QMM LIBRARY -- A Mod for LIBLINEAR
A regularized empirical risk minimization software developed at Sabanci University
Project News
-
We plan to update our software regularly based on user needs. You can contact
Hakan Erdogan for further information.
-
QMM library website is up!
-
Currently developing the website...
Introduction
QMM library is a general solver for binary and multi-class regularized empirical risk classification. Currently only binary classification with coordinate descent method is functional.
- Based on majorize minimize (MM) method
- Monotonicity guarantees
- Possible to use many different loss functions
- Supports L1, L2 or elastic net regularization
How to run our project
QMM library is a modification of the liblinear software. You can use the same input and output formats. We have added new types of solvers to the liblinear software to enable use of our algorithms.
To enable using QMM library use the options -s 7 through -s 10.
-s type : set type of solver (default 1)
0 -- L2-regularized logistic regression
1 -- L2-regularized L2-loss support vector classification (dual)
2 -- L2-regularized L2-loss support vector classification (primal)
3 -- L2-regularized L1-loss support vector classification (dual)
4 -- multi-class support vector classification by Crammer and Singer
5 -- L1-regularized L2-loss support vector classification
6 -- L1-regularized logistic regression
7 -- MMCD - please specify loss, curv
8 -- MMCD_SM - soft-max method
9 -- MMCD_SG - sub-gradient
10 -- MMCD_SIMPLE - please specify loss, curv
-l loss_type : L1, L2, LOG, HU1, HU2, LS
0 -- L1
1 -- L2
2 -- LOG
3 -- HU1
4 -- HU2
5 -- LS
-u curv_type : MC, OC, NC
0 -- MC
1 -- OC
2 -- NC
-r alpha : regularization parameter, between 0 and 1
0 -- L1 regularization
1 -- L2 regularization
0<r<1 -- elastic net
-t tau : loss parameter for HU1, HU2 and L1 losses, typical 0.1
-p epsi : curv parameter, minimum curvature for NC, typical 0.001
-i initmodel : initial model file to start iterations
-h chat_level : how much should I talk?
0 -- minimal
1 -- calc and print obj
-c cost : set the parameter C (default 1)
-e epsilon : set tolerance of termination criterion
-s 0 and 2
|f'(w)|_2 <= eps*min(pos,neg)/l*|f'(w0)|_2,
where f is the primal function and pos/neg are # of
positive/negative data (default 0.01)
-s 1, 3, and 4
Dual maximal violation <= eps; similar to libsvm (default 0.1)
-s 5 and 6
|f'(w)|_inf <= eps*min(pos,neg)/l*|f'(w0)|_inf,
where f is the primal function (default 0.01)
-s 7
|w-w^prev|_inf <= eps*|w|_inf,
-B bias : if bias >= 0, instance x becomes [x; bias]; if < 0, no bias term added (default -1)
-wi weight: weights adjust the parameter C of different classes (see README for details)
-v n: n-fold cross validation mode
-q : quiet mode (no outputs)
Download
Links
The current release (October 2011) of our software
can be obtained by downloading the zip file.
This is a modification of liblinear version 1.5.1 .
The package includes the source code in C/C++.
Documentation
and Codes
Codes will be published here.
For abstract of the project English / Turkish
Papers implementing our mod
-
Mehmet Umut Sen, Hakan Erdogan, "Basit Birlestirici Tipleri için Dogrusal Olmayan Siniflandirici Birlestirme," IEEE SIU 2011, Kemer, Nisan 2011.
-
Ibrahim Saygin Topkaya, Mehmet Umut Sen, Mustafa Berkay Yilmaz, Hakan Erdogan, "Görsel-Isitsel Tandem Siniflandiricilar ve Birlesimleri ile Konusma Tanima Basarisini Arttirma," IEEE SIU 2011, Kemer, Nisan 2011.
Please
send comments and suggestions to Hakan Erdogan.