exec/abcdir/README.md

In simoncarrignon/spreadrt: Model of Twitter Retweet Cascade Spread

ABC - SLRUM - TWITTER-SPREAD

This Folder abcdir is a working folder yet not well organized but that aims at storing all scipt needed to run and analyse the model available in twitter-spread usings ABC.

I will keep some note here on the name of the folders: usually VF or TF mean "True False/ Vrai Faux". This generally note that the abc has been run with the score computed against both distribution of false and true tweet. testneutralRumors means Rumors ie we test 3Dcascade on the distribution of RUMORS and not on all cascades.

Content:

All fine abc*.R (like: abcRandom.R, abcRumors.R,...) are script able to run ABC using MPI on their own. They generate priors, run the experiment using parLapply function from Rmpi package. They create output folder where the scrore and the parameters of the simulations will be stored.

This is ok for relatively quick model that all last around the same time. This is not the case with the 3D model that can sometime generate really really slow simulations, which will then ask SLURM for hours when just one node will need this hours. To try to optimize this we wrote a "generateParameters.R"

• file in the folder slurm-mpi/ are script used to run bunch of experiment via mpi and SLURM.

file generateParameters.R

The idea behind this script is to generate a huge set of parameter that can be split given their expected running time. The times used in the current version of this folder have been calculated by running bunch of test and using a log-linear fitting. Equation and should be somewhere in the scripts here .

To generate the folder with all parametesr use something like:

Rscript generateParameters.R numfolder_stored_parameters $folder_stored_parameters

This will be read using the script readParameters.R and run using Cascade3D. readParameters.R is wrapped within slurm-mpi/Rmpiscript-mn4_splitted_large and slurm-mpi/Rmpiscript-mn4_splitted_small scripts that wrap the call within SLURM scheduling system and MPI.

Then you can use the slurm-mpi/send_batch_simu file that look like:

folder_stored_parameterss=nonneutralRumors
outputSimulation=nonneutralRumorsVF
for i in {3..5};
do
    echo "running $folder_stored_parameterss/$folder_stored_parameterss$i"
    sbatch Rmpiscript-mn4_splitted_large $folder_stored_parameterss/$folder_stored_parameterss$i $outputSimulation "large"
    sbatch Rmpiscript-mn4_splitted_small $folder_stored_parameterss/$folder_stored_parameterss$i $outputSimulation "small"
done

simoncarrignon/spreadrt documentation built on July 8, 2019, 3:34 a.m.