[jira] (SYSTEMML-1010) Perftest 0.11 release and related improvements
Title: Message Title Glenn Weidner commented on SYSTEMML-1010 Re: Perftest 0.11 release and related improvements Created similar umbrella for performance testing on 0.12 release candidate(s) including test times results - note different and smaller cluster used for 0.12. Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1010) Perftest 0.11 release and related improvements
Title: Message Title Glenn Weidner edited a comment on SYSTEMML-1010 Re: Perftest 0.11 release and related improvements Created similar umbrella [SYSTEMML-1217] for performance testing on 0.12 release candidate(s) including test times results - note different and smaller cluster used for 0.12. Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1219) Improve instructions and tips for running performance tests
Title: Message Title Glenn Weidner created an issue SystemML / SYSTEMML-1219 Improve instructions and tips for running performance tests Issue Type: Sub-task Assignee: Unassigned Created: 31/Jan/17 05:45 Priority: Minor Reporter: Glenn Weidner There is a readme file at https://github.com/apache/incubator-systemml/blob/master/scripts/perftest/README.TXT which provides good description of scripts used for running performance tests. Add supplemental information for additional configuration tips as part of release process document at http://apache.github.io/incubator-systemml/release-process.html#performance-suite or separate best practices. Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1218) Perftest: Investigate default parameters for long-running tests
Title: Message Title Glenn Weidner created an issue SystemML / SYSTEMML-1218 Perftest: Investigate default parameters for long-running tests Issue Type: Sub-task Assignee: Unassigned Created: 31/Jan/17 05:34 Priority: Minor Reporter: Glenn Weidner To shorten release process test time, investigate updating the perftest to bring scenarios S, M, and L to 2-3 days. For example, some multinomial (e.g., MSVM) can take a long time at 80g which might be related to classes default classes parameter (150). Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1217) Perftest 0.12 release and related improvements
Title: Message Title Glenn Weidner updated an issue SystemML / SYSTEMML-1217 Perftest 0.12 release and related improvements Change By: Glenn Weidner Attachment: times_Stats_80g_v11.txt Attachment: times_Stats_80g_rc1.txt Attachment: times_Regression_80g_v11.txt Attachment: times_Regression_80g_rc2.txt Attachment: times_Regression_80g_rc1.txt Attachment: times_Regression_8g_v11.txt Attachment: times_Regression_8g_rc1.txt Attachment: times_Multinomial_80g_v11_mlreg_sparse.txt Attachment: times_Multinomial_80g_v11_mlreg_dense_run3.txt Attachment: times_Multinomial_80g_v11_mlreg_dense_run2.txt
[jira] (SYSTEMML-1217) Perftest 0.12 release and related improvements
Title: Message Title Glenn Weidner commented on SYSTEMML-1217 Re: Perftest 0.12 release and related improvements Attached times.txt results for various test runs (e.g., 0.12 release candidate 1, latest release 0.11, 0.12 release candidate 2) across different data sizes and in some cases type (dense/sparse). The times file names correspond to the script names (runAllBinomial, runAllMultinomial, runAllRegression, etc) plus the data set size (8g medium, 80g large). A different cluster than SYSTEMML-1010 was used and only had 5 nodes available so the times obtained with v11 on same 5-node cluster were compared. Note data was generated in separate script runs (i.e., genBinomial, genMultinomial, etc) such that the resulting times only include run time (no data generation). Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1217) Perftest 0.12 release and related improvements
Title: Message Title Glenn Weidner created an issue SystemML / SYSTEMML-1217 Perftest 0.12 release and related improvements Issue Type: Umbrella Affects Versions: SystemML 0.12 Assignee: Glenn Weidner Created: 31/Jan/17 05:09 Priority: Major Reporter: Glenn Weidner Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1214) Singular Value Decomposition
Title: Message Title Imran Younus updated an issue SystemML / SYSTEMML-1214 Singular Value Decomposition Change By: Imran Younus Implement scalable version of singular value decomposition. This can be done using scalable QR decomposition, where matrix R is small enough to fit on the driver, so that local svd can be performed on R.Ref:An Improved Algorithm for Computing the Singular Value Decomposition, T Chan, ACM Transactions on Mathematical Software (TOMS) Trans. Math. Soft. 8 , March pp. 72-83 1982 Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1214) Singular Value Decomposition
Title: Message Title Imran Younus updated an issue SystemML / SYSTEMML-1214 Singular Value Decomposition Change By: Imran Younus Implement scalable version of singular value decomposition. This can be done either by using scalable QR decomposition or by computing gram , where matrix R is small enough to fit on the driver, so that local svd can be performed on R . Ref:An Improved Algorithm for Computing the Singular Value Decomposition, T Chan, ACM Transactions on Mathematical Software (TOMS), March 1982 Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1216) implement local svd function
Title: Message Title Imran Younus created an issue SystemML / SYSTEMML-1216 implement local svd function Issue Type: New Feature Assignee: Unassigned Created: 31/Jan/17 01:39 Priority: Major Reporter: Imran Younus SystemML currently provides several local matrix decompositions (qr(), lu(), cholesky()). But local version of svd is missing. This is also needed to scalable SVD implementation. Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1213) Implement scalable matrix decompositions and inverse
Title: Message Title Imran Younus commented on SYSTEMML-1213 Re: Implement scalable matrix decompositions and inverse related PR https://github.com/apache/incubator-systemml/pull/368 Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1215) Scalable Matrix Inverse
Title: Message Title Imran Younus updated an issue SystemML / SYSTEMML-1215 Scalable Matrix Inverse Change By: Imran Younus Implement scalable matrix inverse algorithms for lower/upper triangular matrices. These algorithms are used by various other scalable linear algebra functions like LU, Cholesky and Solve. Also, implement inverse of a square matrix using QR decomposition. Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1215) Scalable Matrix Inverses
Title: Message Title Imran Younus updated an issue SystemML / SYSTEMML-1215 Scalable Matrix Inverses Change By: Imran Younus Summary: Scalable Lower/Upper Matrix Inverses Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1215) Scalable Matrix Inverse
Title: Message Title Imran Younus updated an issue SystemML / SYSTEMML-1215 Scalable Matrix Inverse Change By: Imran Younus Summary: Scalable Matrix Inverses Inverse Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1215) Scalable Lower/Upper Matrix Inverses
Title: Message Title Imran Younus created an issue SystemML / SYSTEMML-1215 Scalable Lower/Upper Matrix Inverses Issue Type: New Feature Assignee: Unassigned Created: 31/Jan/17 01:03 Priority: Major Reporter: Imran Younus Implement scalable matrix inverse algorithms for lower/upper triangular matrices. These algorithms are used by various other scalable linear algebra functions like LU, Cholesky and Solve. Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1214) Singular Value Decomposition
Title: Message Title Imran Younus created an issue SystemML / SYSTEMML-1214 Singular Value Decomposition Issue Type: New Feature Assignee: Unassigned Created: 31/Jan/17 00:57 Priority: Major Reporter: Imran Younus Implement scalable version of singular value decomposition. This can be done either by using QR decomposition or by computing gram matrix. Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
[jira] (SYSTEMML-1213) Implement scalable matrix decompositions and inverse
Title: Message Title Imran Younus created an issue SystemML / SYSTEMML-1213 Implement scalable matrix decompositions and inverse Issue Type: Epic Assignee: Unassigned Attachments: ScalableLinearAlgebraSystemML.docx Components: Algorithms Created: 31/Jan/17 00:39 Environment: spark 2.0 Priority: Major Reporter: Imran Younus This jira tracks implementation of the following linear algebra operations: Matrix Inverse QR, LU, Cholesky and SVD decomposition Linear Solve
[jira] (SYSTEMML-1212) Link to main website in header of project documentation
Title: Message Title Deron Eriksson commented on SYSTEMML-1212 Re: Link to main website in header of project documentation The project documentation exists in docs in incubator-systemml, following the Spark model. This is separate from the website. Spark deploys versioned docs to their site but the project docs live in spark rather than spark-website. Adding a link in the header of the docs ensures that you can always get back to main website from the project docs no matter where the project docs are deployed. cc Frederick Reiss Add Comment This message was sent by Atlassian JIRA (v6.3.15#6346-sha1:dbc023d)
