def __init__(self):
self.tasks = []
self.index = util.Index()
self.vocab = util.Index()
self.ingredients = [self.index.index(k) for k in INGREDIENTS]
self.recipes = {
self.index.index(k): set(self.index.index(vv) for vv in v)
for k, v in RECIPES.items()
}
#print self.recipes
#self.hints = [
# (self.index.index(k), tuple(self.vocab.index(vv) for vv in v))
# for k, v in HINTS
#]
self.hints = []
for k, v in HINTS:
self.hints.append((self.index.index(k), v))
for w in v:
self.vocab.index(w)
self.kind_to_obs = {}
self.obs_to_kind = {}
for k in self.ingredients:
self.kind_to_obs[k] = len(self.kind_to_obs)
self.obs_to_kind[self.kind_to_obs[k]] = k
self.n_obs = (
2 * WINDOW_SIZE * WINDOW_SIZE * len(self.kind_to_obs)
+ len(self.index)
+ 4)
self.n_act = N_ACTIONS
self.n_actions = self.n_act
self.n_features = self.n_obs
self.is_discrete = True
self.max_hint_len = 3
self.n_vocab = len(self.vocab)
self.random = util.next_random()
self.START = START
self.STOP = STOP
self.tasks = []
for i, (goal, steps) in enumerate(self.hints):
self.tasks.append(Minicraft2Task(i, goal, None))
self.n_tasks = len(self.tasks)
self.n_train = len(TRAIN_IDS)
self.n_val = 0
self.n_test = len(TEST_IDS)
self.demos = {}
gflags.DEFINE_boolean("infer_hyp", False, "use hypotheses at test time")
gflags.DEFINE_string("restore", None, "model to restore")
gflags.DEFINE_float("concept_prior", None,
"place a normal prior on concept representations")
gflags.DEFINE_integer(
"adapt_reprs", 100,
"number of representations to sample when doing adaptation")
gflags.DEFINE_integer(
"adapt_samples", 1000,
"number of episodes to spend evaluating sampled representations")
N_EMBED = 64
N_HIDDEN = 64
random = util.next_random()
class Policy(object):
def __init__(self, task):
self.task = task
self.t_state = tf.placeholder(tf.float32, (None, task.n_features))
self.t_action = tf.placeholder(tf.int32, (None, ))
self.t_reward = tf.placeholder(tf.float32, (None, ))
self.t_hint = tf.placeholder(tf.int32, (None, None))
self.t_hint_len = tf.placeholder(tf.int32, (None, ))
self.t_task = tf.placeholder(tf.int32, (None, ))
self.t_last_hyp = tf.placeholder(tf.int32, (None, ), "last_hyp")
self.t_last_hyp_hidden = tf.placeholder(tf.float32,
else:
new_ingredients.append(ing)
ingredients = new_ingredients
specialized = []
for ingredient in ingredients:
if "?" in ingredient:
specialized.append(ingredient.replace("?", variants.pop(0)))
else:
specialized.append(ingredient)
#print goal, specialized
HINTS.append((goal, [START] + specialized + [STOP]))
for ingredient in INGREDIENTS:
HINTS.append((ingredient, [START, ingredient, STOP]))
util.next_random().shuffle(HINTS)
#for x in RECIPES.items():
# print x
#
#print
#for x in HINTS:
# print x
#for i, (k, v) in enumerate(HINTS):
# print i, k, v
#HINTS = [
# ("wood", ["wood"]),
# ("ore", ["ore"]),