Project

General

Profile

How to Build Professor on geniegpvm01.fnal.gov

This notes describe, as of May 2019, the procedure of building Professor tuning toolkit on the GENIE dedicated virtual machine at FNAL (geniegpvm01.fnal.gov).

If you have specific questions, please try to contact Julia Yarba:

Please bear in mind that it uses 3rd party packages that are centrally installed in the CVMFS disk space in a form of UPS products..

NOTE: This procedure has been tried on other resources (e.g. Wilson cluster at FNAL, with gvv7.3.0 or gcc v6.4.0).

WARNING: These instructions may or may not work for building Professor2.3.0.
As of May 2019, everything worked well when building Professor2.3.0 (w/Eigen-3.3.5) on the FNAL Wilson cluster (headnode: tev.fnal.gov).
However, an initial attempt to build Professor2.3.0 on geniegpvm01 has failed - work in progress.

UPDATE: As of June 2019, Professor2.3.1 is available, and it builds by the instructions of this document on either Wilson cluster or geniegpvm01.fnal.gov.
If you are interested in Professor2.3.1, please bear in mind the following:
It is NOT distributed via hepforge but is available via bitbucket: wget https://bitbucket.org/iamholger/professor/downloads/Professor-2.3.1.tar.gz
It is recommended to build with Eigen-3.3.7 and YODA-1.7.6
YODA-1.7.6 is available via hepforge or bitbucket: wget https://yoda.hepforge.org/downloads/YODA-1.7.6.tar.gz

STEP-1: Login to geniegpvm01.fnal.gov

STEP-2: Setup necessary externals, as UPS products

source /cvmfs/fermilab.opensciencegrid.org/products/genie/bootstrap_genie_ups.sh
setup root v6_12_06a -q e17:prof

NOTE: Your compiler will be gcc v7_3_0

NOTE: Package pyDOE (needed later) is NOT a part of standard numpy distro !!!

STEP-3: Make sure matplotlib is installed; if not one can install using pip:

pip install matplotlib -t /path/to/your/local/matplotlib
export PYTHONPATH=/path/to/your/local/matplotlib:$PYTHONPATH

NOTE: One can check under interactive python if matplotlib is correctly installed:

python
>>> import matplotlib
>>> exit()

If matplotlib is still missing, one will get an error message (before exiting, that is).

STEP-4: Make sure iMinuit is installed

Trying to import it under interactive python; if not, install it using pip:

pip install iMinuit -t /path/to/you/local/iMinuit
export PYTHONPATH=/path/to/your/local/iMinuit:$PYTHONPATH

Double check interactively if iMinuit is properly in your PYTHONPATH:

python
>>> import iminuit
>>> exit()

GENERAL REMARK: There's a slight chance that running pip will screw up the font in a terminal window...

STEP-5: Download Professor

wget http://www.hepforge.org/archive/professor/Professor-2.2.2.tar.gz

NOTE-1: As mentioned at the beginning of this document, Professor2.3.0 may or may not build on a given node.
However, Professor2.3.1 has been recently released, and it seems to build fine, and is operational.
Bear in mind that it is available via bitbucket but not necessarily via hepforge.
NOTE-2: According to Professor's development team, it may make sense to wait for Professor-2.4 series, as Professor-2.3 is "nothing special".
It should still be noted that Professor2.3.x series offers a "prototype" of a feature that will allow to automatically selected polynomial order in parametrization but the authors say it is "not very scientific".
All in all, in Professor 2.3.0 this is feature is not operational due to a bug that has been fixed in recent release 2.3.1.

STEP-6: Download YODA-1.6.7

wget http://www.hepforge.org/archive/yoda/YODA-1.6.7.tar.gz

NOTE: The YODA1.7.0 "configure" script crashed on tev.fnal.gov on m4_ifblank.
It happened with gcc6.4.0 compiler (that is set together with e.g. art, etc.).
If trying to build e.g. in "native" tev.fnal.gov environment (gcc4-series, etc.), other issues are present.
However, on geniegpvm01 we're building with gcc v7_3_0 so it may make sense to re-check more recent versions.

STEP-7: 6. Download Eigen-3.3.4

See the following URL for details: http://eigen.tuxfamily.org/index.php?title=Main_Page

wget http://bitbucket.org/eigen/eigen/get/3.3.4.tar.gz

NOTE: tarball is called 3.3.4.tar.gz (yes, that's how it's called, no less)

STEP-8: Un-tar Eigen-3.3.4

NOTE: It un-tar's into "eigen-eigen-5a0156e40feb" (hahaha).
Maybe better to mv eigen-eigen-5a0156e40feb to eigen_3.3.4 ???

tar -xzf 3.3.4.tar.gz
mv eigen-eigen-5a0156e40feb eigen-3.3.4

STEP-9: Build ("build" ?) Eigen (it may take a few minutes)

Take a look at the INSTALL notes in the directory - there're installation instructions there (the procedure uses cmake).

NOTE: The INSTALL notes don't explicitly say, but the build procedure requites cmake 2.8.5 or later (while on geniegpvm01 the default is cmake 2.6-patch4)

One may want to setup a modern version of cmake, e.g.

setup cmake v3_12_1 

Note that apparently Eigen requires an "out-of-source" build, so one should first create a build directory for it, and do the actual build from there.
However, if one can still istall into eigen-3.3.4 (or whatever you'd call it), if one wishes (via -DCMAKE_INSTALL_PREFIX).

mkdir eigen-3.3.4-build
cd eigen-3.3.4-build
cmake  -DCMAKE_INSTALL_PREFIX=/path/to/eigen-3.3.4 /path/to/eigen-3.3.4

( e.g. cmake  -DCMAKE_INSTALL_PREFIX=../eigen-3.3.4  ../eigen-3.3.4 )

make install

STEP-10: Un-tar and build Professor, add it to various paths

Instructions are available in Professor-2.2.2/README and online, via the following URL:
http://professor.hepforge.org/docs.sphinx220/install.html#required-packages

NOTE: Professor does NOT require out-of-source build

tar -xzf Professor-2.2.2.tar.gz
cd Professor-2.2.2
CXXFLAGS="-I../eigen-3.3.4/include/eigen3 -O4" make all

( generic way: CXXFLAGS="-I/LOCATION_OF_EIGEN_HEADERS -O4" make all )

export PATH=/path/to/Professor-2.2.2/bin:$PATH
export LD_LIBRARY_PATH=/path/to/Professor-2.2.2/lib:$LD_LIBRARY_PATH
export PYTHONPATH=/path/to/Professor-2.2.2/lib/python2.7/site-packages:$PYTHONPATH

Example: assuming that one is in the Professor-2.2.2 top dir, one can do as follows:

export PATH=$PWD/bin:$PATH
export LD_LIBRARY_PATH=$PWD/lib:$LD_LIBRARY_PATH
export PYTHONPATH=$PWD/lib/python2.7/site-packages:$PYTHONPATH

  • STEP-11: Give a try to Professor's most basic utility prof2-ncoeffs

prof2-ncoeffs 3

The output should look like the following:

Polynomial order     Minimum samples
         0                   1
         1                   4
         2                  10
         3                  20
         4                  35
         5                  56
         6                  84
         7                 120
         8                 165
         9                 220
        10                 286

STEP-12: Un-tar, configure, and build YOUA-1.6.7, with Root extension

Bear in mind that installation into the build directory isn't supported.
For this reason you'll need a separate install directory.

tar -xzf YODA-1.6.7.tar.gz
mkdir YODA-1.6.7-install
cd YODA-1.6.7
PYTHON=$PYTHONHOME/bin/python ./configure --prefix=$PWD/../YODA-1.6.7-install --enable-root
make && make install

NOTE: If one tries to configure as stated in the instructions
./configure --prefix=$PWD/../YODA-1.6.7-install --enable-root
i.e. without leading PYTHON=$PYTHONHOME/bin/python, the configuration procedure will
eventually chock up on not being able to locate python.
So do exactly as recommended above (it was a verbal recommendation from Holger Sch.).

NOTE: If building on a multi-core node, you can do a parallel build. e.g.
make -j4 && make install

STEP-13: Extend paths to YODA

export LD_LIBRARY_PATH=/path-to/YODA-1.6.7-install/lib:$LD_LIBRARY_PATH
export PATH=/path-to/YODA-1.6.7-install/bin:$PATH
export PYTHONPATH=/path-to/YODA-1.6.7-install/lib/python2.7/site-packages:$PYTHONPATH

E.g. to export PYTHONPATH from from the YODA-1.6.7-install top dir, one would do

export PYTHONPATH=$PWD/lib/python2.7/site-packages:$PYTHONPATH 

NOTE: Technically speaking NO YODA is needed to run prof2-ncoeffs, prof2-sample, prof2-ipol, or even prof2-tune.
However, prof2-tune will kind of complain of not finding YODA, thus not writing YODA nistos, but just "tunes" output.
In principle, YODA would be useful since it has yoda2root features, etc. (but fortunately it's not mandatory)

STEP-14: You may want to install pyDOE

If you plan to e.g. tuning with the --scan-n option, you may want to install pyDOE as well.

pip install pyDOE -t /path/to/you/local/pyDOE
export PYTHONPATH=/path/to/your/local/pyDOE:$PYTHONPATH

Double check interactively if pyDOE is properly in your PYTHONPATH:

python
>>> import pyDOE
>>> exit()

STEP-15: Professor in 5 minutes

For those new to Professor and who wishes to see/learn its basics "in 5 minutes", there is a nice example ("Toy Professor") by Robert Hatcher:
https://cdcvs.fnal.gov/redmine/projects/professor-based-parameter-variation/wiki/Creating_Toy_DataMC#Seed-File-Structure

NOTE: In order to try this example, you may need simps package. If it is not in your environment, you may want to install it

pip  install   simpy -t /path/to/your/local/sympy
export PYTHONPATH=/path/to/your/local/sympy:$PYTHONPATH

The create_toy_data.py operates in YODA.

However, the Root-based script would look as follows:

#! /usr/bin/env python2.7
""" 
This is a script to create data for toy "problem" for professor
Uses values in params.json  (professor uses params.dat)
""" 

import random
import ROOT 
import json
from numpy.random import poisson
from sympy import symbols

myfile = open("params.json","r");

myjson = json.load(myfile)
print "myjson object: ",myjson

params = myjson["params"]
print "params object: ",params

a = params["a"]
b = params["b"]
c = params["c"]

x, y, z = symbols('x y z')
func1 = 100*(400.-c*100.*pow(abs(x-5.0*b)/10.0,2+a))
print 'func1: ',func1

def eval3_1(xx,yy,zz):
    v = func1.subs(x,xx)
    print "func1: ",func1,'xx=',xx,' v=',v
    return v

ROOT.TH1.SetDefaultSumw2()
nbins=50
xmin=0.0
xmax=20.0
h1 = ROOT.TH1D( 'h1_foo1', 'toy data', nbins, xmin, xmax )
# for b in range(1,h1.GetNbinsX()+1):
for b in range(h1.GetNbinsX()):
    xmid = h1.GetBinLowEdge(b+1) + 0.5*h1.GetBinWidth(b+1)
    n = eval3_1(xmid,0,0)
    if (n<=0):
        continue
    # mean for this bin ... now poisson
    nthrows = poisson(n)
    print b, n, nthrows
    for toss in range(nthrows):
        h1.Fill(xmid)

ofile = ROOT.TFile( "hists.root", "RECREATE" )
h1.Write()
ofile.Close()

# end-of-script

STEP-16: Try a mini-example (from Geant4)

It's in the followng area:

/genie/app/yarba_j/Professor-Install-Trial/Geant4-FTF-Professor

So please cd to that area, and take a look.
This is a very minimal "pilot" example from Geant4 Fritiof model (FTF), and it's based on 25 points in the 3-parameter space.
There're 2 directories there, scan and data.
The data directory contains a Root file which histograms representing a group of experimental data (from IAEA/Ishibashi, ITEP771, and NA49).
In addition there're params.dat and params.json files that contain Geant4 default settings of the FTF parameters that we want to fit.
The scan directory contains 25 subdirectories, and each subdirectory contains a Root file with simulated histograms (observables) that correspond to the experimental data mentioned above, plus params files with the FTF parameters settings used in this particular simulation.

Try the following steps:

prof2-ipol  scan  - it'll create file ipol.dat (Eigen interpolation of the data)

prof2-tune -d data ipol.dat      - this is the actual fit

NOTE: if one uses --scan-n NPoints option it actually greatly improves fit results, e.g.

prof2-tune -d data ipol.dat --scan-n 25

The output is written to the "tunes" directory.

HAVE FUN !!!