Fast implementations of several probabilistic models.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
code Register `adjust_gradient` in `Preconditioner` Jun 14, 2017
doc Slightly improved documentation. Mar 27, 2015
media Changed README. Sep 15, 2013
tutorials
.gitignore
.travis.yml
CHANGELOG.md Prepared release Jul 16, 2016
Dockerfile
LICENSE Added license file and made some minor changes. Nov 20, 2014
Makefile Fixed some warnings on Mac. Mar 9, 2015
README.md
epydoc.config Minor documentation fixes. Apr 17, 2014
epydoc.css Started more seriously documenting the code. Mar 8, 2013
requirements.txt Added Dockerfile Mar 9, 2018
setup.m Add regularizer support to Matlab interface. Mar 17, 2015
setup.py

README.md

Conditional Modeling Toolkit

samples

Fast implementations of several probabilistic models. Examples:

  • MCGSM (mixture of conditional Gaussian scale mixtures; Theis et al., 2012)
  • MCBM (mixture of conditional Boltzmann machines)
  • FVBN (fully-visible belief network; Neal, 1992)
  • GLM (generalized linear model; Nelder & Wedderburn, 1972)
  • MLR (multinomial logistic regression)
  • STM (spike-triggered mixture model; Theis et al., 2013)

Requirements

Python Interface

  • Python >= 2.7.0
  • NumPy >= 1.6.1
  • automake >= 1.11.0
  • libtool >= 2.4.0
  • Pillow >= 2.6.1 (optional)

Matlab Interface

  • Matlab >= R2009b
  • MEX compatible compiler

I have tested the code with the above versions, but older versions might also work.

Python Example

from cmt.models import MCGSM
from cmt.transforms import WhiteningPreconditioner

# load data
input, output = load('data')

# preprocessing
wt = WhiteningPreconditioner(input, output)

# fit a conditional model to predict outputs from inputs
model = MCGSM(dim_in=input.shape[0],
              dim_out=output.shape[0],
              num_components=8,
              num_scales=6,
              num_features=40)
model.initialize(*wt(input, output))
model.train(*wt(input, output), parameters={
        'max_iter': 1000,
        'threshold': 1e-5})

# evaluate log-likelihood [nats] on the training data
loglik = model.loglikelihood(*wt(input, output)) + wt.logjacobian(input, output)

Matlab Example

% load the data
data = load('data.mat')

% fit a generalized linear model to the data
model = cmt.GLM(10, cmt.ExponentialFunction, cmt.Poisson));
model.train(data.input, data.output, 'maxIter', 1000, 'threshold', 1e-5);

Python Interface Installation

Linux

Make sure autoconf, automake and libtool are installed.

apt-get install autoconf automake libtool

Go to ./code/liblbfgs and execute the following:

./autogen.sh
./configure --enable-sse2
make CFLAGS="-fPIC"

Once the L-BFGS library is compiled, go back to the root directory and execute:

python setup.py build
python setup.py install

Mac OS X

First, make sure you have recent versions of automake and libtool installed. The versions that come with Xcode 4.3 didn't work for me. You can use Homebrew to install newer ones.

brew install autoconf automake libtool
brew link autoconf automake libtool

Next, go to ./code/liblbfgs and execute the following:

./autogen.sh
./configure --disable-dependency-tracking --enable-sse2
make CFLAGS="-arch x86_64 -arch i386"

Once the L-BFGS library is compiled, go back to the root directory and execute:

python setup.py build
python setup.py install

Matlab Interface Installation

Normally it should be enough to download one of the prebuild binaries of our cmt matlab interface. If you want to build it yourself, the following instructions are for you.

Build lbfgs library

The first step is always to follow the above instructions to build liblbfgs for your platform.

On Windows this is done by opening code\liblbfgs\lbfgs.sln in Visual Studio. In Visual Studio, make sure to select the "Release" configuration and the right platform (probably "x64"). Then build the "lib" project by selecting it with a right click and choosing "build". The resulting file lbfgs.lib should be found in the subfolder x64\Release (or "Release" for x86 systems, in which case you have to change that path in the setup script).

Building the mex interface in Matlab

Next open Matlab and make sure a valid mex compiler can be found:

mex -setup

If that is not the case, please follow the official MathWorks documentation to install a supported compiler and check again.

Then execute setup.m from the root folder of cmt in Matlab:

setup

The distribute folder should now contain all the files needed to run the CMT toolbox from within matlab. Add it to your matlab path to use it.