tocarina/howto3/data/old.swp/03usingit/01setup.txt

81 lines
1.8 KiB
Plaintext

<subsection Setup>
<frame>
I use the Dataset provided in this Paper (arXiv:1902.09914)
<list>
<e>up to 600k Anti-#k_T# jets in the Training Set with:</e>
<e>#p_T# between $550 \cdot \textrm{GeV}$ and $650 \cdot \textrm{GeV}$</e>
<e>$R_{i}^{2} = \eta_{i}^{2} + \phi_{i}^{2} \leq {0.8}^{2}$</e>
<e>the 4 vectors in each event are sorted by #p_t#</e>
<e>and are preprocessed here into</e>
<l2st>
<e>#flag#: a constant</e>
<e>$\Delta{\eta}$: $\eta = \log{\left(\frac{p + p_{3}}{p - p_{3}} \right)} / 2$, and $\Delta{\eta} = \eta - \operatorname{mean}{\left(\eta \right)}$</e>
<e> $\Delta{\phi}$: $\phi = \operatorname{arctan_{2}}{\left(p_{2},p_{1} \right)}$, and $\Delta{\phi} = \phi - \operatorname{mean}{\left(\phi \right)}$</e>
<e>$lp_{T}$: $p_{T}^{2} = p_{1}^{2} + p_{2}^{2}$, and $lp_{T} = - \log{\left(\frac{p_{T}}{p_{T}^{jet}} \right)}$</e>
<ignore>
<e>flag (a constant)
<e>#Eq(eta,ln((p+p_3)/(p-p_3))/2)#
<e>#Eq(phi,atan2(p_2,p_1))#
<e>#Eq(ln(p_t_jet/p_t),ln(sqrt((p_1_jet**2+p_2_jet**2)/(p_1**2+p_2**2))))#
</ignore>
</l2st>
</list>
</frame>
<ignore>
</ignore>
<repeat w="['none']">
<frame title="Setup" label="setup_?i?">
<split>
<que wid="0.5">
<list>
<e>Preproccessing</e>
<l2st>
<e>Sort by the transverse momentum</e>
</l2st>
<e>Encoder</e>
<l2st>
<e>Learn a graph (topK: connect each node to K neighbours)</e>
<e>Run graph updates</e>
<e>4 nodes -> 1 node</e>
</l2st>
<e>Decoder</e>
<l2st>
<e>1 node -> 4 nodes </e>
<e>Run graph updates</e>
<e>Sort again by the transverse momentum</e>
</l2st>
</list>
</que>
<que wid="0.48">
<list>
<e>50k jets</e>
<e>Learning rate of #0.0003#</e>
<e>Batch size of 200</e>
<e>Train until the loss does not improve for 30 Epochs</e>
<e>Compression size of 7</e>
</list>
<i f="history200" wmode="True"></i>
</que>
</split>
</frame>
</repeat>