systemml-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Niketan Pansare" <npan...@us.ibm.com>
Subject Open tasks: Integration with MLPipeline
Date Thu, 03 Dec 2015 22:32:21 GMT


Hi all,

In this email, I list the open tasks related to integration with
MLPipeline. This allows external developers to contribute to the SystemML
project until our JIRA server is up and running.

1. Make the existing Logistic regression wrapper more robust:
- Extend the wrapper or the DML script to handle zero-based labels (either
throw an error or support zero-based labels).

2. Improve the performance of the Logistic regression wrapper:
- Profile the wrapper to find potential bottlenecks. The candidates for
bottlenecks are RDDConverterUtilsExt.vectorDataFrameToBinaryBlock and line
153-158 in LogisticRegressionModel.

3.  Perform detailed performance analysis of the converter utils.
- Also explore the usability aspect of these utils.

4. Add MLPipeline wrappers for existing scripts.
- Refer to
https://github.com/apache/incubator-systemml/tree/master/scripts/algorithms
to pick the algorithm and
http://apache.github.io/incubator-systemml/algorithms-reference.html to
understand the assumptions as well as parameters to the given algorithm.
- A good algorithm to start with is L2SVM:
http://apache.github.io/incubator-systemml/algorithms-classification.html#binary-class-support-vector-machines
https://github.com/apache/incubator-systemml/blob/master/scripts/algorithms/l2-svm.dml
https://github.com/apache/incubator-systemml/blob/master/scripts/algorithms/l2-svm-predict.dml

5. Add the documentation for MLPipeline wrappers to
http://apache.github.io/incubator-systemml/index.html

References:
1. Existing Logistic regression wrappers:
https://github.com/apache/incubator-systemml/blob/master/src/main/java/org/apache/sysml/api/ml/LogisticRegression.java
https://github.com/apache/incubator-systemml/blob/master/src/main/java/org/apache/sysml/api/ml/LogisticRegressionModel.java

2. Converter utils:
https://github.com/apache/incubator-systemml/blob/master/src/main/java/org/apache/sysml/runtime/instructions/spark/utils/RDDConverterUtilsExt.java

Thanks,

Niketan Pansare
IBM Almaden Research Center
E-mail: npansar At us.ibm.com
http://researcher.watson.ibm.com/researcher/view.php?person=us-npansar

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message