自己写的函数
dataset.getbatch返回来的np array中不知道为什么含有nan,
采取
np.count_nonzero(batch_x != batch_x)
进行判断,如果有nan则continue,否则训练
for i in range(4000):
batch_x, batch_y= dataset.getbatch(step) NAN_NUMB=np.count_nonzero(batch_x != batch_x) if NAN_NUMB>0: continue
sess.run(train, feed_dict={x: batch_x, y_: batch_y})