TensorFlow：评估具有负维度的张量切片

Question

使用具有未指定尺寸的占位符，如何切割矢量并打印它？

x = tf.placeholder(tf.float32, [None, 784])
for ep in range(1000):
    batch_xs, batch_ys = mnist.train.next_batch(100)
    sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
    sl = tf.slice(y, [0,0], [1,784])

sess.run(sl)和sl.eval()会生成InvalidArgumentError消息Shape [-1,784] has negative dimensions。