Merge remote-tracking branch 'origin/eager_eval' into eager_eval

This commit is contained in:
Alexander Munch-Hansen 2018-05-13 23:55:02 +02:00
commit 2c02689577
3 changed files with 413 additions and 315 deletions

View File

@ -73,29 +73,49 @@ int no_block_at_to_idx(int to_state, int player) {
else return 1;
}
int can_bear_off(int board[], int player, int from_idx) {
int can_bear_off(int board[], int player, int from_idx, int to_idx) {
int* checker_idxs = idxs_with_checkers_of_player(board, player);
if (player == 1) {
int moving_backmost_checker = 1;
int bearing_directly_off = 0;
int all_checkers_in_last_quadrant = 1;
/* Check if bearing directly off */
if (player == 1 && to_idx == 25) bearing_directly_off = 1;
else if (player == -1 && to_idx == 0) bearing_directly_off = 1;
for (int i = 1; i <= checker_idxs[0]; i++) {
if ( !((checker_idxs[i] >= from_idx) &&
(checker_idxs[i] >= 19)) ) {
free(checker_idxs);
return 0;
if (player == 1 ) {
/* Check if all checkers are in last quardrant */
if (checker_idxs[i] < 19) {
all_checkers_in_last_quadrant = 0;
break;
}
/* Check if moving backmost checker */
if (checker_idxs[i] < from_idx) {
moving_backmost_checker = 0;
if (!bearing_directly_off) break;
}
} else {
for (int i = 1; i <= checker_idxs[0]; i++) {
if ( !((checker_idxs[i] <= from_idx) &&
(checker_idxs[i] <= 6)) ) {
free(checker_idxs);
return 0;
if (checker_idxs[i] > 6) {
all_checkers_in_last_quadrant = 0;
break;
}
if (checker_idxs[i] > from_idx) {
moving_backmost_checker = 0;
if (!bearing_directly_off) break;
}
}
}
free(checker_idxs);
return 1;
if (all_checkers_in_last_quadrant &&
(bearing_directly_off || moving_backmost_checker)) return 1;
else return 0;
}
@ -124,7 +144,7 @@ int is_move_valid(int board[], int player, int face_value, int move[]) {
&& bear_in_if_checker_on_bar(board, player, from_idx)
&& checkers_at_from_idx(from_state, player)
&& no_block_at_to_idx(to_state, player)
&& (!bearing_off || can_bear_off(board, player, from_idx))
&& (!bearing_off || can_bear_off(board, player, from_idx, to_idx))
;
}

28
report_docs.txt Normal file
View File

@ -0,0 +1,28 @@
<christoffer> Alexander og jeg skrev noget af vores bachelorprojekt om til C her i fredags.
<christoffer> Man skal virkelig passe på sine hukommelsesallokeringer.
<Jmaa> Ja, helt klart.
<christoffer> Jeg fandt et memory leak, der lækkede 100 MiB hukommelse i sekundet.
<Jmaa> Hvilken del blev C-ificeret?
<Jmaa> Damned
<christoffer> Årsagen var at vi gav et objekt med tilbage til Python uden at dekrementere dets ref-count, så fortolkeren stadig troede at nogen havde brug for det.
<christoffer> Den del af spillogikken, der tjekker om træk er gyldige.
<christoffer> Det bliver kaldt ret mange tusinde gange pr. spil, så vi tænkte at der måske kunne være lidt optimering at hente i at omskrive det til C.
<Jmaa> Ok, så I har ikke selv brugt alloc og free. Det er alligevel noget.
<christoffer> Metoden selv blev 7 gange hurtigere!
<Jmaa> Wow!
<christoffer> Jo. Det endte vi også med at gøre.
<christoffer> Vi havde brug for lister af variabel størrelse. Det endte med en struct med et "size" felt og et "list" felt.
<Jmaa> Inkluderer det speedup, frem og tilbagen mellem C og python?
<christoffer> Det burde det gøre, ja!
<Jmaa> Gjorde det nogen stor effekt for hvor hurtigt I kan evaluere?
<christoffer> Jeg tror ikke at der er særligt meget "frem og tilbage"-stads. Det ser ud til at det kode man skriver bliver kastet ret direkte ind i fortolkeren.
<christoffer> Det gjorde en stor forskel for når vi laver 1-ply.
<christoffer> "ply" er hvor mange træk man kigger fremad.
<christoffer> Så kun at kigge på det umiddelbart næste træk er 0-ply, hvilket er det vi har gjort indtil nu
<christoffer> 1-ply var for langsomt. Det tog ca. 6-7 sekunder at evaluere ét træk.
<christoffer> Alexander lavede lidt omskrivninger, så TensorFlow udregnede det hurtigere og fik det ned på ca. 3-4 sekunder *pr. spil*.
<christoffer> Så skrev vi noget af det om til C, og nu er vi så på ca. 2 sekunder pr. spil med 1-ply, hvilket er ret vildt.
<christoffer> Det er så godt at Python-fortolkeren kan udvides med C!
<christoffer> caspervk, kan I optimere jeres bachelorprojekt med et par C-moduler?
<Jmaa> Det er en hel lille sektion til rapporten det der.
<christoffer> Yeah. Kopierer bare det her verbatim ind.

782
test.py
View File

@ -141,6 +141,56 @@ class TestIsMoveValid(unittest.TestCase):
# TODO: More tests for bearing off are needed
def test_bear_off_non_backmost(self):
board = ( 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 1, 1,
0 )
self.assertEqual(Board.is_move_valid(board, 1, 2, (23, 25)), True)
self.assertEqual(Board.is_move_valid(board, 1, 1, (24, 25)), True)
self.assertEqual(Board.is_move_valid(board, 1, 2, (24, 26)), False)
def test_bear_off_quadrant_limits_white(self):
board = ( 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 1,
1, 1, 1, 1, 1, 1,
0 )
self.assertEqual(Board.is_move_valid(board, 1, 2, (23, 25)), False)
self.assertEqual(Board.is_move_valid(board, 1, 1, (24, 25)), False)
def test_bear_off_quadrant_limits_black(self):
board = ( 0,
-1, -1, -1, -1, -1, -1,
-1, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0 )
self.assertEqual(Board.is_move_valid(board, -1, 2, (2, 0)), False)
self.assertEqual(Board.is_move_valid(board, -1, 1, (1, 0)), False)
def test_bear_off_quadrant_limits_white_2(self):
board = ( 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
1, 0, 0, 0, 0, 1,
0 )
self.assertEqual(Board.is_move_valid(board, 1, 1, (24, 25)), True)
def test_bear_off_quadrant_limits_black_2(self):
board = ( 0,
-1, 0, 0, 0, 0, -1,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0 )
self.assertEqual(Board.is_move_valid(board, -1, 1, (1, 0)), True)
class TestNumOfChecker(unittest.TestCase):
def test_simple_1(self):
board = ( 0,
@ -552,372 +602,372 @@ class TestLegalMoves(unittest.TestCase):
self.assertEqual(Board.calculate_legal_states(board, -1, (4,3)), expected_board_set)
# class TestBoardFlip(unittest.TestCase):
# def test_flip_board(self):
# board = (0,
# -14, -1, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 1,
# 0)
# expected_board = ( 0,
# -1, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 1, 14,
# 0 )
# self.assertEqual(Board.flip(board), expected_board)
# def test_flip_board_bar(self):
# board = (2,
# 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0,
# -7)
# expected_board = (7,
# 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0,
# 0, 0, 0, 0, 0, 0,
# -2)
# self.assertEqual(Board.flip(board), expected_board)
# def test_flip_board_extensive(self):
# board = (4,
# -5, -1, 0, 4, 3, 0,
# 0, -1, 0, -5, 0, 0,
# 0, 3, 0, 0, 0, 0,
# 0, 0, 0, -1, 0, 1,
# -2)
# expected_board = (2,
# -1, 0, 1, 0, 0, 0,
# 0, 0, 0, 0, -3, 0,
# 0, 0, 5, 0, 1, 0,
# 0, -3, -4, 0, 1, 5,
# -4)
# self.assertEqual(Board.flip(board), expected_board)
# def test_inverse(self):
# board = (4,
# -5, -1, 0, 4, 3, 0,
# 0, -1, 0, -5, 0, 0,
# 0, 3, 0, 0, 0, 0,
# 0, 0, 0, -1, 0, 1,
# -2)
# self.assertEqual(Board.flip(Board.flip(board)), board)
# def test_tesauro_initial(self):
# board = Board.initial_state
# expected = (1,1,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0.0,
# 0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 1,1,1,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,0,0,
# 0.0,
# 0,
# 1,
# 0
# )
# import numpy as np
# self.assertTrue((Board.board_features_tesauro(board, 1) ==
# np.array(expected).reshape(1, 198)).all())
# def test_tesauro_bars(self):
# board = list(Board.initial_state)
# board[1] = 0
# board[0] = 2
# board[24] = 0
# board[25] = -2
# board = tuple(board)
# expected = (0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1.0,
# 0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 1,1,1,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1.0,
# 0,
# 1,
# 0
# )
# import numpy as np
# self.assertTrue((Board.board_features_tesauro(board, 1) ==
# np.array(expected).reshape(1, 198)).all())
# def test_tesauro_home(self):
# board = list(Board.initial_state)
# board[1] = 0
# board[24] = 0
# board = tuple(board)
# expected = (0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0.0,
# 2,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 1,1,1,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0.0,
# 2,
# 1,
# 0
# )
# import numpy as np
# self.assertTrue((Board.board_features_tesauro(board, 1) ==
# np.array(expected).reshape(1, 198)).all())
# def test_tesauro_black_player(self):
# board = Board.initial_state
# expected = (1,1,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0.0,
# 0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 1,1,1,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,1,1,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 0,0,0,0,
# 1,1,0,0,
# 0.0,
# 0,
# 0,
# 1
# )
# import numpy as np
# self.assertTrue((Board.board_features_tesauro(board, -1) ==
# np.array(expected).reshape(1, 198)).all())
class TestBoardFlip(unittest.TestCase):
def test_flip_board(self):
board = (0,
-14, -1, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 1,
0)
expected_board = ( 0,
-1, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 1, 14,
0 )
self.assertEqual(Board.flip(board), expected_board)
def test_flip_board_bar(self):
board = (2,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
-7)
expected_board = (7,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0,
-2)
self.assertEqual(Board.flip(board), expected_board)
def test_flip_board_extensive(self):
board = (4,
-5, -1, 0, 4, 3, 0,
0, -1, 0, -5, 0, 0,
0, 3, 0, 0, 0, 0,
0, 0, 0, -1, 0, 1,
-2)
expected_board = (2,
-1, 0, 1, 0, 0, 0,
0, 0, 0, 0, -3, 0,
0, 0, 5, 0, 1, 0,
0, -3, -4, 0, 1, 5,
-4)
self.assertEqual(Board.flip(board), expected_board)
def test_inverse(self):
board = (4,
-5, -1, 0, 4, 3, 0,
0, -1, 0, -5, 0, 0,
0, 3, 0, 0, 0, 0,
0, 0, 0, -1, 0, 1,
-2)
self.assertEqual(Board.flip(Board.flip(board)), board)
def test_tesauro_initial(self):
board = Board.initial_state
expected = (1,1,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0.0,
0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
1,1,1,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,0,0,
0.0,
0,
1,
0
)
import numpy as np
self.assertTrue((Board.board_features_tesauro(board, 1) ==
np.array(expected).reshape(1, 198)).all())
def test_tesauro_bars(self):
board = list(Board.initial_state)
board[1] = 0
board[0] = 2
board[24] = 0
board[25] = -2
board = tuple(board)
expected = (0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1.0,
0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
1,1,1,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1.0,
0,
1,
0
)
import numpy as np
self.assertTrue((Board.board_features_tesauro(board, 1) ==
np.array(expected).reshape(1, 198)).all())
def test_tesauro_home(self):
board = list(Board.initial_state)
board[1] = 0
board[24] = 0
board = tuple(board)
expected = (0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0.0,
2,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
1,1,1,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0.0,
2,
1,
0
)
import numpy as np
self.assertTrue((Board.board_features_tesauro(board, 1) ==
np.array(expected).reshape(1, 198)).all())
def test_tesauro_black_player(self):
board = Board.initial_state
expected = (1,1,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0.0,
0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
1,1,1,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,1,1,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
0,0,0,0,
1,1,0,0,
0.0,
0,
0,
1
)
import numpy as np
self.assertTrue((Board.board_features_tesauro(board, -1) ==
np.array(expected).reshape(1, 198)).all())
if __name__ == '__main__':