ValueError: Попытка преобразовать тензор в тензор и не удалось. Ошибка: аргумент должен быть плотным тензором: - PullRequest
Новая
       46

ValueError: Попытка преобразовать тензор в тензор и не удалось. Ошибка: аргумент должен быть плотным тензором:

2 голосов
/ 15 марта 2019

Когда я вырезаю линию 

tf.reshape(rewards_list, [-1, 25])

Я получаю сообщение об ошибке 

ValueError: Cannot feed value of shape (1, 1, 25) for Tensor 'Placeholder_3:0', which has shape '(?, 25)'

Но когда я оставляю это там, я получаю сообщение об ошибке в заголовке. 

ValueError: Tried to convert 'tensor' to a tensor and failed. Error: Argument must be a dense tensor: [array([[0.4758947]], dtype=float32)] - got shape [1, 1, 1], but wanted [1].

Я не понимаю, что происходит. Как может rewards_list быть обеими этими фигурами? 

observations = tf.placeholder('float32', shape=[None, num_stops]) # Current game states : r[stop], r[next_stop], r[third_stop]
actions = tf.placeholder('int32',shape=[None])  # 0 - num-stops for actions taken
rewards = tf.placeholder('float32',shape=[None])  # +1, -1 with discounts

# Model
Y = tf.layers.dense(observations, 200, activation=tf.nn.relu)
Ylogits = tf.layers.dense(Y, num_stops)

# sample an action from predicted probabilities
sample_op = tf.random.categorical(logits=Ylogits, num_samples=1)


# loss
cross_entropies = tf.losses.softmax_cross_entropy(onehot_labels=tf.one_hot(actions,num_stops), logits=Ylogits)

loss = tf.reduce_sum(rewards * cross_entropies)

# training operation
optimizer = tf.train.RMSPropOptimizer(learning_rate=0.001, decay=.99)
train_op = optimizer.minimize(loss)

visited_stops = []
steps = 0

with tf.Session() as sess:

    sess.run(tf.global_variables_initializer())

    # Start at a random stop, initialize done to false
    current_stop = random.randint(0, len(r) - 1)
    done = False

    # reset everything    
    while not done: # play a game in x steps   

        observations_list = []
        actions_list = []
        rewards_list = []

        # List all stops and their scores
        observation = r[current_stop]

        # Add the stop to a list of non-visited stops if it isn't
        # already there
        if current_stop not in visited_stops:
            visited_stops.append(current_stop)

        # decide where to go
        action = sess.run(sample_op, feed_dict={observations: [observation]})

        # play it, output next state, reward if we got a point, and whether the game is over
        #game_state, reward, done, info = pong_sim.step(action)
        new_stop = int(action)


        reward = r[current_stop][action]

        if len(visited_stops) == num_stops:
            done = True

        if steps >= BATCH_SIZE:
            done = True

        steps += 1

        observations_list.append(observation)
        actions_list.append(action)
        rewards_list.append(reward)







        #rewards_list = np.reshape(rewards, [-1, 25])
        current_stop = new_stop

    #processed_rewards = discount_rewards(rewards, args.gamma)
    #processed_rewards = normalize_rewards(rewards, args.gamma)

    tf.reshape(rewards_list, [-1, 25])

    sess.run(train_op, feed_dict={observations: [observations_list],
                                 actions: [actions_list],
                                 rewards: rewards_list})
...