Implement machine Learning altorithm LBFGS with Logistic Regression on REEF
- REEF 0.9-SNAPSHOT
- Wake 0.9-SNAPSHOT
- TANG 0.9-SNAPSHOT
- shimoga 0.1-SNAPSHOT
mvn clean install
$ ./run.sh
Currently 6 threads are working. One for Controller task, Five for Compute tasks.
- Distributed group communication
- ControllerTask scatters data to ComputeTasks
- ComputeTasks compute according to the algorithm and reduce the result back
- we just average the result data.(Since it is coefficient)
- Java ported l-bfgs algorithm
- Use this algorithm for fit to logistic regression
- File loading api
- I gived up
- I just put raw data on source code like MatMul example
A researcher is interested in how variables, such as GRE (Graduate Record Exam scores), GPA (grade point average) and prestige of the undergraduate institution, effect admission into graduate school. The response variable, admit/don't admit, is a binary variable.
STDOUT.txt from driver
****************** RESULT ******************
Controller merged data : 0.9999978847998146 0.9999957553215022 2.8
********************************************
Supposed to trained coefficient data. Currently No meaning since I didn't implement logistic regression well.
- liblbfgs
- The MIT License
- REEF
- Apache License, Version 2.0
- L-BFGS
- C port of LBFGS algorithm
- http://www.chokkan.org/software/liblbfgs/
- Notes on CG and LM-BFGS optimization of logistic regression
- http://www.citeulike.org/user/rlsummerscales/article/9706575
- REEF
- Group Communication(MatMul example)
- https://github.com/Microsoft-CISL/shimoga/tree/master/src/main/java/com/microsoft/reef/examples/groupcomm/matmul