madlib-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From iyerr3 <>
Subject [GitHub] madlib pull request #229: SVM: Add minibatch as a new solver
Date Fri, 19 Jan 2018 17:41:07 GMT
GitHub user iyerr3 opened a pull request:

    SVM: Add minibatch as a new solver

    Additional author: Nikhil Kak <>
    This work is based on the original work by Xiaocheng Tang <>
in #75. 
    This PR adds two main features: 
    1. A Minibatch solver that takes as input a batch of data
    2. Minibatching for SVM 

You can merge this pull request into a Git repository by running:

    $ git pull feature/svm_minibatch

Alternatively you can review and apply these changes as the patch at:

To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:

    This closes #229
commit 8dde3fc0e42be6fbd97585cc046060b84d624da1
Author: Rahul Iyer <riyer@...>
Date:   2018-01-08T21:21:16Z

    New module: Add minibatch as additional optimization framework

commit 8b0af20c79af41d6c5fe023b32bd885f15df5bd1
Author: Rahul Iyer <riyer@...>
Date:   2018-01-08T21:23:55Z

    SVM: Add minibatch capabilities

commit 2526d61f36e740df47e6b103e485133deb99ec43
Author: Rahul Iyer <riyer@...>
Date:   2018-01-09T00:55:34Z

    Add new dataset for batch

commit 6a8649771d7950d376285b24e25070fd524be519
Author: Nikhil Kak <nkak@...>
Date:   2018-01-09T18:52:18Z

    Add install-check test for svm minibatch

commit b5d1adbc5b501640cb0230d5622143d9b25ce4f5
Author: Nikhil Kak <nkak@...>
Date:   2018-01-10T18:27:57Z

    Add predict call for svm installcheck test

commit a943b1ac6c162a33eaf269d775061ddc559dd360
Author: Rahul Iyer <riyer@...>
Date:   2018-01-10T23:30:00Z

    Update model in getLoss function

commit 971403769f65a05e95a07d2df44c8a30120d025c
Author: Nikhil Kak <nkak@...>
Date:   2018-01-11T00:46:07Z

    Refactor svm minibatch to add comments and update variable names.
    We are now using a ColumnVector instead of MappedColumnVector because
    the minibatch transition function wasn't able to convert ColumnVector to
    MappedColumnVector. This required us to not rebind tuple.depVar and
    instead just assign it to y.

commit c86a36ca149a99d569ba54ddaaabe585f729df59
Author: Nikhil Kak <nkak@...>
Date:   2018-01-11T21:47:16Z

    Add classification test for svm minibatch and add relevane asserts.

commit a43bff2c0392770148fba2d04fb75c19d48803ee
Author: Rahul Iyer <riyer@...>
Date:   2018-01-12T00:42:13Z

    SVM: Fix classification with minibatching
    - Unnest minibatch array data to get all dependent labels
    - Transform dependent labels to 1/-1 by unnesting and then rebuilding
    - Update install-check to use text labels and better thresholds for

commit 5d29f4ed7eaa8bdac16b72136f70aeca7997b54d
Author: Nikhil Kak <nkak@...>
Date:   2018-01-12T01:30:21Z

    Use correct count for averaging the gradient/loss

commit b9a69ddabdade4a3f57577de391d0f093ed93630
Author: Rahul Iyer <riyer@...>
Date:   2018-01-12T21:48:45Z

    SVM: Fix minibatch data in install-check

commit 60bda8a147d3a354329c9720fd41bb13650799a9
Author: Rahul Iyer <riyer@...>
Date:   2018-01-18T00:09:49Z

    Add assert for data validation (+ comment updates)



View raw message