EA_benchmarking

(Work In Progress) Documentation

• https://ea-benchmarking.readthedocs.io/en/latest/

Prerequisites

• MPI 1.0
• MPICH 3.3.2
• NEURON > 7.6 (see installation instructions here.
• CoreNeuron (see installation instructions here.
• SLURM
  • This is a limitation since it generally requires that you are running this code on a HPC cluster. If you'd like to try a version that doesn't require HPC, try this repo.
  • If you do check out this sub-repo, please reach out to me at [email protected] and I am happy to help with any techinical issues in set up! Thank you.
• NeuroGPU
  • CUDA (10 >)
  • GCC (8.3.0 >)
• Other requirements like (BluePyOpt)[https://github.com/BlueBrain/BluePyOpt] and (eFEL)[https://efel.readthedocs.io/en/latest/] are listed in requirements.tst

Installation

• conda env create -f env.yml --name benchmarking_env (preferred)

  • conda activate benchmarking_env

• pip install -r requirements.txt

  • You may encounter an issue with installing mpi4py if you do not have MPI and MPICH installed. See (this link).[https://stackoverflow.com/questions/28440834/error-when-installing-mpi4py]

• Then you'll need to compile NEURON/neuroGPU on your system so:

  • sh compile_neuron.sh

Run Demo Experiment

• cd into scripts/slurm_launch/
• sh neuron_batch_simple.sh

Running Experiments

• We create text files like the one below to specify experiments and these plans can be found in:
  • scripts/slurm_launch/plans
  • They look like this:

nGens=1
offspring=3000,3000,3000,3000,3000,3000,3000,3000,3000
cpuTrials=80
N=1,2,4,8,16,32,64,128,256
n_stims=6,6,6,6,6,6,6,6,6
n_sfs=20,20,20,20,20,20,20,20,20
clean=False 

• you can launch them by using:
• scripts/slurm_launch/meta_chain.sh

Plotting / Parsing Experiments

• once you have ran some experiments you should have files the outputs folders.

• cd scripts/plotting_scripts

• sh plot_all_neuron.sh

• This will parse the output logs from the experiments and then consolidate the results in the outputs folder.

• If you want to compare between different experiments in different outputs folders you can use meta_plot.py which parses the outputs in subdirectories : Neuron_EA, NeuroGPU_EA and CoreNeuron_EA.