Mailing-List: contact dev-help@mahout.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@mahout.apache.org
Date: Tue, 5 Feb 2013 21:59:15 +0000 (UTC)
From: "Sebastian Schelter (JIRA)" <jira@apache.org>
To: dev@mahout.apache.org
Message-ID: <JIRA.12630597.1359938428719.249313.1360101555025@arcas>
In-Reply-To: <JIRA.12630597.1359938428719@arcas>
References: <JIRA.12630597.1359938428719@arcas>
Subject: [jira] [Commented] (MAHOUT-1148) QR Decomposition is too slow
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/MAHOUT-1148?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13571776#comment-13571776 ] 

Sebastian Schelter commented on MAHOUT-1148:
--------------------------------------------

I ran a quick mini benchmark using ALS to factorize the movielens1M dataset, (I didn't check correctness of the results). Numbers indicate a 2x to 3x increase in performance:

{noformat}
[2] features, [3] iterations: OLD [1233] ms <-> NEW [1254] ms
[2] features, [3] iterations: OLD [760] ms <-> NEW [926] ms
[2] features, [3] iterations: OLD [758] ms <-> NEW [768] ms
[2] features, [3] iterations: OLD [768] ms <-> NEW [774] ms
[2] features, [3] iterations: OLD [772] ms <-> NEW [760] ms
[5] features, [3] iterations: OLD [1116] ms <-> NEW [1015] ms
[5] features, [3] iterations: OLD [1110] ms <-> NEW [981] ms
[5] features, [3] iterations: OLD [1124] ms <-> NEW [986] ms
[5] features, [3] iterations: OLD [1116] ms <-> NEW [1008] ms
[5] features, [3] iterations: OLD [1105] ms <-> NEW [989] ms
[10] features, [3] iterations: OLD [2577] ms <-> NEW [1825] ms
[10] features, [3] iterations: OLD [2578] ms <-> NEW [1863] ms
[10] features, [3] iterations: OLD [2566] ms <-> NEW [1834] ms
[10] features, [3] iterations: OLD [2587] ms <-> NEW [1827] ms
[10] features, [3] iterations: OLD [2623] ms <-> NEW [1839] ms
[20] features, [3] iterations: OLD [11211] ms <-> NEW [5863] ms
[20] features, [3] iterations: OLD [11106] ms <-> NEW [5820] ms
[20] features, [3] iterations: OLD [11041] ms <-> NEW [5851] ms
[20] features, [3] iterations: OLD [11047] ms <-> NEW [5837] ms
[20] features, [3] iterations: OLD [11027] ms <-> NEW [5748] ms
[30] features, [3] iterations: OLD [30781] ms <-> NEW [13929] ms
[30] features, [3] iterations: OLD [30900] ms <-> NEW [13971] ms
[30] features, [3] iterations: OLD [30917] ms <-> NEW [13890] ms
[30] features, [3] iterations: OLD [30933] ms <-> NEW [13984] ms
[30] features, [3] iterations: OLD [30902] ms <-> NEW [13964] ms
[50] features, [3] iterations: OLD [118351] ms <-> NEW [41957] ms
[50] features, [3] iterations: OLD [118682] ms <-> NEW [42165] ms
[50] features, [3] iterations: OLD [118605] ms <-> NEW [42302] ms
[50] features, [3] iterations: OLD [118698] ms <-> NEW [42219] ms
{noformat}
                
> QR Decomposition is too slow
> ----------------------------
>
>                 Key: MAHOUT-1148
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-1148
>             Project: Mahout
>          Issue Type: Bug
>            Reporter: Ted Dunning
>
> A user reported that QR decomposition is too slow.  I coded up a replacement that can be 10x faster under certain cases and the new version is also tested.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira