def test_check_imperfect_solution(self):
n = 4
cols = 10
rows = 5
dg = DataGenerator()
some_instance_np_array = np.array(
[[1, 10, 1], [2, 10, 2], [1, 10, 3], [5, 10, 4], [1, 10, 1]])
solution_checker = SolutionChecker(n, cols, rows)
self.assertEqual(
solution_checker.get_reward(some_instance_np_array),
(10 * 5) / (cols * rows)
)
def test_check_imperfect_solution_count_files_2(self):
n = 4
cols = 10
rows = 5
dg = DataGenerator()
some_instance_np_array = np.array(
[[1, 10, 1], [6, 10, 2], [2, 10, 3], [1, 10, 4], [1, 10, 1]])
solution_checker = SolutionChecker(n, cols, rows)
self.assertEqual(
solution_checker.get_reward(some_instance_np_array, count_tiles=True),
1 / n
)
def perform_simulation(self, state):
'''
Performs the simulation until legal moves are available.
If simulation ends by finding a solution, a root state starting from this simulation is returned
'''
solution_tiles_order = []
depth = 0
simulation_root_state = state # in case simulation ends in solution; these states are the solution
if len(state.tiles) == 0:
print('perform_simulation called with empty tiles')
return ALL_TILES_USED, simulation_root_state, solution_tiles_order
val = self.val
while True:
val += 1
if len(state.tiles) == 0:
print('solution found in simulation')
return ALL_TILES_USED, simulation_root_state, solution_tiles_order
valid_moves = SolutionChecker.get_valid_next_moves(state, state.tiles, val=val, colorful_states=self.colorful_states)
if not valid_moves:
return depth, simulation_root_state, solution_tiles_order
next_random_tile_index = random.randint(0, len(valid_moves) -1)
success, new_board = SolutionChecker.get_next_turn(
state, valid_moves[next_random_tile_index], val, destroy_state=True,
colorful_states=self.colorful_states
)
self.n_tiles_placed += 1
solution_tiles_order.append(valid_moves[next_random_tile_index])
if success == ALL_TILES_USED:
print('grid is full')
# no LFB on grid; probably means grid is full
solution_tiles_order.append(valid_moves[next_random_tile_index])
return ALL_TILES_USED, simulation_root_state, solution_tiles_order
elif success == NO_NEXT_POSITION_TILES_UNUSED:
print('no next position with unused tiles')
return depth, simulation_root_state, solution_tiles_order
elif success == TILE_CANNOT_BE_PLACED:
# cannot place the tile. return depth reached
return depth, simulation_root_state, solution_tiles_order
else:
new_tiles = SolutionChecker.eliminate_pair_tiles(state.tiles, valid_moves[next_random_tile_index])
new_state = State(board=new_board, tiles=new_tiles, parent=state)
new_state.score = -1 # because no choice is performed for sequent actions
state.children.append(new_state)
state = new_state
depth += 1
return depth, simulation_root_state, solution_tiles_order
def test_check_perfect_solution(self):
n = 20
w = 40
h = 40
dg = DataGenerator()
some_instance_visual = dg.gen_instance_visual(n, w, h)
perfect_bin_configuration = sorted(some_instance_visual, key=lambda x: (x[2][0], x[2][1]))
some_instance_np_array = dg._transform_instance_visual_to_np_array(some_instance_visual)
solution_checker = SolutionChecker(n, h, w)
self.assertEqual(
solution_checker.get_reward(np.array(perfect_bin_configuration)),
0
)
def test_check_imperfect_solution_count_tiles(self):
n = 4
cols = 10
rows = 5
dg = DataGenerator()
some_instance_visual = dg.gen_instance_visual(n, cols, rows)
# NOTE: first bin always repeated
some_instance_np_array = np.array(
[[1, 10, 1], [2, 10, 2], [1, 10, 3], [5, 10, 4], [1, 10, 1]])
solution_checker = SolutionChecker(n, cols, rows)
self.assertEqual(
solution_checker.get_reward(some_instance_np_array, count_tiles=True),
1 / n
)
def test_get_next_turn_only_success(self):
success, new_board, = SolutionChecker.get_next_turn(
self.state, self.tile, val=1, get_only_success=True, destroy_state=True
)
# assert board is destroyed i.e. equal to new_board
self.assertIsNone(new_board)
def test_get_next_lfb_on_grid_5(self):
state = np.array([
[1, 1, 1, 1],
[1, 1, 1, 1],
[0, 0, 0, 1]
])
res = SolutionChecker.get_next_lfb_on_grid(state)
self.assertEqual(res, (2, 0))
def get_next_turn(self, state, tile, val=1):
new_board = np.copy(state.board)
next_position = SolutionChecker.get_next_lfb_on_grid(new_board)
# one without the other should not be possible
if not next_position and len(state.tiles) == 0:
print('solution found!')
return ALL_TILES_USED, None
elif not next_position:
return NO_NEXT_POSITION_TILES_UNUSED, None
success, new_board = SolutionChecker.place_element_on_grid_given_grid(
tile, SolutionChecker.get_next_lfb_on_grid(new_board), val,
new_board, get_cols(new_board), get_rows(new_board))
if not success:
# cannot place the tile. this branch will not be considered
return TILE_CANNOT_BE_PLACED, None
return True, new_board
def test_get_possible_tile_actions_given_grid(self):
grid = np.array([
[1, 1, 1, 0, 0],
[1, 1, 1, 0, 0],
[1, 1, 1, 0, 0],
])
tiles = [[1, 2], [2, 1], [4, 5], [1, 1], [1, 1], [3, 2], [3, 3], [2, 3]]
possible_tiles_to_place = SolutionChecker.get_possible_tile_actions_given_grid(grid, tiles)
self.assertEqual(possible_tiles_to_place, [[1, 2], [2, 1], [1, 1], [1, 1], [3, 2]])
def test_get_valid_next_moves_1(self):
self.board = np.array([
[1, 1, 1, 1],
[1, 1, 1, 0],
[0, 0, 0, 0]
])
self.tiles = [(2, 1), (1, 2)]
self.state = State(self.board, self.tiles)
next_moves = SolutionChecker.get_valid_next_moves(self.state, self.tiles)
self.assertEqual(next_moves, [(2, 1)])
def __init__(self, tiles, board, strategy='max_depth'):
self.initial_tiles, self.initial_board = tiles, board
self.state = State(self.initial_board, self.initial_tiles)
self.solution_checker = SolutionChecker(len(tiles), get_rows(board), get_cols(board))
self.strategy=strategy
self.n_tiles_placed = 0
self.solution_tiles_order = []
# for MCTS vis purposes
self.colorful_states = True
def test_get_next_turn_destroy_board(self):
success, new_board, = SolutionChecker.get_next_turn(
self.state, self.tile, val=1, get_only_success=False, destroy_state=True
)
self.assertEqual(success, True)
# assert board is destroyed i.e. equal to new_board
np.testing.assert_array_equal(
self.state.board,
new_board
)
def test_get_valid_actions_indexed_given_grid(self):
grid = np.array([
[1, 1, 1, 0, 0],
[1, 1, 1, 0, 0],
[1, 1, 1, 0, 0],
])
tiles = [[1, 2], [2, 1], [0, 0], [4, 5], [1, 1], [1, 1], [3, 2], [3, 3], [2, 3]]
possible_tiles_indexes = SolutionChecker.get_valid_tile_actions_indexes_given_grid(
grid, tiles)
self.assertEqual(possible_tiles_indexes, [1, 1, 0, 0, 1, 1, 1, 0, 0])
def test_get_valid_next_moves_2(self):
self.board = np.array([
[1, 1, 1, 1],
[1, 1, 1, 0],
[0, 0, 0, 0]
])
tiles_list = [(2, 1), (1, 2), (3, 3), (4, 4), (1, 1)]
self.tiles = sorted(tiles_list, key=lambda x: (x[1], x[0]))
self.state = State(self.board, self.tiles)
next_moves = SolutionChecker.get_valid_next_moves(self.state, self.tiles)
self.assertEqual(next_moves, [(1, 1), (2, 1)])
def initialize(self, height, width, n_tiles):
Game.__init__(self)
self.height = height
self.width = width
self.n_tiles = n_tiles
dg = DataGenerator()
self.tiles, solution = dg.gen_instance(n_tiles, width, height)
# print(self.tiles)
_tiles_ints = SolutionChecker.get_tiles_with_orientation(self.tiles)
self._base_board = Board(height, width, _tiles_ints, n_tiles)
def __init__(self, tiles, board):
self.Qsa = {} # stores Q values for s,a (as defined in the paper)
self.Nsa = {} # stores #times edge s,a was visited
self.Ns = {} # stores #times board s was visited
self.Ps = {} # stores initial policy (returned by neural net)
self.Es = {} # stores game.getGameEnded ended for board s
self.Vs = {} # stores game.getValidMoves for board s
self.initial_tiles, self.initial_board = tiles, board
self.state = State(self.initial_board, self.initial_tiles)
self.solution_checker = SolutionChecker(len(tiles), get_rows(board),
get_cols(board))
def test_bin_outside_border(self):
n = 20
h = 50
w = 50
solution_checker = SolutionChecker(n, h, w)
#
# 11 -------------
# | |
# | | |
# | | |
# -----------------------|
# 40
solution_checker.LFBs = SortedKeyList([], key=lambda x: (x[1], x[0]))
solution_checker.LFBs.add((40, 11))
_bin = (10, 10)
self.assertFalse(solution_checker.is_bin_outside_borders(_bin))
_bin = (12, 10)
self.assertTrue(solution_checker.is_bin_outside_borders(_bin))
def build_reward(self): # reorder input % tour and return tour length (euclidean distance)
self.permutations = tf.stack( # this just creates a vectors with repeating idxs so we can gather later
[
tf.tile(tf.expand_dims(tf.range(self.batch_size,dtype=tf.int32), 1), [1, self.n+1]),
self.tour
], 2
)
if self.is_training==True:
self.ordered_input_ = tf.gather_nd(self.input_,self.permutations)
else:
self.ordered_input_ = tf.gather_nd(tf.tile(self.input_,[self.batch_size,1,1]),self.permutations)
solution_checker = SolutionChecker(self.n, self.w, self.h)
sess = tf.Session()
rewards = tf.py_func(
solution_checker.get_rewards, [self.ordered_input_, self.count_non_placed_tiles, self.combinatorial_reward], tf.float32)
self.reward = rewards
tf.summary.scalar('reward_mean', tf.reduce_mean(rewards))
def get_only_random_predictions(examples):
total_random_correct = 0
with open("tensorboard/random_predictions.csv", 'w') as f:
for i in range(2755):
total_count = 0
total_random_correct = 0
for example in examples:
expected = np.argmax(example[2])
expected_tile = example[1][expected]
random_prediction = random.randint(
0,
SolutionChecker.get_n_nonplaced_tiles(example[1]) - 1)
if np.array_equal(expected_tile,
example[1][random_prediction]):
total_random_correct += 1
total_count += 1
accuracy = (total_random_correct / total_count) * 100
f.write(f'{accuracy}\n')
print(accuracy)
return
def test_get_next_turn(self):
success, new_board, = SolutionChecker.get_next_turn(
self.state, self.tile, val=1, get_only_success=False, destroy_state=False
)
self.assertEqual(success, True)
np.testing.assert_array_equal(
new_board,
np.array([
[1, 1, 1, 1],
[1, 1, 1, 1],
[0, 0, 0, 1]
])
)
# assert board is not destroyed
np.testing.assert_array_equal(
self.state.board,
np.array([
[1, 1, 1, 1],
[1, 1, 1, 0],
[0, 0, 0, 0]
])
)
def main(options):
#N_TILES = 8
#HEIGHT = 8
#WIDTH = 8
HEIGHT = 25
WIDTH = 25
N_TILES = 50
for (WIDTH, HEIGHT) in [(30, 30)]:
#for N_TILES in [50]:
SCALAR_TILES = True
predict_move_index = True
g = Game(HEIGHT, WIDTH, N_TILES)
dg = DataGenerator(WIDTH, HEIGHT)
# from alpha.binpack.tensorflow.NNet import NNetWrapper as nn
from alpha.binpack.keras.NNet import NNetWrapper as nn
nnet = nn(g, scalar_tiles=SCALAR_TILES)
n_tiles, height, width = N_TILES, HEIGHT, WIDTH
if options['load_model']:
nnet.load_checkpoint()
else:
# place tiles one by one
# generate pair x and y where x is stack of state + tiles
print('Preparing examples')
N_EXAMPLES = 1000
examples = get_n_examples(N_EXAMPLES,
width,
height,
n_tiles,
dg,
scalar_tiles=SCALAR_TILES)
if options['load_examples']:
with open(
f'models/train_examples_{N_TILES}_{HEIGHT}_{WIDTH}.pickle',
'rb') as f:
train_examples = pickle.load(f)
else:
train_examples = get_examples(examples,
N_TILES,
height,
width,
dg,
return_binary_mask=True,
predict_move_index=True,
scalar_tiles=SCALAR_TILES)
with open(
f'models/train_examples_{N_TILES}_{HEIGHT}_{WIDTH}.pickle',
'wb') as f:
pickle.dump(train_examples, f)
if options['load_val_examples']:
with open(
f'models/validation_examples_{N_TILES}_{HEIGHT}_{WIDTH}.pickle',
'rb') as f:
validation_examples = pickle.load(f)
else:
N_EXAMPLES = 100
validation_examples = get_n_examples(N_EXAMPLES,
width,
height,
n_tiles,
dg,
scalar_tiles=SCALAR_TILES)
validation_examples = get_examples(validation_examples,
N_TILES,
height,
width,
dg,
from_file=False,
return_binary_mask=True,
predict_move_index=True,
scalar_tiles=SCALAR_TILES,
shuffle_tiles_times=1)
validation_examples = get_val_examples_not_in_overlap(
train_examples, validation_examples)
with open(
f'models/validation_examples_{N_TILES}_{HEIGHT}_{WIDTH}.pickle',
'wb') as f:
pickle.dump(validation_examples, f)
if not options['load_model']:
print(f'In total: {len(train_examples)} train examples')
print(f'In total: {len(validation_examples)} validation examples')
if options['load_model']:
nnet.load_checkpoint()
else:
nnet.train(train_examples, validation_examples)
nnet.save_checkpoint()
np.set_printoptions(
formatter={'float': lambda x: "{0:0.2f}".format(x)}, linewidth=115)
total_correct = 0
total_random_correct = 0
total_max_col_correct = 0
total_max_row_correct = 0
total_biggest_tile_correct = 0
total_smallest_tile_correct = 0
total_most_common_tile_correct = 0
total_count = 0
n_empty_tiles_with_fails = [0] * (N_TILES + 1)
if False:
# overlap was 39/1868 between val and train
get_overlap_between_examples(train_examples, validation_examples)
return
if False:
get_only_random_predictions(validation_examples)
return
for example in validation_examples:
prediction = nnet.predict([example[0], example[1]])
random_prediction = random.randint(
0,
SolutionChecker.get_n_nonplaced_tiles(example[1]) - 1)
output_str = ''
if VISUALIZE_PREDICTIONS:
output_str += f'----------------------------------------------------------'
output_str += '\n'
if predict_move_index:
_prediction = prediction
if VISUALIZE_PREDICTIONS:
output_str += f'{prediction}\n'
max_index = np.argmax(prediction)
_prediction_index = max_index
if SCALAR_TILES:
expected = np.argmax(example[2])
else:
expected = np.argmax(example[1])
# if not scalar_tiles:
# print('grid state')
# print(example[0][:, :, 0])
# print(state_to_tiles_dims(example[0], dg))
# print('expected')
if SCALAR_TILES:
expected_tile = example[1][expected]
prediction_tile = example[1][_prediction_index]
if VISUALIZE_PREDICTIONS:
output_str += f'{example[1].tolist()}\n'
else:
expected_tile = dg.get_matrix_tile_dims(
example[0][:, :, expected + 1])
prediction_tile = dg.get_matrix_tile_dims(
example[0][:, :, _prediction_index + 1])
# print(expected, expected_tile)
#print('prediction')
# print(_prediction)
#print(_prediction_index, prediction_tile)
if VISUALIZE_PREDICTIONS:
output_str += f'{example[0]}\n'
output_str += f'expected: {expected_tile}, got: {prediction_tile}'
output_str += f'random: {example[1][random_prediction]}'
if SCALAR_TILES:
widest_tile = example[1][0]
highest_tile = example[1][0]
biggest_tile = example[1][0]
smallest_tile = example[1][0]
counter = Counter()
for i, tile in enumerate(example[1]):
if tile[1] > widest_tile[1]:
widest_tile = tile
elif tile[1] == widest_tile[1]:
if tile[0] > widest_tile[0]:
widest_tile = tile
if tile[0] > highest_tile[0]:
highest_tile = tile
elif tile[0] == highest_tile[0]:
if tile[1] > highest_tile[1]:
highest_tile = tile
if tile[1] * tile[0] > (biggest_tile[1] *
biggest_tile[0]):
biggest_tile = tile
if tile[1] * tile[0] < (smallest_tile[1] *
smallest_tile[0]):
smallest_tile = tile
counter[tuple(tile.tolist())] += 1
if np.array_equal(expected_tile, widest_tile):
total_max_col_correct += 1
if np.array_equal(expected_tile, highest_tile):
total_max_row_correct += 1
if np.array_equal(expected_tile, biggest_tile):
total_biggest_tile_correct += 1
if np.array_equal(expected_tile, smallest_tile):
total_smallest_tile_correct += 1
most_common_tile = np.array(counter.most_common(1)[0][0])
if np.array_equal(most_common_tile, np.array([0, 0])):
most_common_tile = np.array(
counter.most_common(2)[1][0])
if np.array_equal(expected_tile, most_common_tile):
total_most_common_tile_correct += 1
if VISUALIZE_PREDICTIONS:
output_str += f'max_tile: {widest_tile}\n'
if np.array_equal(expected_tile, prediction_tile):
total_correct += 1
# visualize predictions
#if not np.array_equal(expected_tile, widest_tile) and VISUALIZE_PREDICTIONS:
# print(output_str)
if VISUALIZE_PREDICTIONS:
print(output_str)
else:
n_empty_tiles_with_fails[count_n_of_non_placed_tiles(
example[1]) // 2] += 1
# print(example[1][random_prediction])
if np.array_equal(expected_tile,
example[1][random_prediction]):
total_random_correct += 1
else:
if expected_tile == prediction_tile:
total_correct += 1
else:
n_empty_tiles_with_fails[count_n_of_non_placed_tiles(
state_to_tiles_dims(example[0], dg)) // 2] += 1
total_count += 1
else:
_prediction = np.reshape(prediction, (width, height))
_prediction = get_prediction_masked(_prediction,
example[0][:, :, 0])
expected = np.reshape(example[1], (width, height))
if VISUALIZE_PREDICTIONS:
# visualize predictions
# print('-' * 50)
# print('grid state')
# print(example[0][:, :, 0])
# print('expected')
# print(expected)
# print('prediction')
# print(_prediction)
pass
if predict_move_index:
with open(f"nn_results/custom_{N_TILES}_{width}_{height}.csv",
'w') as f:
output_str = (
f'{width}-{height}\n'
f'In total guessed;{100*(total_correct/ total_count)}; {total_correct}/{total_count}\n'
f'Random baseline; {100*(total_random_correct/total_count)}\n'
f'Max col tile baseline; {100*(total_max_col_correct/total_count)}\n'
f'Max row tile baseline; {100*(total_max_row_correct/total_count)}\n'
f'Most common tile baseline; {100*(total_most_common_tile_correct/total_count)}\n'
f'Max area tile baseline; {100*(total_biggest_tile_correct/total_count)}\n'
f'Min area tile baseline; {100*(total_smallest_tile_correct/total_count)}'
)
f.write(output_str)
print(output_str)
print(n_empty_tiles_with_fails)
print('-' * 100)
if not PREDICT_FULL_EXAMPLES:
# return
continue
tiles_left = []
for example in examples:
if SCALAR_TILES:
state, tiles, _ = example
tiles = get_tiles_with_orientation(tiles.tolist())
else:
tiles, _ = example
# tiles = dg.get_matrix_tile_dims(tiles)
grid = np.zeros((width, height))
tiles_left.append(
play_using_prediction(nnet,
width,
height,
tiles,
grid,
N_TILES,
dg,
predict_move_index,
scalar_tiles=SCALAR_TILES))
# [0, 6, 4, 2, 2, 2, 0, 4, 4, 8, 6, 2, 2, 6, 6, 8, 6, 4, 4, 4]
print(tiles_left)
print(np.sum(tiles_left) / len(tiles_left))
def play_using_prediction(nnet,
width,
height,
tiles,
grid,
n_tiles,
dg,
predict_move_index=False,
verbose=False,
scalar_tiles=False):
if scalar_tiles:
tiles = tiles
else:
tiles = state_to_tiles_dims(tiles, dg)
while True:
tiles_left = len(tiles)
if tiles_left == 0:
print(
f"Success: game ended with {tiles_left / ORIENTATIONS} tiles left unplaced."
)
return 0
_tiles_ints = SolutionChecker.get_possible_tile_actions_given_grid(
grid, tiles)
if len(_tiles_ints) == 0 and tiles_left:
print(
f"game ended with {tiles_left / ORIENTATIONS} tiles left unplaced."
)
return tiles_left / ORIENTATIONS
np.random.shuffle(_tiles_ints)
if scalar_tiles:
_tiles = tiles_to_np_array(
SolutionChecker.pad_tiles_with_zero_scalars(
_tiles_ints, ORIENTATIONS * n_tiles - len(_tiles_ints)))
# state = state.squeeze()
prediction = nnet.predict([grid, tiles_to_np_array(_tiles)])
else:
tiles_in_matrix_shape = dg._transform_instance_to_matrix(
_tiles_ints, only_one_orientation=True)
tiles_in_matrix_shape = pad_tiles_with_zero_matrices(
tiles_in_matrix_shape,
n_tiles * ORIENTATIONS - tiles_in_matrix_shape.shape[2], width,
height)
state = np.dstack((np.expand_dims(grid,
axis=2), tiles_in_matrix_shape))
prediction = nnet.predict(state)
if not predict_move_index:
if len(prediction) == 2: # if we are also predicting v
prediction, v = prediction
prediction = np.reshape(prediction, (width, height))
# get the probability matrix
prediction = get_prediction_masked(prediction, state[:, :, 0])
if verbose:
print('-' * 50)
print(grid, prediction)
print(tiles)
if scalar_tiles:
_ttiles = tiles
else:
_ttiles = state_to_tiles_dims(state, dg)
solution_tile = get_best_tile_by_prediction(
grid,
_ttiles,
prediction,
dg,
predict_move_index=predict_move_index)
if verbose:
print(solution_tile)
if solution_tile is None:
print(
f"game ended with {tiles_left / ORIENTATIONS} tiles left unplaced."
)
return tiles_left / ORIENTATIONS
success, grid = SolutionChecker.place_element_on_grid_given_grid(
solution_tile[0],
solution_tile[1],
val=1,
grid=grid,
cols=width,
rows=height)
if not success:
print(
f"game ended with {tiles_left / ORIENTATIONS} tiles left unplaced."
)
return tiles_left / ORIENTATIONS
if scalar_tiles:
tiles = [tuple(x) for x in tiles]
tiles = SolutionChecker.eliminate_pair_tiles(tiles, solution_tile[0])
return 0
def get_best_tile_by_prediction(grid,
tiles,
prediction,
dg,
predict_move_index=True):
'''
1. mask invalid moves
2. renormalize the probability distribution
3. for each tile sum up which would be the probability of placing that tile in LFB
4. return the new grid
Returns the tile which is the best in format [(rows, cols), position_to_place]
'''
next_lfb = SolutionChecker.get_next_lfb_on_grid(grid)
max_probability = -math.inf
max_index = 0
best_tile = None
rows, cols = grid.shape
tile_probabilities = defaultdict(int)
tile_counts = defaultdict(int)
tiles_to_iterate_on = tiles
SOFTMAX = False
best_tile_individual = None
best_prediction = 0
for i, tile in enumerate(tiles_to_iterate_on):
tile = tuple(tile)
success, _ = SolutionChecker.place_element_on_grid_given_grid(
tile,
next_lfb,
val=1,
grid=grid,
cols=cols,
rows=rows,
get_only_success=True)
if not success:
continue
if predict_move_index and SOFTMAX:
'''
The best tile is predicted by taking the sum of the tiles
with the same (width, height)
'''
if tuple(tile) == (0, 0):
continue
tile_probabilities[tile] += prediction[i]
tile_counts[tile] += 1
elif predict_move_index:
if tuple(tile) == (0, 0):
continue
if prediction[i] > best_prediction:
best_tile_individual = tile
best_prediction = prediction[i]
# if tile[0] * tile[1] > best_prediction:
# best_tile_individual = tile
# best_prediction = tile[0] * tile[1]
else:
probability = np.sum(prediction[next_lfb[0]:next_lfb[0] + tile[0],
next_lfb[1]:next_lfb[1] + tile[1]])
# scale with area
# probability = probability / (tile[0] * tile[1])
if probability > max_probability:
max_index = i
max_probability = probability
best_tile = [tile, next_lfb]
if predict_move_index and SOFTMAX:
max_tile = None
max_val = -math.inf
for k in tile_probabilities.keys():
v = tile_probabilities[k] / tile_counts[k]
if True:
if v > max_val:
max_tile = k
max_val = v
else:
if k[0] * k[1] > max_val:
max_tile = k
max_val = k[0] * k[1]
if max_tile:
best_tile = [max_tile, next_lfb]
elif predict_move_index:
best_tile = [best_tile_individual, next_lfb]
# print(tile_probabilities)
# print(tile_counts)
if not best_tile:
print('No valid tile placement found')
return best_tile
def predict(self, temp=1, N=3000):
initial_state = self.state
state = self.state
available_tiles = state.tiles
prev_state = state
solution_found = False
depth = 0
self.val = 1
while len(state.tiles):
tile_placed = False
states = []
print(len(state.tiles))
best_score = 0
for i, tile in enumerate(state.tiles):
success, new_board = SolutionChecker.get_next_turn(state, tile, self.val, destroy_state=False)
self.n_tiles_placed += 1
if success == ALL_TILES_USED:
print('solution found!')
solution_found = True
return initial_state, depth, solution_found
if success == TILE_CANNOT_BE_PLACED:
# cannot place the tile. this branch will not be considered
states.append(None)
continue
else:
tile_placed = True
new_tiles = SolutionChecker.eliminate_pair_tiles(state.tiles, tile)
new_state = State(
board=new_board, tiles=new_tiles, parent=state)
state.children.append(new_state)
simulation_result, solution_tiles_order = self.perform_simulations(new_state, N=N, record_simulations=False)
if simulation_result == ALL_TILES_USED:
print('solution found in simulation!')
print(tile)
solution_found = True
# update scores of states found in simulation
new_state.score = len(state.tiles) / ORIENTATIONS - 1
_state = new_state.children[0]
while _state.children:
_state.score = (len(_state.tiles) / ORIENTATIONS) - 1
_state = _state.children[0]
if state.tile_placed:
self.solution_tiles_order.extend([state.tile_placed] + [tile] + solution_tiles_order)
else:
self.solution_tiles_order.extend([tile] + solution_tiles_order)
return initial_state, depth, solution_found
new_state.score = simulation_result
if new_state.score > best_score:
best_tile = tile
best_score = new_state.score
new_state.tile_placed = tile
state.solution_tiles_order.append(tile)
states.append(new_state)
if not tile_placed:
# no tile was placed, it's a dead end; end game
return initial_state, depth, solution_found
# PERFORMANCE:
# for visualization this can be changed
# all tiles will be 1 inside the frame for performance reasons
self.val += 1
depth += 1
best_action = get_max_index(states)
prev_state = state
new_state = states[best_action]
print(best_tile, prev_state.tile_placed)
if prev_state.tile_placed:
self.solution_tiles_order.append(prev_state.tile_placed)
state = new_state
print('Solution found!')
solution_found = True
return initial_state, depth, solution_found
def test_eliminate_pair_tiles_3(self):
tiles_list = [(1, 1), (2, 1), (1, 1), (1, 1), (2, 1), (1, 1)]
new_tiles = SolutionChecker.eliminate_pair_tiles(tiles_list, tile_to_remove=(1, 1))
self.assertEqual(new_tiles, [(2, 1), (1, 1), (2, 1), (1, 1)])
def test_eliminate_pair_tiles_when_pair_is_not_present(self):
tiles_list = [(1, 8), (2, 1), (1, 1), (1, 1), (2, 1), (1, 1)]
with self.assertRaises(ValueError):
SolutionChecker.eliminate_pair_tiles(tiles_list, tile_to_remove=(1, 8))
def test_pad(self):
tiles = [[1, 2], [2, 1], [4, 5]]
padded_tiles = SolutionChecker.pad_tiles_with_zero_scalars(tiles, 2)
self.assertEqual(len(padded_tiles), 5)
self.assertEqual([[1, 2], [2, 1], [4, 5], [0, 0], [0, 0]], padded_tiles)
for i in tqdm(range(10)): # test instance
seed_ = 1+random.randint(0, 10000)
dg = DataGenerator() # Create Data Generator
input_batch = dg.train_batch(
1, actor.n, actor.w, actor.h, actor.dimension,
seed=i,
freeze_first_batch=config.freeze_first_batch
)
feed = {actor.input_: input_batch} # Get feed dict
tour, reward = sess.run([actor.tour, actor.reward], feed_dict=feed) # sample tours
j = np.argmin(reward) # find best solution
best_permutation = tour[j][:-1]
predictions_length.append(reward[j])
solution_checker = SolutionChecker(actor.n, actor.w, actor.h)
# TODO: find how this is called in numpy (sort by index)
bins = []
for el in best_permutation:
bins.append(input_batch[0][el])
solution_checker.get_reward(bins)
grid = solution_checker.grid
print('reward',reward[j])
solution_checker.visualize_grid()
#dataset.visualize_2D_trip(input_batch[0][best_permutation])
#dataset.visualize_sampling(tour)
# dataset.visualize_2D_trip(opt_tour)
def get_examples(given_examples,
n_tiles,
height,
width,
dg,
from_file=False,
return_binary_mask=False,
predict_v=False,
predict_move_index=True,
scalar_tiles=False,
shuffle_tiles_times=20):
examples = []
for i, _example in enumerate(given_examples):
# print(f'{i}/{len(given_examples)}')
if scalar_tiles:
state, tiles, solution = _example
state = state.squeeze()
else:
state, solution = _example
grid = np.zeros([height, width])
for solution_index, solution_tile in enumerate(solution):
solution_copy = np.copy(solution)
solution_order = np.array(solution_copy[solution_index:])
solution_tile_dims = solution_tile[:2]
orig_solution_order = solution_order
for i in range(shuffle_tiles_times): # tile permutations
solution_order = np.copy(orig_solution_order)
_tiles_ints = [x[:ORIENTATIONS] for x in solution_order]
_tiles_ints = get_tiles_with_orientation(_tiles_ints)
_tiles_ints = SolutionChecker.get_possible_tile_actions_given_grid(
grid, _tiles_ints)
np.random.shuffle(_tiles_ints)
if scalar_tiles:
tiles = tiles_to_np_array(
SolutionChecker.pad_tiles_with_zero_scalars(
_tiles_ints,
ORIENTATIONS * n_tiles - len(_tiles_ints)))
else:
tiles = dg._transform_instance_to_matrix(
_tiles_ints, only_one_orientation=True)
tiles = pad_tiles_with_zero_matrices(
tiles, ORIENTATIONS * n_tiles - tiles.shape[2], width,
height)
state = np.dstack((np.expand_dims(grid, axis=2), tiles))
pi = solution_to_solution_matrix(
solution_tile,
cols=width,
rows=height,
return_binary_mask=False).flatten()
# v = N_TILES - solution_index
v = 1
if solution_index == len(solution) - 1:
continue
if predict_move_index:
n_possible_tiles = SolutionChecker.get_n_nonplaced_tiles(
_tiles_ints)
if n_possible_tiles == 1: # if only one action-tile placement is possible
pass
else:
_tiles_ints = SolutionChecker.np_array_to_tiles(
_tiles_ints)
solution_index = _tiles_ints.index(solution_tile_dims)
if scalar_tiles:
if INDIVIDUAL_TILES:
split_tiles = np.array(tiles)
split_tiles = np.split(split_tiles,
split_tiles.shape[0])
else:
split_tiles = tiles
example = [
grid.copy(), split_tiles,
one_hot_encode(_tiles_ints, solution_tile_dims,
len(tiles))
]
else:
example = [
state,
one_hot_encode(_tiles_ints, solution_tile_dims,
state.shape[2] - 1)
]
examples.append(example)
# print(_tiles_ints, solution_tile_dims, one_hot_encode(_tiles_ints, solution_tile_dims, state))
else:
example = [state, pi]
if predict_v:
example.append(v)
examples.append(example)
success, grid = SolutionChecker.place_element_on_grid_given_grid(
solution_tile[:ORIENTATIONS],
solution_tile[2],
val=1,
grid=grid,
cols=width,
rows=height)
return examples
def test_get_next_lfb_on_grid_3(self):
state = np.array([[1, 1, 1], [1, 1, 1]])
res = SolutionChecker.get_next_lfb_on_grid(state)
self.assertIsNone(res)
