[
https://issues.apache.org/jira/browse/MADLIB-1168?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16444594#comment-16444594
]
ASF GitHub Bot commented on MADLIB-1168:
----------------------------------------
Github user jingyimei commented on a diff in the pull request:
https://github.com/apache/madlib/pull/265#discussion_r182843994
--- Diff: RELEASE_NOTES ---
@@ -9,6 +9,56 @@ commit history located at
https://github.com/apache/madlib/commits/master.
Current list of bugs and issues can be found at
https://issues.apache.org/jira/browse/MADLIB.
—-------------------------------------------------------------------------
+MADlib v1.14:
+
+Release Date: 2018-April-28
+
+New features:
+* New module - Balanced datasets: A sampling module to balance
classification
+ datasets by resampling using various techniques including
undersampling,
+ oversampling, uniform sampling or user-defined proportion sampling
+ (MADLIB-1168)
+* Mini-batch: Added a mini-batch optimizer for MLP and a preprocessor
function
+ necessary to create batches from the data (MADLIB-1200, MADLIB-1206)
+* k-NN: Added weighted averaging/voting by distance (MADLIB-1181)
+* Summary: Added additional stats: number of positive, negative, zero
values and
+ 95% confidence intervals for the mean (MADLIB-1167)
+* Encode categorical: Updated to produce lower-case column names when
possible
+ (MADLIB-1202)
+* MLP: Added support for already one-hot encoded categorical dependent
variable
+ in a classification task (MADLIB-1222)
+* Pagerank: Added option for personalized vertices that allows higher
weightage
+ for a subset of vertices which will have a higher jump probability as
+ compared to other vertices and a random surfer is more likely to
+ jump to these personalization vertices (MADLIB-1084)
+
+Bug fixes:
+ - Fixed issue with invalid calls of construct_array that led to
problems
+ in Postgresql 10 (MADLIB-1185)
+ - Added newline between file concatenation during PGXN install
(MADLIB-1194)
+ - Fixed upgrade issues in knn (MADLIB-1197)
+ - Added fix to ensure RF variable importance are always non-negative
+ - Fixed inconsistency in LDA output and improved usability
+ (MADLIB-1160, MADLIB-1201)
+ - Fixed MLP and RF predict for models trained in earlier versions to
+ ensure misisng optional parameters are given appropriate default
values
+ (MADLIB-1207)
+ - Fixed a scenario in DT where no features exist due categorical
columns
+ with single level being dropped led to the database crashing
+ - Fixed step size initialization in MLP based on learning rate policy
+ (MADLIB-1212)
+ - Fixed PCA issue that leads to failure when grouping column is a TEXT
type
+ (MADLIB-1215)
+ - Fixed cat levels output in DT when grouping is enabled (MADLIB-1218)
+ - Fixed and simplified initialization of model coefficients in MLP
+ - Removed source table dependency for predicting regression models in
MLP
+ (MADLIB-1223)
+ - Print loss of first iteration in MLP (MADLIB-1228)
+
--- End diff --
We should mention MADLIB-1209 Neural net related bug fix.
> Balance datasets
> ----------------
>
> Key: MADLIB-1168
> URL: https://issues.apache.org/jira/browse/MADLIB-1168
> Project: Apache MADlib
> Issue Type: New Feature
> Components: Module: Sampling
> Reporter: Frank McQuillan
> Assignee: ssoni
> Priority: Major
> Fix For: v1.14
>
> Attachments: MADlib Balance Datasets Requirements.pdf,
> MADlib_Balance_Datasets_Requirements_v2.pdf
>
>
> From [1] here is the motivation behind balancing datasets:
> “Most classification algorithms will only perform optimally when the number
> of samples of each class is roughly the same. Highly skewed datasets, where
> the minority is heavily outnumbered by one or more classes, have proven to be
> a challenge while at the same time becoming more and more common.
> One way of addressing this issue is by re-sampling the dataset as to offset
> this imbalance with the hope of arriving at a more robust and fair decision
> boundary than you would otherwise.
> Re-sampling techniques can be divided in these categories:
> * Under-sampling the majority class(es).
> * Over-sampling the minority class.
> * Combining over- and under-sampling.
> * Create ensemble balanced sets.”
> There is an extensive literature on balancing datasets. The plan for MADlib
> in the initial phase is to offer basic functionality that can be extended in
> later phases based on feedback from users.
> Please see attached document for proposed scope of this story.
> References
> [1] imbalance-learn Python project
> http://contrib.scikit-learn.org/imbalanced-learn/stable/index.html
> https://github.com/scikit-learn-contrib/imbalanced-learn
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
