How do I generate my own images from some public data sets?

[mk123qwe]

such as Top Quark Tagging Reference Dataset https://zenodo.org/record/2603256#.XYM14egzZPY

[GiacomoBoldrini]

Hi mk123qwe, Thank you for taking attention into my project. Sorry for the lack of explanation in the ReadMe but i will answer your question. All these scripts take as input a .root file, output of the delphes detector stage from processes involving some quark or some gluons as output of the madgraph parton level generation (identified by a unique Particle ID ranging from 1-8 for soft light quark and anti-quark and 22 for gluons). Therefore you first need to generate parton, particle and detector level simulation with some MC generator (i used madgraph pythia and delphes) e.g in madgraph: generate p p > g g, generate p p > q q.

However i think you don’t need this stage as your link already provide you with some structured array of variables from the events...so i think you can skip all the scripts about images reading and create your own...i’m sorry but my code does not aim to generate images from a general data format but it is specific from .root arising from delphes!

Let me know if you need some more infos, i did not open your .h5 and atm i cannot...if you’re in trouble we can work out a solution according to what i’ve done in my scripts.

[mk123qwe]

Hi mk123qwe, Thank you for taking attention into my project. Sorry for the lack of explanation in the ReadMe but i will answer your question. All these scripts take as input a .root file, output of the delphes detector stage from processes involving some quark or some gluons as output of the madgraph parton level generation (identified by a unique Particle ID ranging from 1-8 for soft light quark and anti-quark and 22 for gluons). Therefore you first need to generate parton, particle and detector level simulation with some MC generator (i used madgraph pythia and delphes) e.g in madgraph: generate p p > g g, generate p p > q q.

However i think you don’t need this stage as your link already provide you with some structured array of variables from the events...so i think you can skip all the scripts about images reading and create your own...i’m sorry but my code does not aim to generate images from a general data format but it is specific from .root arising from delphes!

Let me know if you need some more infos, i did not open your .h5 and atm i cannot...if you’re in trouble we can work out a solution according to what i’ve done in my scripts.

I am not a physicist，but I'm interested in images and deep learning.I don't know how to get some jets images from these events.Most events are stored in h5 or root.I've noticed that some public data is available on the Internet, but it's hard for me to understand how these data are processed into images. Thank you for your help.

[GiacomoBoldrini]

Hi, sorry for the late reply i hope it's not too late. From your files .h5 description i see that you have acces to the top 200 components (particles) inside the jets. You will need the Pt, the Eta and Phi of each of them along with the Pt,eta,phi of the jet (just mean eta and phi and sum the pt of all the particles) then you can construct your image with the following: ` def create_jet_image(image_vector, bins=100, eta=[-0.8, 0.8], phi=[-0.8, 0.8]):

    etas = np.linspace(eta[0], eta[1], bins)
    phis = np.linspace(phi[0], phi[1], bins)

    im_to_plt = []
    count = 0
    for jet_im in tqdm(image_vector):

        im = np.zeros((bins,bins))
        for i in range(bins-1):
            for j in range(bins-1):
                eta_inf = etas[i]
                eta_sup = etas[i+1]
                phi_inf = phis[j]
                phi_sup = phis[j+1]
                for el in jet_im:
                    if (el[0] > eta_inf) & (el[0] < eta_sup) & (el[1] > phi_inf) & (el[1] < phi_sup):
                        im[i,j] += el[2]
        im_to_plt.append(im)
        count += 1

    return np.array(im_to_plt)`

image vector is a vector as follows: [[eta_1, phi_1, Pt_1], [eta_2, phi_2, Pt_2], ... , [eta_n, phi_n, Pt_n]] where 1,2,...,n are the n particles inside the jet.

im_to_plt is you bin x bin image.

Hope it was helpful