NameError：在封閉範圍內賦值之前引用的自由變量'ar'

Question

我在使用tensorflow的簡單代碼中使用tf.while_loop。它給了我這個錯誤

ar = tf.cond(tf.less(arg[i], 1), lambda: tf.concat([ar,[1.0]], axis = 0), lambda: tf.concat([ar,[0.0]], axis = 0))

NameError: free variable 'ar' referenced before assignment in enclosing scope

代碼::

a = tf.constant([[[1,2],[4,3]],[[5,6],[8,7]]]) 

arg = tf.argmax(a, axis =2) 

arg = tf.cast(tf.reshape(arg,[4]), tf.int32) 



i =tf.constant(0) 
c = lambda i: tf.less(i,4) 

def body(i): 

    ar = tf.cond(tf.less(arg[i], 1), lambda: tf.concat([ar,[1.0]], axis = 0), lambda: tf.concat([ar,[0.0]], axis = 0)) 

    print(i) 

    i = tf.add(i,1) 

    return i 

ar = tf.constant([1]) 

r = tf.while_loop(c,body,[i]) 

sess =tf.InteractiveSession() 
sess.run(tf.global_variables_initializer()) 

sess.run(r) 

print(sess.run(ar)) 

Answer 1

在body功能添加

global ar 

線以上

ar = tf.cond(tf.less(arg[i], 1), lambda: tf.concat([ar,[1.0]], axis = 0), lambda: tf.concat([ar,[0.0]], axis = 0))