Mailing-List: contact dev-help@madlib.incubator.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@madlib.incubator.apache.org
From: mktal <git@git.apache.org>
To: dev@madlib.incubator.apache.org
Reply-To: dev@madlib.incubator.apache.org
References: <git-pr-4-incubator-madlib@git.apache.org>
In-Reply-To: <git-pr-4-incubator-madlib@git.apache.org>
Subject: [GitHub] incubator-madlib pull request: SVM: Add cross validation
 support a...
Content-Type: text/plain
Message-Id: <20151125083650.2916CE03CE@git1-us-west.apache.org>
Date: Wed, 25 Nov 2015 08:36:50 +0000 (UTC)

Github user mktal commented on a diff in the pull request:

    https://github.com/apache/incubator-madlib/pull/4#discussion_r45840119
  
    --- Diff: src/ports/postgres/modules/svm/svm.py_in ---
    @@ -440,55 +595,68 @@ def _process_epsilon(is_svc, args):
     def _extract_params(schema_madlib, params, module='SVM'):
         # NOTICE: the type of values in params_default should be consistent with
         # the types specified in params_types
    -    params_default = {'init_stepsize': 0.01,
    -                      'decay_factor': 0.9,
    -                      'max_iter': 100,
    -                      'tolerance': 1e-10,
    -                      'lambda': 1.0,
    -                      'norm': 'L2',
    -                      'n_folds': 0,
    -                      'epsilon': 0.01,
    -                      'eps_table': ''}
    -
    -    params_types = {'init_stepsize': float,
    -                    'decay_factor': float,
    -                    'max_iter': int,
    -                    'tolerance': float,
    -                    'lambda': list,
    -                    'norm': str,
    -                    'n_folds': int,
    -                    'epsilon': float,
    -                    'eps_table': str}
    +    params_default = {
    --- End diff --
    
    Early stopping in optimization serves the similar purposes as regularizations do. Specifying those optimization parameters as lists allows users to CV over them so that they are chosen to provide the `best` generalization ability while avoiding over- or under-fitting. Another practical concern is to make it easier for users to choose a proper init_stepsize which varies from case to case depending on both the model and the data. 


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---