Caffe_SampleRun

A sample environment to start playing with Caffe on the hepgpu1 servers at Manchester

---

VERY IMPORTANT: This repository will work only on the hepgpu1 servers since it contains explicit path references. Don't try using it anywhere else (it won't work).

Also, before getting started, if you have no idea what Caffe does, or some of the details are lost to you, take a look at this great intuitive short introductions to Multi-Layer Perceptrons and CNN:

Multi-Layer Perceptrons Tutorial

Convolutional Neural Network Tutorial

And if you want to know how all that applies to Caffe, please take a look at these short Caffe Tutorial Slides and at their Tour section.

---

How to use it:

Start using it Caffe is a simple as running the following two commands:

source sourceMe.sh

to setup the Caffe environment. You'll get some warning due to LArLite/LArCV trying to re-build the code (and you having no permissions to do it), but just ignore them.

Then to start Caffe, just run:

./runCaffe.sh

And then just wait for the CNN to work its magic.

You might want to do that in a screen session, since you probably want Caffe to be running for a long time.

Once Caffe has been running for a while, you can start making plots! Just run the following command:

./generatePlots.sh

to convert all the log files in your LogFiles directory in accuracy/loss plots (under your Plots directory).

Discrimination plots coming soon...

---

Samples:

You can find some signal (Heavy Sterile Neutrinos) samples at this path:

/data/LArCNN/Samples_LArCV/HeavySterileNeutrinos_DecayMuPi_Mass0.3_EventSize50k

And some background (BNB $\nu_{\mu}$) samples at this path:

/data/LArCNN/Samples_LArCV/BNB_NuMu_EventSize50k

---

Contacts:

If you have any query, contact salvatore.porzio@postgrad.manchester.ac.uk.