lgray

from uproot_methods import TLorentzVectorArray
import awkward as ak
from coffea.nanoaod import NanoEvents

def make_labeled_p4(x, indices, itype):
    p4 = TLorentzVectorArray.from_ptetaphim(x.pt, x.eta, x.phi, x.mass)
    return ak.JaggedArray.zip(p4=p4,
                              ptype=itype*x.pt.ones_like().astype(np.int),
                              pidx=indices
                              charge=x.charge)

lgray

import os
import os.path as osp
import math

import numpy as np
import torch
import gc
import torch.nn as nn
import torch.nn.functional as F
import torch_geometric.transforms as T

lgray

import torch
import torch_geometric

from torch import nn
from torch_geometric.nn.conv import EdgeConv

class EdgeNetWithCategoriesJittable(nn.Module):
    def __init__(self, input_dim=3, hidden_dim=8, output_dim=4, n_iters=1, aggr='add',
             norm=torch.tensor([1./500., 1./500., 1./54., 1/25., 1./1000.])):
        super(EdgeNetWithCategoriesJittable, self).__init__()

lgray

Sharing from a friend.....
This chilled me to the bone.  This problem is outrageously out of control....and this...so insightful.  TIME TO WAKE UP!
(I am not the original creator of this post)
“Can you tell the difference between these kids? Can you tell if they’re sad or asking for help? Could you pick them out in a crowd of 1000 people? By requiring kids to wear masks - whether it’s in stores, in schools, or a public place - you are robbing them of their ONE hope of being found in the event that they get taken - their faces.
Did you know that more than 2,000 missing-child reports are filed each day, and that many of them can be found when parents provide specific details about their physical appearance and a photo of THEIR FACE?
Did you KNOW that a child in AMERICA is over 66,000 x more likely to be human trafficked than to get COVID-19?
So DO YOU REALIZE that by requiring children over the age of 2 to wear a mask, you are making child abduction & human trafficking SO MUCH EASIER on the offenders?
Don’t bel

lgray

import torch
import numpy as np
import awkward as ak

def make_embedding_truth(truth_label_by_hits, device='cpu'):
    truth_ordering = np.argsort(truth_label_by_hits)
    uniques, counts = np.unique(truth_label_by_hits, return_counts=True)
    truth_indices = ak.JaggedArray.fromcounts(counts, truth_ordering) # delete noise?   
    pairwise_truth = truth_indices.choose(2)
    other_samples = np.zeros(pairwise_truth.counts.sum() + (pairwise_truth.counts==0).sum(), dtype=np.int64)

lgray

    def forward(self, x, batch: OptTensor=None):
        x = self.datanorm * x
        x = self.inputnet(x)        
        
        for ec in self.edgeconvs:
            edge_index = knn_graph(x, self.k, batch, loop=False, flow=ec.flow)
            x = ec(x, edge_index)
        
        out = self.output(x) # output the embedding directly from the dgcnn
        

lgray

#!/bin/bash

cmsrel CMSSW_11_1_0_pre6
cd CMSSW_11_1_0_pre6/src
cmsenv
git cms-init
git remote add thomas-cmssw [email protected]:tklijnsma/cmssw.git
git fetch thomas-cmssw dev-finecalo:dev-finecalo
git cms-merge-topic tklijnsma:dev-finecalo
git checkout dev-finecalo

lgray

pt_true = []
matched_pt_true = []
pred_pt = []

eta_true = []
matched_eta_true = []
pred_eta = []

phi_true = []
matched_phi_true = []

lgray

class GraphMETNetwork(nn.Module):
    def __init__ (self, continuous_dim, cat_dim, output_dim=1, hidden_dim=32, conv_depth=1):
        super(GraphMETNetwork, self).__init__()
        
        self.embed_charge = nn.Embedding(3, hidden_dim//4)
        self.embed_pdgid = nn.Embedding(8, hidden_dim//4)
        self.embed_pv = nn.Embedding(2, hidden_dim//4)
        
        self.embed_continuous = nn.Sequential(nn.Linear(continuous_dim,hidden_dim),
                                             nn.ReLU(),

lgray

class GraphMETNetwork(nn.Module):
    def __init__ (self, continuous_dim, cat_dim, output_dim=1, hidden_dim=32, conv_depth=1):
        super(GraphMETNetwork, self).__init__()
        
        self.embed_charge = nn.Embedding(3, hidden_dim//4)
        self.embed_pdgid = nn.Embedding(8, hidden_dim//4)
        self.embed_pv = nn.Embedding(2, hidden_dim//4)
        
        self.embed_continuous = nn.Sequential(nn.Linear(continuous_dim,hidden_dim//2),
                                              nn.ELU(),