madlib-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Neki, Atsushi" <neki.atsu...@jp.fujitsu.com>
Subject RE: Multinomial Regression: Failed with a msg "Hessian or gradient is not finite."
Date Fri, 14 Apr 2017 06:56:42 GMT
Nagayasu-san, Frank


> Great findings. I guess we are reaching the root cause.

> So, I guess, at least, we can avoid the segfault by using gcc-4.x.

Thanks a lot for your corporation.
By using gcc-4.x, MADLib works so nicely on my environment.

Regards,
--
Atsushi Neki

-----Original Message-----
From: Satoshi Nagayasu [mailto:snaga@uptime.jp] 
Sent: Monday, April 10, 2017 5:32 PM
To: user@madlib.incubator.apache.org
Subject: Re: Multinomial Regression: Failed with a msg "Hessian or gradient is not finite."

2017-04-10 17:19 GMT+09:00 Neki, Atsushi <neki.atsushi@jp.fujitsu.com>:
>> > Do you have any chance to build the PostgreSQL binary from the source code by
yourself, and try it?
>
> It's still Failed.
> Same result with before. (Attached error log)
>
> [NG]  MADLib 1.10(binary that built from source by using gcc-5.4)
>          + PG 9.6(binary that built from source by using gcc-5.4)
>          + Ubuntu 16.04

Great findings. I guess we are reaching the root cause.

gcc-5.4 and/or some components which MADlib relies might have some problem in the binary level.
I will try reproducing it on my box asap.

So, I guess, at least, we can avoid the segfault by using gcc-4.x.

Regards,
--
Satoshi Nagayasu <snaga@uptime.jp>

Mime
View raw message