Madlib on greenplum error 22003

时间:2013-10-25 18:58:40

标签: greenplum

我正在尝试在大型数据集上测试madlib linregr_train函数。对于某些独立变量的组合,我得到以下错误:

ERROR:22003:“0.000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000

有时,如果我对数据集进行采样,这些错误就会消失。以前有没有其他人遇到过这个问题?我可以设置哪个选项可以让我将这些数字舍入为零,或者让我知道结果是向下舍入而不是出错?

1 个答案:

答案 0 :(得分:0)

双倍的最小可能值约为4.94e-324,具体取决于您使用的框架/平台。在你的帖子中,我在小数点后计数320个零,在最后一个小数点计数15个数字。你非常接近极限。有关详细信息,请参阅Wikipedia page

在问题发生的地方进行本地化,然后尝试防止数字过小。