CMS Machine Learning at Level 1 Trigger Tutorial

Setup

This tutorial is designed to run on lxplus8 using a Python environment from CMSSW. In addition we install some extra Python packages, and we use Xilinx's Vivado HLS software from a container on /cvmfs. This tutorial is designed to get you working on the notebooks with minimal setup. For real ML@L1T development work we recommend using a high-memory, high single-core-performance PC with the latest Xilinx tools.

Note with the recent switch to lxplus9 as the alias for lxplus, you need to specify lxplus8 to run this tutorial!

To set up the environment and launch the jupyter notebooks:

From your laptop, open a terminal and use your CERN computing account to ssh to lxplus8, e.g.:

ssh <user>@lxplus8.cern.ch

Note which lxplus8 node your are connected to, e.g. lxplus812.cern.ch, we will need to refer to the exact server in a later step. From the terminal on lxplus8, first clone this repository

git clone https://gitlab.cern.ch/fastmachinelearning/cms_mlatl1t_tutorial
cd cms_mlatl1t_tutorial

Setup the software environment and launch the Jupyter Hub with:

bash start_notebooks.sh # for bash
# source start_notebooks.sh # for zsh

Take note of the output:

    Or copy and paste one of these URLs:
        http://localhost:8888/?token=<long string>
     or http://127.0.0.1:8888/?token=<long string>

The port may or may not be 8888, depending on other users of the machine. Note which port your are given.

Then from another terminal on your laptop, forward the port on lxplus to your laptop like this:

ssh -N -f -L <local port>:localhost:<lxplus port> <user>@lxplus8<node>.cern.ch

e.g.

ssh -N -f -L 8888:localhost:8888 <user>@lxplus812.cern.ch

You made need to change the local port (left hand side) if you see a message like this:

bind [127.0.0.1]:8888: Address already in use
channel_setup_fwd_listener_tcpip: cannot listen to port: 8888
Could not request local forwarding.

You made need to change the remote port (right hand side) if the output from bash start_notebooks.sh showed a different port number (use the one printed there).

Authors and acknowledgment

@ssummers @thaarres @alobanov