DataFrames Exercise

You can do this exercise in a notebook or as a script (or even in the REPL!).

Load and Clean-up a DataFrame

Open the reduced Higgs ML dataset, that you will find in the assets area of this tutorial, if you cloned it or are running via Binder (or you can get it here).

N.B. To understand the structure of the data, have a look at the Higgs ML Website and the CERN Opendata Portal.

First, clean the dataset, replacing all of the numerical \(-999.0\) values with the correct Julia missing values.

Plot Primary Variables for Signal and Background

For the signal and background events (Label == 's', Label == 'b' respectively), plot the tau and lepton \(p_T\) and \((\eta, \phi)\).

This is a nice way to start to visualise if this is a significant variable for distinguishing signal from background.

If scatter plots are too busy, you could subsample the data using the usual Julia slice notation, [start:step:end].

Summary Values

Using the Statistics package, find the mean and standard deviation for the tau and lepton \(p_T\) (try to use the original data frame and groupby).

Derive Data and Plot

Calculate the distance metric between the tau and the lepton, \(d = \sqrt{(\eta_\tau - \eta_{lep})^2 + (\phi_\tau - \phi_{lep})^2}\)

Remember that \(\delta\phi\) should be normalised.

Next Steps

For this exercise we haven’t really time to do much more than this initial data exploration. If you wanted to take this further, a nice direction is to learn the basics of one of the nice Julia ML packages (MJL.jl or Flux.jl) and start training some ML networks to distinguish the Higgs events from the background. After all, that is exactly the raison d’être of this dataset!