tesnsorflow unsorted segment mean 使用注意事项

 简单介绍

 

输出会根据对应id大小排序

seg = tf.constant([1,1,0,0])
data = tf.constant([.3,.4,11,11])
sess.run(tf.unsorted_segment_mean(data,seg,4))
>>array([11.        ,  0.35000002,  0.        ,  0.        ]

sess.run(tf.unsorted_segment_mean(data,seg,2))
>>array([11.        ,  0.35000002], dtype=float32)

即使非连续的也无所谓，也会排序，由于只能显示两个id5对应的被截掉了 

seg = tf.constant([5,5,0,0])
data = tf.constant([.3,.4,11,11])
sess.run(tf.unsorted_segment_mean(data,seg,2))
Out[45]: array([11.,  0.], dtype=float32)

如果显示全部一定要用最大id

seg = tf.constant([5,5,0,0])
data = tf.constant([.3,.4,11,11])
sess.run(tf.unsorted_segment_mean(data,seg,6))
Out[46]: 
array([11.        ,  0.        ,  0.        ,  0.        ,  0.        ,
        0.35000002], dtype=float32)