Skip to content

Commit

Permalink
Update readme, add doc directory and tutorials (#70)
Browse files Browse the repository at this point in the history
  • Loading branch information
@vissarion
vissarion committed Apr 4, 2020
1 parent ddcbfeb commit 3ec8e41
Show file tree
Hide file tree
Showing 31 changed files with 697 additions and 623 deletions.
2 changes: 1 addition & 1 deletion R-proj/src/direct_sampling.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -117,7 +117,7 @@ Rcpp::NumericMatrix direct_sampling(Rcpp::Nullable<Rcpp::List> body = R_NilValue
unsigned int jj = 0;

for (typename std::list<Point>::iterator rpit = randPoints.begin(); rpit!=randPoints.end(); rpit++, jj++) {
RetMat.col(jj) = Eigen::Map<VT>(&(*rpit).get_coeffs()[0], (*rpit).dimension());
RetMat.col(jj) = rpit->getCoefficients();
}
return Rcpp::wrap(RetMat);
}
19 changes: 12 additions & 7 deletions R-proj/src/sample_points.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -197,6 +197,7 @@ Rcpp::NumericMatrix sample_points(Rcpp::Nullable<Rcpp::Reference> P = R_NilValue
}
}


if (Rcpp::as<Rcpp::List>(random_walk).containsElementNamed("nburns")) {
nburns = Rcpp::as<int>(Rcpp::as<Rcpp::List>(random_walk)["nburns"]);
if (nburns < 0) {
Expand Down Expand Up @@ -233,7 +234,8 @@ Rcpp::NumericMatrix sample_points(Rcpp::Nullable<Rcpp::Reference> P = R_NilValue
HP.normalize();
if (gaussian) {
StartingPoint = StartingPoint - mode;
HP.shift(Eigen::Map<VT>(&mode.get_coeffs()[0], mode.dimension()));
HP.shift(mode.getCoefficients());

}
break;
}
Expand All @@ -252,7 +254,7 @@ Rcpp::NumericMatrix sample_points(Rcpp::Nullable<Rcpp::Reference> P = R_NilValue
if (billiard && diam < 0.0) VP.comp_diam(diam);
if (gaussian) {
StartingPoint = StartingPoint - mode;
VP.shift(Eigen::Map<VT>(&mode.get_coeffs()[0], mode.dimension()));
VP.shift(mode.getCoefficients());
}
break;
}
Expand All @@ -271,7 +273,7 @@ Rcpp::NumericMatrix sample_points(Rcpp::Nullable<Rcpp::Reference> P = R_NilValue
if (billiard && diam < 0.0) ZP.comp_diam(diam);
if (gaussian) {
StartingPoint = StartingPoint - mode;
ZP.shift(Eigen::Map<VT>(&mode.get_coeffs()[0], mode.dimension()));
ZP.shift(mode.getCoefficients());
}
break;
}
Expand All @@ -296,7 +298,7 @@ Rcpp::NumericMatrix sample_points(Rcpp::Nullable<Rcpp::Reference> P = R_NilValue
}
if (gaussian) {
StartingPoint = StartingPoint - mode;
VPcVP.shift(Eigen::Map<VT>(&mode.get_coeffs()[0], mode.dimension()));
VPcVP.shift(mode.getCoefficients());
}
break;
}
Expand Down Expand Up @@ -342,14 +344,17 @@ Rcpp::NumericMatrix sample_points(Rcpp::Nullable<Rcpp::Reference> P = R_NilValue
MT RetMat(dim, numpoints);
unsigned int jj = 0;


for (typename std::list<Point>::iterator rpit = randPoints.begin(); rpit!=randPoints.end(); rpit++, jj++) {
if (gaussian) {
RetMat.col(jj) = Eigen::Map<VT>(&(*rpit).get_coeffs()[0], (*rpit).dimension()) +
Eigen::Map<VT>(&mode.get_coeffs()[0], mode.dimension());

RetMat.col(jj) = rpit->getCoefficients() + mode.getCoefficients();

} else {
RetMat.col(jj) = Eigen::Map<VT>(&(*rpit).get_coeffs()[0], (*rpit).dimension());
RetMat.col(jj) = (*rpit).getCoefficients();
}
}

return Rcpp::wrap(RetMat);

}
260 changes: 13 additions & 247 deletions README.md
View file
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
**VolEsti** is a `C++` library for volume approximation and sampling of convex bodies (*e.g.* polytopes) with an `R` and limited `python` interface.
**VolEsti** is a `C++` library for volume approximation and sampling of convex bodies (*e.g.* polytopes) with an `R` and limited `python` interface. **VolEsti** is part of the [GeomScale](https://geomscale.github.io) project.

[![CRAN status](https://www.r-pkg.org/badges/version/volesti)](https://cran.r-project.org/package=volesti)
[![CRAN downloads](https://cranlogs.r-pkg.org/badges/volesti)](https://cran.r-project.org/package=volesti)
Expand All @@ -11,259 +11,25 @@
[![CircleCI master](https://circleci.com/gh/GeomScale/volume_approximation/tree/master.svg?style=shield)](https://circleci.com/gh/GeomScale/volume_approximation/tree/master)
[![CircleCI master](https://circleci.com/gh/GeomScale/volume_approximation/tree/develop.svg?style=shield)](https://circleci.com/gh/GeomScale/volume_approximation/tree/develop)

### R Interface
------------
### Documentation

#### Install Rcpp package

* Install package-dependencies: `Rcpp`, `RcppEigen`, `BH`, `lpSolveAPI`.

1. Then use devtools package to install `volesti` Rcpp package. In folder /root/R-proj Run:
```r
Rcpp::compileAttributes()
library(devtools)
devtools::build()
devtools::install()
library(volesti)
```
2. You can use Rstudio as well to open `volesti.Rproj` and then click `build source Package` and then `Install and Restart` in `Build` at the menu bar.

#### Generate CRAN version

To generate the CRAN version of the R package follow the instructions below:

1. From the command line navigate to folder `/cran_gen`. Then Run:
```r
source('genCRANpkg.R')
```

2. Open genCRANpkg.R script with `Rstudio` and run it.

#### Run volesti from `R`
* The main function is `volume()`. It can be used to approximate the volume of a convex polytope given as a set of linear inequalities or a set of vertices (d-dimensional points) or as a Minkowski sum of segments (zonotope). There are two algorithms that can be used. The first is `SequenceOfBalls` and the second is `CoolingGaussian` (see References).
* The function `sample_points()` can be used to sample points from a convex polytope with uniform or spherical gaussian target distribution.
* The function `round_polytope()` can be used to round a convex polytope.
* The function `rand_rotate()` can be used to apply a random rotation to a convex polytope.

For more details you can read the documentation in folder `/inst/doc`.

#### Create pdf documentation from Rd files
* Install volesti library.
* In `R` mode (or in Rstudio) Run
```
pack = "volesti"
path = find.package(pack)
system(paste(shQuote(file.path(R.home("bin"), "R")),
"CMD", "Rd2pdf", shQuote(path)))
```
* The pdf will be created and saved in R-proj folder.
* We give such a documentation in /R-proj/doc folder.

### C++ Interface
------------

#### Compile C++ sources and run tests

To compile the C++ code you need the [lp_solve](http://lpsolve.sourceforge.net/5.5/) library. For example, for Unix/Linux you need `liblpsolve55.so` (this is available from the library's [webpage](http://lpsolve.sourceforge.net/5.5/) as well as a package in several linux distributions e.g. [debian](https://packages.debian.org/stretch/liblpsolve55-dev)). You have to specify the path to `liblpsolve55.so`, by running, in folder test:
```
cmake -DLP_SOLVE=_PATH_TO_LIB_FILE_ .
make
```
For example: `-DLP_SOLVE=/usr/lib/lpsolve/liblpsolve55.so`

You can run the tests by `cmake test` or `ctest -jK` where `K` the number of `CPU` threads. By adding the option `--verbose` to `ctest` you get more information about the tests, *e.g.* time per test, volume computed and the name of the polytope or convex body.

#### Polytope input

The current version of the software assumes that the polytope is given in the form of linear inequalities i.e. {x \in R^d : Ax <= b} where A is a matrix of dimension m *x* d and b a vector of dimension m or as a set of m vertices {\in R^d} or as a Minkowski sum of m segments {\in R^d}. The input is described in an `.ine`-file (H-polytopes) or in a `.ext` file (V-polytopes or zonotopes). The `.ine` file is described as follows:

```
various comments
begin
m d+1 numbertype
b -A
end
various options
```

The `.ext` file is described as follows:
```
various comments
begin
m d+1 numbertype
1 v_1
.. ...
1 v_m
end
various options
```
In V-polytope case v_i are vertices and in zonotope case they are segments.

This filestype (or similar) is used by a number of other software in polyhedral computation (e.g. `cdd`, `vinci`, `latte`). In the current version of the software, the options are treated as comments and the numbertype as C++ double type.
If your input has equality constraints then you have to transform it in the form that only contain linear inequalities which described above by using some other software. We recommend to use latte https://www.math.ucdavis.edu/~latte for this transformation.

#### Run volesti from command line

After successful compilation you can use the software by command line. For example, the following command `./vol -h` will display a help message about the program's available options.

###### Example

To estimate the volume of the 10-dimensional hypercube first prepare the file `cube10.ine` as follows:

```
cube10.ine
H-representation
begin
20 11 real
1 1 0 0 0 0 0 0 0 0 0
1 0 1 0 0 0 0 0 0 0 0
1 0 0 1 0 0 0 0 0 0 0
1 0 0 0 1 0 0 0 0 0 0
1 0 0 0 0 1 0 0 0 0 0
1 0 0 0 0 0 1 0 0 0 0
1 0 0 0 0 0 0 1 0 0 0
1 0 0 0 0 0 0 0 1 0 0
1 0 0 0 0 0 0 0 0 1 0
1 0 0 0 0 0 0 0 0 0 1
1 -1 0 0 0 0 0 0 0 0 0
1 0 -1 0 0 0 0 0 0 0 0
1 0 0 -1 0 0 0 0 0 0 0
1 0 0 0 -1 0 0 0 0 0 0
1 0 0 0 0 -1 0 0 0 0 0
1 0 0 0 0 0 -1 0 0 0 0
1 0 0 0 0 0 0 -1 0 0 0
1 0 0 0 0 0 0 0 -1 0 0
1 0 0 0 0 0 0 0 0 -1 0
1 0 0 0 0 0 0 0 0 0 -1
end
input_incidence
```

Then to use SequenceOfBalls (SOB) algorithm run the following command:
```
./vol -f1 cube_10.ine
```

which returns 17 numbers:
```d m #experiments exactvolOr-1 approxVolume [.,.] #randPoints walkLength meanVol [minVol,maxVol] stdDev errorVsExact maxminDivergence time timeChebyshevBall```

To use CoolingGaussian (CG) algorithm run the following command:
```
./vol -f1 cube_10.ine -cg
```
which returns the same output as before.

To estimate the volume of a 10-dimensional V-cross polytope described in `cross_10.ext` as follows:
```
cross_10.ext
V-representation
begin
20 11 integer
1 1 0 0 0 0 0 0 0 0 0
1 0 1 0 0 0 0 0 0 0 0
1 0 0 1 0 0 0 0 0 0 0
1 0 0 0 1 0 0 0 0 0 0
1 0 0 0 0 1 0 0 0 0 0
1 0 0 0 0 0 1 0 0 0 0
1 0 0 0 0 0 0 1 0 0 0
1 0 0 0 0 0 0 0 1 0 0
1 0 0 0 0 0 0 0 0 1 0
1 0 0 0 0 0 0 0 0 0 1
1 -1 0 0 0 0 0 0 0 0 0
1 0 -1 0 0 0 0 0 0 0 0
1 0 0 -1 0 0 0 0 0 0 0
1 0 0 0 -1 0 0 0 0 0 0
1 0 0 0 0 -1 0 0 0 0 0
1 0 0 0 0 0 -1 0 0 0 0
1 0 0 0 0 0 0 -1 0 0 0
1 0 0 0 0 0 0 0 -1 0 0
1 0 0 0 0 0 0 0 0 -1 0
1 0 0 0 0 0 0 0 0 0 -1
end
hull
incidence
```
Run:
```
./vol -f2 cross_10.ext
```
which returns the same output as before.

To estimate the volume of a 4-dimensional zonotope defined by the Minkowski sum of 8 segments described in `zonotope_4_8.ext` as follows:
```
zonotope_4_8.ext
Zonotpe
begin
8 5 real
1 0.981851 -0.188734 -0.189761 0.0812645
1 -0.0181493 0.811266 -0.189761 0.0812645
1 -0.0181493 -0.188734 0.810239 0.0812645
1 -0.0181493 -0.188734 -0.189761 1.08126
1 -0.177863 0.437661 -0.0861379 -0.674634
1 0.737116 -0.204646 -0.540973 -0.471883
1 -0.684154 0.262324 0.292341 -0.265955
1 -0.802502 -0.740403 0.0938152 0.0874131
end
hull
incidence
```
Run:
```
./vol -f3 zonotope_4_8.ext
```
Flag `-v` enables the print mode.

#### Generate polytopes

You can use executable `generator` to generate polytopes (hypercubes, simplices, cross polytopes, skinny hypercubes (only in H-representation), product of two simplices (only in H-representation) and zonotoes. For example:

1. To generate a 10-dimensional hypercube in H-representation run:
```
./generate -cube -h -d 10
```

2. To generate a 20-dimensional simplex in V-representaion run:
```
./generate -simplex -v -d 20
```

3. To generate a 5-dimensional zonotope defined by 10 segments run:
```
./generate -zonotope -d 5 -m 10
```

Command `./generate -help` will display a help message about the program's available options.

#### Sampling

You can sample from a convex polytope uniformly or from the spherical gaussian distribution. For example:

1. To sample uniformly from the 10-dimensional hypercube, run:
```
./vol -f1 cube_10.ine -rand -nsample 1000
```
Flag -nsample declares the number of points we wish to sample (default is 100).

2. To sample from the gaussian distribution, run:
```
./vol -f1 cube_10.ine -rand -nsample 1300 -gaussian -variance 1.5
```
Flag `-variance` declares the variance (default is 1.0). The center of the spherical gaussian is the Chebychev center for H-polytopes, or the origin for zonotopes. For V-polytopes is the chebychev center of the simplex that is defined by a random choice of d+1 vertices.

3. To sample from a zonotope described in zonotope.ext file run:
```
./vol -f3 zonotope.ext -rand -nsample 1500
```
For V-polytopes use flag `-f2` before the `.ext` file. In all cases use flag `-v` to print the excecutional time.
* [Using the R interface](doc/r_interface.md)
* [Using the C++ Interface](doc/cpp_interface.md)
* [Wikipage with tutorials and demos](https://github.com/GeomScale/volume_approximation/wiki)
* [Tutorial given to PyData meetup](https://vissarion.github.io/tutorials/volesti_tutorial_pydata.html)

### Credits

Copyright (c) 2012-2018 Vissarion Fisikopoulos
Copyright (c) 2018 Apostolos Chalkis
Copyright (c) 2012-2020 Vissarion Fisikopoulos
Copyright (c) 2018-2020 Apostolos Chalkis

You may redistribute or modify the software under the GNU Lesser General Public License as published by Free Software Foundation, either version 3 of the License, or (at your option) any later version. It is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY.

Main development by Vissarion Fisikopoulos while he was affiliated with University of Athens (UoA, Greece), University of Brussels (ULB, Belgium), and Chalkis Apostolos affiliated with University of Athens (UoA, Greece).
Main development started in 2012 by Vissarion Fisikopoulos while he was affiliated with University of Athens (UoA, Greece), and then with University of Brussels (ULB, Belgium). The main sampling and volume algorithms were developed at that time.

Later, Chalkis Apostolos affiliated with University of Athens (UoA, Greece) and mentored by V.F. and partially supported by GSoC'2018, 2019 programs enhanced volesti with more algorithms for sampling and volume as well as an `R` interface.

We acknowledge several contributions by the open-source community, most notably a `python` interface by Pedro Zuidberg Dos Martires affiliated with KU Leuven.

### Publications

Expand Down
Loading

0 comments on commit 3ec8e41

Please sign in to comment.