bioinf-jku / librfn Goto Github PK

Rectified Factor Networks

License: Other

C++ 91.63% Cuda 4.77% C 3.27% Makefile 0.33%

librfn's Introduction

librfn: Rectified Factor Networks

Rectified Factor Networks (RFNs) are an unsupervised technique that learns a non-linear, high-dimensional representation of its input. The underlying algorithm has been published in

Rectified Factor Networks, Djork-Arné Clevert, Andreas Mayr, Thomas Unterthiner, Sepp Hochreiter, NIPS 2015.

librfn is implemented in C++ and can be easily integrated in existing code bases. It also contains a high-level Python wrapper for ease of use. The library can run in either CPU or GPU mode. For larger models the GPU mode offers large speedups and is the recommended mode.

librfn has been written by Thomas Unterthiner and Djork-Arné Clevert. Sparse matrix support was added by Balázs Bencze and Thomas Adler.

Installation

(optional) Adjust the Makefile to your needs
Type make to start the building process
To use the python wrapper, just copy rfn.py and librfn.so into your working directory.

Requirements

To run the GPU code, you require a CUDA 7.5 (or higher) compatible GPU. While in theory CUDA 7.0 is also supported, it contains a bug that results in a memory leak when running librfn (and your program is likely to crash with an out-of-memory error).

If you do not have access to a GPU, you can disable GPU support by setting USEGPU = no in the Makefile.

Note that librfn makes heavy use of BLAS and LAPACK, so make sure to link it to a high-quality implementation to get optimal speed (e.g. OpenBLAS or MKL) by modifying the Makefile.

Usage

Python wrapper: librfn-python repository.
R extension: librfn-R repository.

Implementation Note

The RFN algorithm is based on the EM algorithm. Within the E-step, the published algorithm includes a projection procedure that can be implemented in several ways (see the RFN paper's supplemental section 9). To make sure no optimzation constraints are violated during this projection, the original publication tries the simplest method first, but backs out to more and more complicated updates if easier method fail (suppl. section 9.5.3). In contrast, librfn always uses the simplest/fastest projection method. This is a simplification/approximation of the original algorithm that nevertheless works very well in practice.

License

librfn was developed by Thomas Unterthiner and is licensed under the General Public License (GPL) Version 2 or higher See License.txt for details.

librfn's People

Contributors

Stargazers

Watchers

Forkers

hihihippp kevinwenya markhofm ihuerga txd866 untom himamis strategist922 shubhampachori12110095 afcarl juexinwang gokceneraslan sailfish009 wangdi2014

librfn's Issues

R package install does not work for clang

Hi - I get an error when I try to install the R package. It may be a clang vs gcc issue, but I'm not really a C++ programmer so I'm not sure. I'm using clang6, which is now recommended by CRAN for R builds on macOS (https://cran.r-project.org/bin/macosx/tools/). Thanks in advance!

My output is:

R CMD build libfrn-R
R CMD INSTALL RFN_0.1.tar.gz

R CMD INSTALL RFN_0.1.tar.gz

installing to library ‘/Library/Frameworks/R.framework/Versions/3.5/Resources/library’
installing source package ‘RFN’ ...
checking for /opt/intel/mkl/include/mkl_blas.h... no
configure: creating ./config.status
config.status: creating src/Makevars
** libs
PKG_LIBS is
BL_LIBS is
/usr/local/clang6/bin/clang++ -Wall -I"/Library/Frameworks/R.framework/Resources/include" -DNDEBUG -DCOMPILE_FOR_R -I -I/usr/local/include -DNOGPU -I"/Library/Frameworks/R.framework/Versions/3.5/Resources/library/Rcpp/include" -I/usr/local/include -std=c++11 -fPIC -Wall -g -O2 -c nist_spblas.cc -o nist_spblas.o
nist_spblas.cc:390:21: error: invalid operands to binary expression ('double'
and 'complex')
y[p->second] += alpha * conj(p->first);
~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~
nist_spblas.cc:481:7: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::sp_conj_axpy' requested here
sp_conj_axpy( alpha * *X, S[i], y, incy);
^
nist_spblas.cc:515:5: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::nondiag_mult_vec_conj_transpose' requested
here
nondiag_mult_vec_conj_transpose(alpha, x, incx, y, incy);
^
nist_spblas.cc:840:5: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::mult_vec_conj_transpose' requested here
mult_vec_conj_transpose(alpha, x, incx, y, incy);
^
nist_spblas.cc:1711:13: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::usmv' requested here
return M->usmv(transa, alpha, x, incx, y, incy);
^
nist_spblas.cc:1928:10: note: in instantiation of function template
specialization 'BLAS_xusmv' requested here
return BLAS_xusmv(
^
nist_spblas.cc:421:10: error: invalid operands to binary expression ('double'
and 'complex')
*Y += alpha * conj(*d) * *X;
~~ ^ ~~~~~~~~~~~~~~~~~~~~~
nist_spblas.cc:518:7: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::mult_conj_diag' requested here
mult_conj_diag(alpha, x, incx, y, incy);
^
nist_spblas.cc:840:5: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::mult_vec_conj_transpose' requested here
mult_vec_conj_transpose(alpha, x, incx, y, incy);
^
nist_spblas.cc:1711:13: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::usmv' requested here
return M->usmv(transa, alpha, x, incx, y, incy);
^
nist_spblas.cc:1928:10: note: in instantiation of function template
specialization 'BLAS_xusmv' requested here
return BLAS_xusmv(
^
nist_spblas.cc:348:17: error: invalid operands to binary expression ('double'
and 'complex')
sum += conj(p->first) * x[p->second];
~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
nist_spblas.cc:453:25: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::sp_conj_dot_product' requested here
y[i] += alpha * sp_conj_dot_product(S[i], x, incx);
^
nist_spblas.cc:521:7: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::nondiag_mult_vec_conj' requested here
nondiag_mult_vec_conj(alpha, x, incx, y, incy);
^
nist_spblas.cc:840:5: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::mult_vec_conj_transpose' requested here
mult_vec_conj_transpose(alpha, x, incx, y, incy);
^
nist_spblas.cc:1711:13: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::usmv' requested here
return M->usmv(transa, alpha, x, incx, y, incy);
^
nist_spblas.cc:1928:10: note: in instantiation of function template
specialization 'BLAS_xusmv' requested here
return BLAS_xusmv(
^
nist_spblas.cc:627:15: error: invalid operands to binary expression ('double'
and 'typename __libcpp_complex_overload_traits::_ComplexType'
(aka 'complex'))
x[jj] /= conj(diag[j]) ;
~~~~~ ^ ~~~~~~~~~~~~~
nist_spblas.cc:885:16: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::transpose_triangular_conj_solve' requested
here
return transpose_triangular_conj_solve(alpha, x, incx);
^
nist_spblas.cc:1737:13: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::ussv' requested here
return M->ussv(transa, alpha, x, incx);
^
nist_spblas.cc:1959:10: note: in instantiation of function template
specialization 'BLAS_xussv' requested here
return BLAS_xussv( transa,
^
nist_spblas.cc:390:21: error: invalid operands to binary expression ('float' and
'complex')
y[p->second] += alpha * conj(p->first);
~~~~~~~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~
nist_spblas.cc:481:7: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::sp_conj_axpy' requested here
sp_conj_axpy( alpha * *X, S[i], y, incy);
^
nist_spblas.cc:515:5: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::nondiag_mult_vec_conj_transpose' requested
here
nondiag_mult_vec_conj_transpose(alpha, x, incx, y, incy);
^
nist_spblas.cc:840:5: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::mult_vec_conj_transpose' requested here
mult_vec_conj_transpose(alpha, x, incx, y, incy);
^
nist_spblas.cc:1711:13: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::usmv' requested here
return M->usmv(transa, alpha, x, incx, y, incy);
^
nist_spblas.cc:2386:10: note: in instantiation of function template
specialization 'BLAS_xusmv' requested here
return BLAS_xusmv(
^
nist_spblas.cc:421:10: error: invalid operands to binary expression ('float' and
'complex')
*Y += alpha * conj(*d) * *X;
~~ ^ ~~~~~~~~~~~~~~~~~~~~~
nist_spblas.cc:518:7: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::mult_conj_diag' requested here
mult_conj_diag(alpha, x, incx, y, incy);
^
nist_spblas.cc:840:5: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::mult_vec_conj_transpose' requested here
mult_vec_conj_transpose(alpha, x, incx, y, incy);
^
nist_spblas.cc:1711:13: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::usmv' requested here
return M->usmv(transa, alpha, x, incx, y, incy);
^
nist_spblas.cc:2386:10: note: in instantiation of function template
specialization 'BLAS_xusmv' requested here
return BLAS_xusmv(
^
nist_spblas.cc:348:17: error: invalid operands to binary expression ('float' and
'complex')
sum += conj(p->first) * x[p->second];
~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
nist_spblas.cc:453:25: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::sp_conj_dot_product' requested here
y[i] += alpha * sp_conj_dot_product(S[i], x, incx);
^
nist_spblas.cc:521:7: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::nondiag_mult_vec_conj' requested here
nondiag_mult_vec_conj(alpha, x, incx, y, incy);
^
nist_spblas.cc:840:5: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::mult_vec_conj_transpose' requested here
mult_vec_conj_transpose(alpha, x, incx, y, incy);
^
nist_spblas.cc:1711:13: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::usmv' requested here
return M->usmv(transa, alpha, x, incx, y, incy);
^
nist_spblas.cc:2386:10: note: in instantiation of function template
specialization 'BLAS_xusmv' requested here
return BLAS_xusmv(
^
nist_spblas.cc:627:15: error: invalid operands to binary expression ('float' and
'typename _libcpp_complex_overload_traits::ComplexType'
(aka 'complex'))
x[jj] /= conj(diag[j]) ;
~~~~~ ^ ~~~~~~~~~~~~~
nist_spblas.cc:885:16: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::transpose_triangular_conj_solve' requested
here
return transpose_triangular_conj_solve(alpha, x, incx);
^
nist_spblas.cc:1737:13: note: in instantiation of member function
'NIST_SPBLAS::TSp_mat::ussv' requested here
return M->ussv(transa, alpha, x, incx);
^
nist_spblas.cc:2417:10: note: in instantiation of function template
specialization 'BLAS_xussv' requested here
return BLAS_xussv( transa,
^
nist_spblas.cc:106:9: warning: private field 'general' is not used
[-Wunused-private-field]
int general;
^
1 warning and 8 errors generated.
make: *** [nist_spblas.o] Error 1
ERROR: compilation failed for package ‘RFN’

Cannot build package: no rule to make target 'gpu_operations.cu'

Hi,

I can't build the package. This is the output from the console:

sudo R CMD INSTALL --configure-args='--with-cuda-home=/usr/lib/nvidia-cuda-toolkit' librfn-R-master

installing to library ‘/usr/local/lib/R/site-library’
installing source package ‘RFN’ ...
checking for /usr/lib/nvidia-cuda-toolkit/bin/nvcc... yes
checking for /usr/lib/nvidia-cuda-toolkit/include/cublas.h... yes
checking for /usr/lib/nvidia-cuda-toolkit/lib64/libcublas.so... yes
checking for /opt/intel/mkl/include/mkl_blas.h... yes
checking for /opt/intel/mkl/lib... yes
checking for /opt/intel/mkl/lib/intel64/libmkl_rt.so... yes
configure: creating ./config.status
config.status: creating src/Makevars
** libs
PKG_LIBS is -L/opt/intel/mkl/lib/intel64 -lmkl_rt -L/usr/lib/nvidia-cuda-toolkit/lib64 -lcublas -lcurand -lcuda -lcudart -lcusolver -lcusparse
BL_LIBS is -L/opt/intel/mkl/lib/intel64 -lmkl_rt
make: *** No rule to make target 'gpu_operations.cu', needed by 'gpu_operations.o'. Stop.
ERROR: compilation failed for package ‘RFN’
removing ‘/usr/local/lib/R/site-library/RFN’

Is this a bug? Or do you have any suggestions how to install?

Cheers!

Recommend Projects

React

A declarative, efficient, and flexible JavaScript library for building user interfaces.
Vue.js

🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript

TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
TensorFlow

An Open Source Machine Learning Framework for Everyone
Django

The Web framework for perfectionists with deadlines.
Laravel

A PHP framework for web artisans
D3

Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

javascript

JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
web

Some thing interesting about web. New door for the world.
server

A server is a program made to process requests and deliver data to clients.
Machine learning

Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Visualization

Some thing interesting about visualization, use data art
Game

Some thing interesting about game, make everyone happy.

Recommend Org

Facebook

We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft

Open source projects and samples from Microsoft.
Google

Google ❤️ Open Source for everyone.
Alibaba

Alibaba Open Source for everyone
D3

Data-Driven Documents codes.
Tencent

China tencent open source team.