1013 (inhibit-read-only t)
1014 (inhibit-point-motion-hooks t))
1015 ;; WARNING: this function assumes DEPL > 0 and SQUARE2 > SQUARE1
1016 (while (/= square1 square2)
1017 (lm-goto-square square1)
1018 (setq square1 (+ square1 depl))
1019 (cond
1020 ((= dy 0) ; Horizontal
1021 (forward-char 1)
1022 (insert-char ?- (1- lm-square-width) t)
1023 (delete-region (point) (progn
1024 (skip-chars-forward " \t")
1025 (point))))
1026 ((= dx 0) ; Vertical
1027 (let ((lm-n 1)
1028 (column (current-column)))
1029 (while (< lm-n lm-square-height)
1030 (setq lm-n (1+ lm-n))
1031 (forward-line 1)
1032 (indent-to column)
1033 (insert-and-inherit ?|))))
1034 ((= dx -1) ; 1st Diagonal
1035 (indent-to (prog1 (- (current-column) (/ lm-square-width 2))
1036 (forward-line (/ lm-square-height 2))))
1037 (insert-and-inherit ?/))
1038 (t ; 2nd Diagonal
1039 (indent-to (prog1 (+ (current-column) (/ lm-square-width 2))
1040 (forward-line (/ lm-square-height 2))))
1041 (insert-and-inherit ?\\))))))
1042 (sit-for 0)) ; Display NOW
1043
1044
1045 ;;;_ + CURSOR MOTION.
1046
1047 ;; previous-line and next-line don't work right with intangible newlines
1048 (defun lm-move-down ()
1049 "Move point down one row on the Lm board."
1050 (interactive)
1051 (if (< (lm-point-y) lm-board-height)
1052 (next-line 1)));;; lm-square-height)))
1053
1054 (defun lm-move-up ()
1055 "Move point up one row on the Lm board."
1056 (interactive)
1057 (if (> (lm-point-y) 1)
1058 (previous-line lm-square-height)))
1059
1060 (defun lm-move-ne ()
1061 "Move point North East on the Lm board."
1062 (interactive)
1063 (lm-move-up)
1064 (forward-char))
1065
1066 (defun lm-move-se ()
1067 "Move point South East on the Lm board."
1068 (interactive)
1069 (lm-move-down)
1070 (forward-char))
1071
1072 (defun lm-move-nw ()
1073 "Move point North West on the Lm board."
1074 (interactive)
1075 (lm-move-up)
1076 (backward-char))
1077
1078 (defun lm-move-sw ()
1079 "Move point South West on the Lm board."
1080 (interactive)
1081 (lm-move-down)
1082 (backward-char))
1083
1084 (defun lm-beginning-of-line ()
1085 "Move point to first square on the Lm board row."
1086 (interactive)
1087 (move-to-column lm-x-offset))
1088
1089 (defun lm-end-of-line ()
1090 "Move point to last square on the Lm board row."
1091 (interactive)
1092 (move-to-column (+ lm-x-offset
1093 (* lm-square-width (1- lm-board-width)))))
1094
1095
1096 ;;;_ + Simulation variables
1097
1098 ;;;_ - lm-nvar
1099 (defvar lm-nvar 0.0075
1100 "Not used.
1101 Affects a noise generator which was used in an earlier incarnation of
1102 this program to add a random element to the way moves were made.")
1103 ;;;_ - lists of cardinal directions
1104 ;;;_ :
1105 (defvar lm-ns '(lm-n lm-s)
1106 "Used when doing something relative to the north and south axes.")
1107 (defvar lm-ew '(lm-e lm-w)
1108 "Used when doing something relative to the east and west axes.")
1109 (defvar lm-directions '(lm-n lm-s lm-e lm-w)
1110 "The cardinal directions.")
1111 (defvar lm-8-directions
1112 '((lm-n) (lm-n lm-w) (lm-w) (lm-s lm-w)
1113 (lm-s) (lm-s lm-e) (lm-e) (lm-n lm-e))
1114 "The full 8 possible directions.")
1115
1116 (defvar lm-number-of-moves
1117 "The number of moves made by the robot so far.")
1118
1119
1120 ;;;_* Terry's mods to create lm.el
1121
1122 ;;;(setq lm-debug nil)
1123 (defvar lm-debug nil
1124 "If non-nil, debugging is printed.")
1125 (defcustom lm-one-moment-please nil
1126 "If non-nil, print \"One moment please\" when a new board is generated.
1127 The drawback of this is you don't see how many moves the last run took
1128 because it is overwritten by \"One moment please\"."
1129 :type 'boolean
1130 :group 'lm)
1131 (defcustom lm-output-moves t
1132 "If non-nil, output number of moves so far on a move-by-move basis."
1133 :type 'boolean
1134 :group 'lm)
1135
1136
1137 (defun lm-weights-debug ()
1138 (if lm-debug
1139 (progn (lm-print-wts) (lm-blackbox) (lm-print-y,s,noise)
1140 (lm-print-smell))))
1141
1142 ;;;_ - Printing various things
1143 (defun lm-print-distance-int (direction)
1144 (interactive)
1145 (insert (format "%S %S " direction (get direction 'distance))))
1146
1147
1148 (defun lm-print-distance ()
1149 (insert (format "tree: %S \n" (calc-distance-of-robot-from 'lm-tree)))
1150 (mapc 'lm-print-distance-int lm-directions))
1151
1152
1153 ;;(setq direction 'lm-n)
1154 ;;(get 'lm-n 'lm-s)
1155 (defun lm-nslify-wts-int (direction)
1156 (mapcar (lambda (target-direction)
1157 (get direction target-direction))
1158 lm-directions))
1159
1160
1161 (defun lm-nslify-wts ()
1162 (interactive)
1163 (let ((l (apply 'append (mapcar 'lm-nslify-wts-int lm-directions))))
1164 (insert (format "set data_value WTS \n %s \n" l))
1165 (insert (format "/* max: %S min: %S */"
1166 (eval (cons 'max l)) (eval (cons 'min l))))))
1167
1168 (defun lm-print-wts-int (direction)
1169 (mapc (lambda (target-direction)
1170 (insert (format "%S %S %S "
1171 direction
1172 target-direction
1173 (get direction target-direction))))
1174 lm-directions)
1175 (insert "\n"))
1176
1177 (defun lm-print-wts ()
1178 (interactive)
1179 (save-excursion
1180 (set-buffer "*lm-wts*")
1181 (insert "==============================\n")
1182 (mapc 'lm-print-wts-int lm-directions)))
1183
1184 (defun lm-print-moves (moves)
1185 (interactive)
1186 (save-excursion
1187 (set-buffer "*lm-moves*")
1188 (insert (format "%S\n" moves))))
1189
1190
1191 (defun lm-print-y,s,noise-int (direction)
1192 (insert (format "%S:lm-y %S, s %S, noise %S \n"
1193 (symbol-name direction)
1194 (get direction 'y_t)
1195 (get direction 's)
1196 (get direction 'noise)
1197 )))
1198
1199 (defun lm-print-y,s,noise ()
1200 (interactive)
1201 (save-excursion
1202 (set-buffer "*lm-y,s,noise*")
1203 (insert "==============================\n")
1204 (mapc 'lm-print-y,s,noise-int lm-directions)))
1205
1206 (defun lm-print-smell-int (direction)
1207 (insert (format "%S: smell: %S \n"
1208 (symbol-name direction)
1209 (get direction 'smell))))
1210
1211 (defun lm-print-smell ()
1212 (interactive)
1213 (save-excursion
1214 (set-buffer "*lm-smell*")
1215 (insert "==============================\n")
1216 (insert (format "tree: %S \n" (get 'z 't)))
1217 (mapc 'lm-print-smell-int lm-directions)))
1218
1219 (defun lm-print-w0-int (direction)
1220 (insert (format "%S: w0: %S \n"
1221 (symbol-name direction)
1222 (get direction 'w0))))
1223
1224 (defun lm-print-w0 ()
1225 (interactive)
1226 (save-excursion
1227 (set-buffer "*lm-w0*")
1228 (insert "==============================\n")
1229 (mapc 'lm-print-w0-int lm-directions)))
1230
1231 (defun lm-blackbox ()
1232 (save-excursion
1233 (set-buffer "*lm-blackbox*")
1234 (insert "==============================\n")
1235 (insert "I smell: ")
1236 (mapc (lambda (direction)
1237 (if (> (get direction 'smell) 0)
1238 (insert (format "%S " direction))))
1239 lm-directions)
1240 (insert "\n")
1241
1242 (insert "I move: ")
1243 (mapc (lambda (direction)
1244 (if (> (get direction 'y_t) 0)
1245 (insert (format "%S " direction))))
1246 lm-directions)
1247 (insert "\n")
1248 (lm-print-wts-blackbox)
1249 (insert (format "z_t-z_t-1: %S" (- (get 'z 't) (get 'z 't-1))))
1250 (lm-print-distance)
1251 (insert "\n")))
1252
1253 (defun lm-print-wts-blackbox ()
1254 (interactive)
1255 (mapc 'lm-print-wts-int lm-directions))
1256
1257 ;;;_ - learning parameters
1258 (defcustom lm-bound 0.005
1259 "The maximum that w0j may be."
1260 :type 'number
1261 :group 'lm)
1262 (defcustom lm-c 1.0
1263 "A factor applied to modulate the increase in wij.
1264 Used in the function lm-update-normal-weights."
1265 :type 'number
1266 :group 'lm)
1267 (defcustom lm-c-naught 0.5
1268 "A factor applied to modulate the increase in w0j.
1269 Used in the function lm-update-naught-weights."
1270 :type 'number
1271 :group 'lm)
1272 (defvar lm-initial-w0 0.0)
1273 (defvar lm-initial-wij 0.0)
1274 (defcustom lm-no-payoff 0
1275 "The amount of simulation cycles that have occurred with no movement.
1276 Used to move the robot when he is stuck in a rut for some reason."
1277 :type 'integer
1278 :group 'lm)
1279 (defcustom lm-max-stall-time 2
1280 "The maximum number of cycles that the robot can remain stuck in a place.
1281 After this limit is reached, lm-random-move is called to push him out of it."
1282 :type 'integer
1283 :group 'lm)
1284
1285
1286 ;;;_ + Randomizing functions
1287 ;;;_ - lm-flip-a-coin ()
1288 (defun lm-flip-a-coin ()
1289 (if (> (random 5000) 2500)
1290 -1
1291 1))
1292 ;;;_ : lm-very-small-random-number ()
1293 ;(defun lm-very-small-random-number ()
1294 ; (/
1295 ; (* (/ (random 900000) 900000.0) .0001)))
1296 ;;;_ : lm-randomize-weights-for (direction)
1297 (defun lm-randomize-weights-for (direction)
1298 (mapc (lambda (target-direction)
1299 (put direction
1300 target-direction
1301 (* (lm-flip-a-coin) (/ (random 10000) 10000.0))))
1302 lm-directions))
1303 ;;;_ : lm-noise ()
1304 (defun lm-noise ()
1305 (* (- (/ (random 30001) 15000.0) 1) lm-nvar))
1306
1307 ;;;_ : lm-fix-weights-for (direction)
1308 (defun lm-fix-weights-for (direction)
1309 (mapc (lambda (target-direction)
1310 (put direction
1311 target-direction
1312 lm-initial-wij))
1313 lm-directions))
1314
1315
1316 ;;;_ + Plotting functions
1317 ;;;_ - lm-plot-internal (sym)
1318 (defun lm-plot-internal (sym)
1319 (lm-plot-square (lm-xy-to-index
1320 (get sym 'x)
1321 (get sym 'y))
1322 (get sym 'sym)))
1323 ;;;_ - lm-plot-landmarks ()
1324 (defun lm-plot-landmarks ()
1325 (setq lm-cx (/ lm-board-width 2))
1326 (setq lm-cy (/ lm-board-height 2))
1327
1328 (put 'lm-n 'x lm-cx)
1329 (put 'lm-n 'y 1)
1330 (put 'lm-n 'sym 2)
1331
1332 (put 'lm-tree 'x lm-cx)
1333 (put 'lm-tree 'y lm-cy)
1334 (put 'lm-tree 'sym 6)
1335
1336 (put 'lm-s 'x lm-cx)
1337 (put 'lm-s 'y lm-board-height)
1338 (put 'lm-s 'sym 3)
1339
1340 (put 'lm-w 'x 1)
1341 (put 'lm-w 'y (/ lm-board-height 2))
1342 (put 'lm-w 'sym 5)
1343
1344 (put 'lm-e 'x lm-board-width)
1345 (put 'lm-e 'y (/ lm-board-height 2))
1346 (put 'lm-e 'sym 4)
1347
1348 (mapc 'lm-plot-internal '(lm-n lm-s lm-e lm-w lm-tree)))
1349
1350
1351
1352 ;;;_ + Distance-calculation functions
1353 ;;;_ - square (a)
1354 (defun square (a)
1355 (* a a))
1356
1357 ;;;_ - distance (x x0 y y0)
1358 (defun distance (x x0 y y0)
1359 (sqrt (+ (square (- x x0)) (square (- y y0)))))
1360
1361 ;;;_ - calc-distance-of-robot-from (direction)
1362 (defun calc-distance-of-robot-from (direction)
1363 (put direction 'distance
1364 (distance (get direction 'x)
1365 (lm-index-to-x (lm-point-square))
1366 (get direction 'y)
1367 (lm-index-to-y (lm-point-square)))))
1368
1369 ;;;_ - calc-smell-internal (sym)
1370 (defun calc-smell-internal (sym)
1371 (let ((r (get sym 'r))
1372 (d (calc-distance-of-robot-from sym)))
1373 (if (> (* 0.5 (- 1 (/ d r))) 0)
1374 (* 0.5 (- 1 (/ d r)))
1375 0)))
1376
1377
1378 ;;;_ + Learning (neural) functions
1379 (defun lm-f (x)
1380 (cond
1381 ((> x lm-bound) lm-bound)
1382 ((< x 0.0) 0.0)
1383 (t x)))
1384
1385 (defun lm-y (direction)
1386 (let ((noise (put direction 'noise (lm-noise))))
1387 (put direction 'y_t
1388 (if (> (get direction 's) 0.0)
1389 1.0
1390 0.0))))
1391
1392 (defun lm-update-normal-weights (direction)
1393 (mapc (lambda (target-direction)
1394 (put direction target-direction
1395 (+
1396 (get direction target-direction)
1397 (* lm-c
1398 (- (get 'z 't) (get 'z 't-1))
1399 (get target-direction 'y_t)
1400 (get direction 'smell)))))
1401 lm-directions))
1402
1403 (defun lm-update-naught-weights (direction)
1404 (mapc (lambda (target-direction)
1405 (put direction 'w0
1406 (lm-f
1407 (+
1408 (get direction 'w0)
1409 (* lm-c-naught
1410 (- (get 'z 't) (get 'z 't-1))
1411 (get direction 'y_t))))))
1412 lm-directions))
1413
1414
1415 ;;;_ + Statistics gathering and creating functions
1416
1417 (defun lm-calc-current-smells ()
1418 (mapc (lambda (direction)
1419 (put direction 'smell (calc-smell-internal direction)))
1420 lm-directions))
1421
1422 (defun lm-calc-payoff ()
1423 (put 'z 't-1 (get 'z 't))
1424 (put 'z 't (calc-smell-internal 'lm-tree))
1425 (if (= (- (get 'z 't) (get 'z 't-1)) 0.0)
1426 (incf lm-no-payoff)
1427 (setf lm-no-payoff 0)))
1428
1429 (defun lm-store-old-y_t ()
1430 (mapc (lambda (direction)
1431 (put direction 'y_t-1 (get direction 'y_t)))
1432 lm-directions))
1433
1434
1435 ;;;_ + Functions to move robot
1436
1437 (defun lm-confidence-for (target-direction)
1438 (apply '+
1439 (get target-direction 'w0)
1440 (mapcar (lambda (direction)
1441 (*
1442 (get direction target-direction)
1443 (get direction 'smell)))
1444 lm-directions)))
1445
1446
1447 (defun lm-calc-confidences ()
1448 (mapc (lambda (direction)
1449 (put direction 's (lm-confidence-for direction)))
1450 lm-directions))
1451
1452 (defun lm-move ()
1453 (if (and (= (get 'lm-n 'y_t) 1.0) (= (get 'lm-s 'y_t) 1.0))
1454 (progn
1455 (mapc (lambda (dir) (put dir 'y_t 0)) lm-ns)
1456 (if lm-debug
1457 (message "n-s normalization."))))
1458 (if (and (= (get 'lm-w 'y_t) 1.0) (= (get 'lm-e 'y_t) 1.0))
1459 (progn
1460 (mapc (lambda (dir) (put dir 'y_t 0)) lm-ew)
1461 (if lm-debug
1462 (message "e-w normalization"))))
1463
1464 (mapc (lambda (pair)
1465 (if (> (get (car pair) 'y_t) 0)
1466 (funcall (car (cdr pair)))))
1467 '(
1468 (lm-n lm-move-up)
1469 (lm-s lm-move-down)
1470 (lm-e forward-char)
1471 (lm-w backward-char)))
1472 (lm-plot-square (lm-point-square) 1)
1473 (incf lm-number-of-moves)
1474 (if lm-output-moves
1475 (message (format "Moves made: %d" lm-number-of-moves))))
1476
1477
1478 (defun lm-random-move ()
1479 (mapc
1480 (lambda (direction) (put direction 'y_t 0))
1481 lm-directions)
1482 (dolist (direction (nth (random 8) lm-8-directions))
1483 (put direction 'y_t 1.0))
1484 (lm-move))
1485
1486 (defun lm-amble-robot ()
1487 (interactive)
1488 (while (> (calc-distance-of-robot-from 'lm-tree) 0)
1489
1490 (lm-store-old-y_t)
1491 (lm-calc-current-smells)
1492
1493 (if (> lm-no-payoff lm-max-stall-time)
1494 (lm-random-move)
1495 (progn
1496 (lm-calc-confidences)
1497 (mapc 'lm-y lm-directions)
1498 (lm-move)))
1499
1500 (lm-calc-payoff)
1501
1502 (mapc 'lm-update-normal-weights lm-directions)
1503 (mapc 'lm-update-naught-weights lm-directions)
1504 (if lm-debug
1505 (lm-weights-debug)))
1506 (lm-terminate-game nil))
1507
1508
1509 ;;;_ - lm-start-robot ()
1510 (defun lm-start-robot ()
1511 "Signal to the Lm program that you have played.
1512 You must have put the cursor on the square where you want to play.
1513 If the game is finished, this command requests for another game."
1514 (interactive)
1515 (lm-switch-to-window)
1516 (cond
1517 (lm-emacs-is-computing
1518 (lm-crash-game))
1519 ((not lm-game-in-progress)
1520 (lm-prompt-for-other-game))
1521 (t
1522 (let (square score)
1523 (setq square (lm-point-square))
1524 (cond ((null square)
1525 (error "Your point is not on a square. Retry !"))
1526 ((not (zerop (aref lm-board square)))
1527 (error "Your point is not on a free square. Retry !"))
1528 (t
1529 (progn
1530 (lm-plot-square square 1)
1531
1532 (lm-store-old-y_t)
1533 (lm-calc-current-smells)
1534 (put 'z 't (calc-smell-internal 'lm-tree))
1535
1536 (lm-random-move)
1537
1538 (lm-calc-payoff)
1539
1540 (mapc 'lm-update-normal-weights lm-directions)
1541 (mapc 'lm-update-naught-weights lm-directions)
1542 (lm-amble-robot)
1543 )))))))
1544
1545
1546 ;;;_ + Misc functions
1547 ;;;_ - lm-init (auto-start save-weights)
1548 (defvar lm-tree-r "")
1549
1550 (defun lm-init (auto-start save-weights)
1551
1552 (setq lm-number-of-moves 0)
1553
1554 (lm-plot-landmarks)
1555
1556 (if lm-debug
1557 (progn
1558 (save-excursion
1559 (set-buffer (get-buffer-create "*lm-w0*"))
1560 (erase-buffer)
1561 (set-buffer (get-buffer-create "*lm-moves*"))
1562 (set-buffer (get-buffer-create "*lm-wts*"))
1563 (erase-buffer)
1564 (set-buffer (get-buffer-create "*lm-y,s,noise*"))
1565 (erase-buffer)
1566 (set-buffer (get-buffer-create "*lm-smell*"))
1567 (erase-buffer)
1568 (set-buffer (get-buffer-create "*lm-blackbox*"))
1569 (erase-buffer)
1570 (set-buffer (get-buffer-create "*lm-distance*"))
1571 (erase-buffer))))
1572
1573
1574 (lm-set-landmark-signal-strengths)
1575
1576 (mapc (lambda (direction)
1577 (put direction 'y_t 0.0))
1578 lm-directions)
1579
1580 (if (not save-weights)
1581 (progn
1582 (mapc 'lm-fix-weights-for lm-directions)
1583 (mapc (lambda (direction)
1584 (put direction 'w0 lm-initial-w0))
1585 lm-directions))
1586 (message "Weights preserved for this run."))
1587
1588 (if auto-start
1589 (progn
1590 (lm-goto-xy (1+ (random lm-board-width)) (1+ (random lm-board-height)))
1591 (lm-start-robot))))
1592
1593
1594 ;;;_ - something which doesn't work
1595 ; no-a-worka!!
1596 ;(defum lm-sum-list (list)
1597 ; (if (> (length list) 0)
1598 ; (+ (car list) (lm-sum-list (cdr list)))
1599 ; 0))
1600 ; this a worka!
1601 ; (eval (cons '+ list))
1602 ;;;_ - lm-set-landmark-signal-strengths ()
1603 ;;; on a screen higher than wide, I noticed that the robot would amble
1604 ;;; left and right and not move forward. examining *lm-blackbox*
1605 ;;; revealed that there was no scent from the north and south
1606 ;;; landmarks, hence, they need less factoring down of the effect of
1607 ;;; distance on scent.
1608
1609 (defun lm-set-landmark-signal-strengths ()
1610
1611 (setq lm-tree-r (* (sqrt (+ (square lm-cx) (square lm-cy))) 1.5))
1612
1613 (mapc (lambda (direction)
1614 (put direction 'r (* lm-cx 1.1)))
1615 lm-ew)
1616 (mapc (lambda (direction)
1617 (put direction 'r (* lm-cy 1.1)))
1618 lm-ns)
1619 (put 'lm-tree 'r lm-tree-r))
1620
1621
1622 ;;;_ + lm-test-run ()
1623
1624 ;;;###autoload
1625 (defalias 'landmark-repeat 'lm-test-run)
1626 ;;;###autoload
1627 (defun lm-test-run ()
1628 "Run 100 Lm games, each time saving the weights from the previous game."
1629 (interactive)
1630
1631 (lm 1)
1632
1633 (dotimes (scratch-var 100)
1634
1635 (lm 2)))
1636
1637
1638 ;;;_ + lm: The function you invoke to play
1639
1640 ;;;###autoload
1641 (defalias 'landmark 'lm)
1642 ;;;###autoload
1643 (defun lm (parg)
1644 "Start or resume an Lm game.
1645 If a game is in progress, this command allows you to resume it.
1646 Here is the relation between prefix args and game options:
1647
1648 prefix arg | robot is auto-started | weights are saved from last game
1649 ---------------------------------------------------------------------
1650 none / 1 | yes | no
1651 2 | yes | yes
1652 3 | no | yes
1653 4 | no | no
1654
1655 You start by moving to a square and typing \\[lm-start-robot],
1656 if you did not use a prefix arg to ask for automatic start.
1657 Use \\[describe-mode] for more info."
1658 (interactive "p")
1659
1660 (setf lm-n nil lm-m nil)
1661 (lm-switch-to-window)
1662 (cond
1663 (lm-emacs-is-computing
1664 (lm-crash-game))
1665 ((or (not lm-game-in-progress)
1666 (<= lm-number-of-moves 2))
1667 (let ((max-width (lm-max-width))
1668 (max-height (lm-max-height)))
1669 (or lm-n (setq lm-n max-width))
1670 (or lm-m (setq lm-m max-height))
1671 (cond ((< lm-n 1)
1672 (error "I need at least 1 column"))
1673 ((< lm-m 1)
1674 (error "I need at least 1 row"))
1675 ((> lm-n max-width)
1676 (error "I cannot display %d columns in that window" lm-n)))
1677 (if (and (> lm-m max-height)
1678 (not (eq lm-m lm-saved-board-height))
1679 ;; Use EQ because SAVED-BOARD-HEIGHT may be nil
1680 (not (y-or-n-p (format "Do you really want %d rows " lm-m))))
1681 (setq lm-m max-height)))
1682 (if lm-one-moment-please
1683 (message "One moment, please..."))
1684 (lm-start-game lm-n lm-m)
1685 (eval (cons 'lm-init
1686 (cond
1687 ((= parg 1) '(t nil))
1688 ((= parg 2) '(t t))
1689 ((= parg 3) '(nil t))
1690 ((= parg 4) '(nil nil))
1691 (t '(nil t))))))))
1692
1693
1694 ;;;_ + Local variables
1695
1696 ;;; The following `outline-layout' local variable setting:
1697 ;;; - closes all topics from the first topic to just before the third-to-last,
1698 ;;; - shows the children of the third to last (config vars)
1699 ;;; - and the second to last (code section),
1700 ;;; - and closes the last topic (this local-variables section).
1701 ;;;Local variables:
1702 ;;;outline-layout: (0 : -1 -1 0)
1703 ;;;End:
1704
1705 (provide 'landmark)
1706
1707 ;;; arch-tag: ae5031be-96e6-459e-a3df-1df53117d3f2
1708 ;;; landmark.el ends here