def run(self, board: Board):
"""
The model performs random actions until there is sufficient data in replay memory
Then, the model will also train off replay memory after every action
"""
sub = 0
max_score = 0
total = 0
for e in range(self.episodes):
print('Running episode', e)
board.reset_board()
game_over = False
total_score = 0
print('Memories', len(self.memory.memories))
print('Epsilon', self.epsilon)
while not game_over:
current_state = Board.get_game_state(board._blocks)
new_board, new_state = self.predict(board)
reward = 0
if new_board == -1:
game_over = True
else:
reward, game_over = board.do_move(new_board)
total_score += reward
self.memory.remember(
(current_state, new_state, reward, game_over))
print('Score', total_score)
self.train()
self.epsilon -= self.decay
total += total_score
max_score = total_score if total_score > max_score else max_score
if total_score < 100:
sub += 1
# generate summary
print('Max score', max_score)
print('Avg score', total / self.episodes)
print('Sub 100', sub)
self.save()
def run(self, board: Board):
sub100 = 0
max_score = 0
total = 0
for e in range(self.episodes):
print('Iteration: ', e)
board.reset_board()
game_over = False
total_score = 0
#print('Memories', len(self.memory.memories))
#print('Epsilon', self.epsilon)
while not game_over:
current_state = Board.get_game_state(board._blocks)
new_board, new_state = self.predict(board)
reward = 0
if new_board == -1:
game_over = True
else:
reward, game_over = board.do_move(new_board)
total_score += reward
self.memory.remember(
(current_state, new_state, reward, game_over))
print('Score', total_score)
self.train()
self.epsilon -= self.decay
total += total_score
max_score = total_score if total_score > max_score else max_score
if total_score < 100:
sub100 += 1
# generate summary
print('Max score', max_score)
print('Avg score', total / self.episodes)
print('Sub 100', sub100)
self.save()
def play(self, board: Board):
possible_boards = board.get_next()
possible_states = []
for b in possible_boards:
possible_states.append(Board.get_game_state(b))
if len(possible_boards) == 0:
board.reset_board()
return
q_values = []
for s in possible_states:
s = np.reshape(s, (1, -1))
q_values.append(self.model.predict(s)[0])
m = np.argmax(q_values)
best_board = possible_boards[m]
reward, game_over = board.do_move(best_board)
if game_over:
self.score = 0
else:
self.score += reward
