def run():
agentDQN = DeepQNetworkAgent()
current_episode = 0
max_episode = 401
for current_episode in range(max_episode):
gameInit(agentDQN, showGui=showGui) #change boolean to display UI
#Decrease exploration epsilon after every episode
agentDQN.explorationEpsilonDecay()
while not snakeML.gameEnded:
state_1 = np.reshape(snakeML.getInputLayer(), (1, SIZE_INPUT_LAYER))
if np.random.random_sample() < agentDQN.explorationEpsilon:
action = np.random.randint(0, 3)
else:
prediction = agentDQN.model.predict(np.reshape(state_1, (1, SIZE_INPUT_LAYER)))
action = np.argmax(prediction)
foodAte, gameEnded = snakeML.nextStep(action)
reward = agentDQN.setReward(foodAte, gameEnded)
state_2 = np.reshape(snakeML.getInputLayer(), (1, SIZE_INPUT_LAYER))
agentDQN.train_short_memory(state_1, action, reward, state_2, gameEnded)
agentDQN.remember(state_1, action, reward, state_2, gameEnded)
agentDQN.replay_memory()
print(f"Game {current_episode}, Score={snakeML.getSnakeLength()-1}")
snakeML.exit()
if((current_episode % 10)==0):
agentDQN.model.save(f"model/episode_{current_episode:03d}.hdf5")
def showSnake(modelName):
model = tf.keras.models.load_model(modelName)
snakeML.newGameML()
snakeML.startGameML()
game_ended = False
num_steps = 0
while not game_ended:
input_layer = snakeML.getInputLayer()
output = model.predict(np.matrix(input_layer))
# predict output
output = np.argmax(output)
len, game_ended = snakeML.nextStep(output)
num_steps += 1
snakeML.exit()
weights = reshapeWeights(population[j], input_layer_units,
hidden_layers_units, output_layer_units)
model.set_weights(weights)
snakeML.newGameML()
snakeML.startGameML(showGui=False)
game_ended = False
num_steps = 0
while not game_ended and (num_steps < max_steps):
input_layer = snakeML.getInputLayer()
output = model.predict(np.matrix(input_layer))
# predict output
output = np.argmax(output)
len[j], game_ended = snakeML.nextStep(output)
num_steps += 1
fitness[j] = calcFitness(len[j], num_steps)
snakeML.exit()
updateCanvas(txtbox_training_id, i, j, fitness[j], len[j])
best_person = np.argmax(fitness)
print(
f"Generation {i}, Top fitness = {fitness[best_person]}, Length = {len[j]}"
)
updateCanvas(txtbox_trained_id, i, "Null", fitness[best_person], len[j])
weights = reshapeWeights(population[best_person], input_layer_units,
hidden_layers_units, output_layer_units)
model.set_weights(weights)
model.save(f"model/model_generation_{i}_best_fitness.h5")
print(f"Generation {i} best model saved.")
print("Creating next generation. This might take a while.")
population = GeneticAlgo.getNextGeneration(population, fitness)
df = pd.DataFrame(population)