LatAnalyze

Description

LatAnalyze is a C++14 library for statistical data analysis based on bootstrap resampling. It has been written with lattice QCD data analysis in mind (hence the name), but its features are not lattice specific and can be used more general statistical context.

Sadly a proper documentation was never written, but some comprehensive examples covering several key features can be found in the examples directory.

The main features are the following:

• Array and matrix types with fast arithmetic operations based on Eigen.
• High-level types for bootstrap sample manipulation (including various statistical estimators and histogramming).
• Mathematical expression parser for runtime defined functions.
• Data I/O in ASCII, XML, and HDF5.
• High-level wrappers to minimisation routines from the GSL, Minuit 2 (optional) and NLopt.
• Non-linear regression with error on independent variables (through total least squares).
• High-level wrappers to numerical integrator and non-linear solver from the GSL.
• High-level functional types for function of model. General functions can be defined from C pointers, C++ objects, strings of mathematical expressions or tabulated data.
• High-level plotting functions based on gnuplot, with the possibility of generating and saving plot scripts.

Installation

The head of the master branch always points to the latest stable release. The develop branch is the main unstable branch of LatAnalyze.

LatAnalyze is written in C++14. It has been successfully built on various Unix platform using reasonably recent versions of Clang or GCC. The only strict dependencies are the GSL and HDF5 (built with C++ support). Additionally, cmake (from 3.11), bison (from 3.0) and flex are necessary to build the library. Unless you use a very exotic system, these tools are standard on any Unix platform and should be already present or easy to install through a package manager. Optional dependencies are Minuit 2 and NLopt.

LatAnalyze is built through CMake. Below are example instructions for compiling and installing LatAnalyze.

Portable quick installation

For a quick installation with all possible extensions execute:

./install-latan.sh <prefix>

in the ci-scripts directory where <prefix> is where you want LatAnalyze (and its dependencies) to be installed. This script will automatically download, build and install GSL, HDF5, Minuit, and NLopt.

Quick installation on macOS with dependencies through Homebrew

All the dependencies of LatAnalyze can be installed through the Homebrew package manager.

brew install automake autoconf libtool bison flex gsl minuit2 nlopt hdf5

Then build the library with cmake

mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=<prefix> -DCMAKE_BUILD_TYPE=Release ..
make -j <n>
make install

where <prefix> should be replaced by the desired prefix for LatAnalyze installation, and <n> is the number of parallel build processes.

Dependencies through Homebrew/Linuxbrew

If you use Homebrew/Linuxbrew to manage packages, the following command can install all the dependencies of LatAnalyze (including optional ones)

Using LatAnalyze in CMake projects

When LatAnalyze is installed, the necessary CMake module configuration files to link against the library are available in the installation prefix.

To link against LatAnalyze in a downstream CMake project, simply use

find_package(LatAnalyze)
target_link_libraries(<my_app> LatAnalyze::LatAnalyze)

where <my_app> is a CMake target. If LatAnalyze was installed in a non-standard prefix, you might have to set CMAKE_PREFIX_PATH accordingly for find_package to be successful.

You can additionally modify the find_package to ensure specific components are supported

find_package(LatAnalyze REQUIRED COMPONENTS <...>)

where <...> is a space separated list of components. Currently available components are

• MINUIT2: Minuit 2 minimisers
• NLOPT: NLopt minimisers

History

v3.6

Build system:

• automake build discontinued, now using CMake
• CI through GitHub Actions

Additions:

• Discrete wavelet transform
• Thread pools
• Correlation dynamic range (CDR) calculation
• Data filtering and CDR optimisation

Changes:

• Significant optimisation of covariance matrix calculation
• Covariance matrix available for any StatArray (not only resampled arrays)

v3.5.1

Various fixes and cleaning of outdated code.

v3.5

Additions:

• 'Impulse' & line type plots
• Plot line width & dash modifiers
• Plot palettes (category10 by default)
• Multivariate Gaussian RNG
• 2-pt fitter 'scan' mode over all possible fit ranges
• Command line utility for plotting data

Changes:

• Complete overhaul of the header structure
• Integration of LatCore in LatAnalyze
• p-value is now a 2-sided chi^2 test, 1-sided value kept as 'chi^2 CCDF'

Fixes:

• Matrix plot data now saving correctly
• Many compatibility fixes

v3.4

Additions:

• latan-config utility to easily compile LatAnalyze-based programs
• Linear and constant models to the 2-point fitter

Changes:

• HDF5 is now a compulsory dependency

Fixes:

• Variance matrix computation fix.

v3.3

Additions:

• Sample plot CL utility.
• Infinity as a math constant.
• Option to dump bootstrap sequence while resampling.
• FFT through the GSL.

Changes:

• GSL integrator accepts infinite bounds.
• latan-sample-combine accepts mixes of DSample and DMatSample.
• More general latan-sample-element command.

v3.2.2

Additions:

• The math interpreter supports inf for infinity.

Changes:

• Vector version of setUnidimData.

Fixes:

• Variance matrix computation fix.

v3.2.1

Fixes:

• Wrong argument number check in latan-resample

v3.2 (needs LatCore 1.1)

Additions:

• 2-pt function fitter latan-2pt-fit
• Tool to extract one element of a matrix sample latan-sample-element
• Band plotting

Changes:

• Sample utilities renamed latan-sample-*
• Resample utility renamed latan-resample

v3.1.2

Fixes:

• HDF5 archive URL update in build scripts

v3.1.1 (needs LatCore 1.0)

Fixes:

• Minuit precision fixed
• Minor fit interface fixes

v3.1

Additions:

• Wrappers to NLopt and GSL minimisers.
• Command-line tool to plot the correlation heatmap of a boostrap sample file.
• I/O functions for DSample type.

Changes:

• Internal random generator removed (obsolete because of C++11 pseudo-random generators).
• Fit interface and XY*Data classes rewritten from scratch for improved flexibility and performance.

Fixes:

• Loads of portability and compatibility fixes and CI with Travis.

v3.0

Commit 7b4f2884a5e99bbfab4d4bd7623f609a55403c39.
First 'stable' version of LatAnalyze in C++. The v2.0 refers to the C version and v1.0 to an old undistributed version.
This version compiles fine on OS X with clang but does have many portability issues to other platforms/compilers, v3.1 is the first real release.