def __init__(self,
epsilon=0.05,
gamma=0.8,
alpha=0.2,
numTraining=900,
extractor=SimpleExtractor(),
**args):
"You can initialize Q-values here..."
args['epsilon'] = epsilon
args['gamma'] = gamma
args['alpha'] = alpha
args['numTraining'] = numTraining
self.featExtractor = extractor
self.index = 0 # This is always Pacman
self.weights = CustomCounter()
self.q_values = CustomCounter()
self.lastAction = None
ReinforcementAgent.__init__(self,
epsilon=epsilon,
gamma=gamma,
alpha=alpha,
numTraining=numTraining)
"*** YOUR CODE HERE ***"
def mat_features(states, extractor=SimpleExtractor(), ftrs=None):
"""Transform a list of states in state matrices"""
if isinstance(states, np.ndarray):
m = list(map(functools.partial(extractor.getMatrixFeatures, features=ftrs), states))
return m
else:
return [extractor.getMatrixFeatures(states, features=ftrs)]
def __init__(self, epsilon=0.1, alpha=0.5, gamma=0.9, discount=0.9):
self.epsilon = epsilon
self.alpha = alpha
self.gamma = gamma
self.discount = discount
self.legalActions = Const.ACTIONS
self.featExtractor = DangerExtractorInstance()
self.featExtractor = SimpleExtractor()
self.weights = [0] * self.featExtractor.featureNum
self.game_state = None
self.train_episodes = 10000
self.env = Env()
def __init__(self, **args):
"You can initialize Q-values here..."
ReinforcementAgent.__init__(self, **args)
self.weights = util.Counter()
#100 episodes of training
self.weights['enemy'] = -10
self.weights['bias'] = -10
self.weights['bullet'] = -200
self.weights['edge'] = -10
self.weights['hitEnemy'] = 10
self.weights['moveFoward'] = 10
self.legalActions = Const.ACTIONS
self.featExtractor = SimpleExtractor()
self.lastAction = 1
def enhancedPacmanFeatures(state, action):
"""
For each state, this function is called with each legal action.
It should return a counter with { <feature name> : <feature value>, ... }
"""
# features = util.Counter()
"*** YOUR CODE HERE ***"
# successor = state.generateSuccessor(0, action)
# foodCount = successor.getFood().count()
# features['foodCount'] = foodCount
#features = neuralDistances(state, action)
featureExtract = SimpleExtractor()
features = featureExtract.getFeatures(state, action)
#print("from enhanced")
#print(features)
#print("features: ",features.items())
# it looks like 'capsule 0' is the problem; its values is a list of values not an int
return features
def __init__(self, epsilon=0.1, alpha=0.5, gamma=0.9, discount=0.9):
self.epsilon = epsilon
self.alpha = alpha
self.gamma = gamma
self.discount = discount
self.legalActions = Const.ACTIONS
self.featExtractor = SimpleExtractor()
self.featExtractor2 = PositionExtractor()
self.weights = util.Counter()
self.game_state = None
self.env = Env()
self.input_num = self.featExtractor2.getFeatureNum()
self.hidden_num = 100
self.output_num = 6
self.W1 = np.random.rand(self.input_num, self.hidden_num)
self.W2 = np.random.rand(self.hidden_num, self.output_num)
import numpy as np
import gym
import matplotlib.pyplot as plt
from tanks import Env
from const import Const
from featureExtractors import DangerExtractor
from featureExtractors import SimpleExtractor
# env = gym.make('CartPole-v0')
train_episodes = 100
learning_rate = 0.01
level_type = "minimal"
game_speed = 1000
env = Env(level_type, game_speed, train_episodes)
gamma = 0.99
featExtractor = SimpleExtractor()
def discount_rewards(r):
""" take 1D float array of rewards and compute discounted reward """
discounted_r = np.zeros_like(r)
running_add = 0
for t in reversed(range(0, r.size)):
running_add = running_add * gamma + r[t]
discounted_r[t] = running_add
return discounted_r
class agent():
def __init__(self, lr, s_size, a_size, h_size):
# These lines established the feed-forward part of the network. The agent takes a state and produces an action.