SigClustering issueshttps://xgitlab.cels.anl.gov/whopkins/sigclustering/-/issues2021-09-13T13:27:25-05:00https://xgitlab.cels.anl.gov/whopkins/sigclustering/-/issues/6Make latent space interpretable2021-09-13T13:27:25-05:00Walter HopkinsMake latent space interpretableSomehow make latent space interpretable and/or represent physically meaningful features (masses, momenta, possible of several input particles).Somehow make latent space interpretable and/or represent physically meaningful features (masses, momenta, possible of several input particles).https://xgitlab.cels.anl.gov/whopkins/sigclustering/-/issues/5Reinforcement-learning based clustering+feature extraction2021-09-13T13:17:07-05:00Walter HopkinsReinforcement-learning based clustering+feature extractionSee if RL can improve the clustering.See if RL can improve the clustering.https://xgitlab.cels.anl.gov/whopkins/sigclustering/-/issues/4Combining clustering and autoencoder loss2021-09-13T13:15:46-05:00Walter HopkinsCombining clustering and autoencoder lossAn autoencoder+clustering algorithm needs to be written to combine both a clustering and reconstruction loss. The clustering loss should be calculable without knowing the number of clusters a priori. As a starting point we could use k-means in combination with the gap statistic.An autoencoder+clustering algorithm needs to be written to combine both a clustering and reconstruction loss. The clustering loss should be calculable without knowing the number of clusters a priori. As a starting point we could use k-means in combination with the gap statistic.https://xgitlab.cels.anl.gov/whopkins/sigclustering/-/issues/3Delphes-based ttMET analysis2021-09-13T13:09:23-05:00Walter HopkinsDelphes-based ttMET analysisWe need to produce numpy arrays based on the ttMET Delphes samples. These should include high-level (mT, large-R jet mass, etc) and low-level (four vectors) quantities.We need to produce numpy arrays based on the ttMET Delphes samples. These should include high-level (mT, large-R jet mass, etc) and low-level (four vectors) quantities.Walter HopkinsWalter Hopkinshttps://xgitlab.cels.anl.gov/whopkins/sigclustering/-/issues/2Performance estimate of autoencoder on V100s2021-09-13T13:07:41-05:00Walter HopkinsPerformance estimate of autoencoder on V100sWe should measure the performance of the autoencoder with the ccMET data on NVIDIA V100s.We should measure the performance of the autoencoder with the ccMET data on NVIDIA V100s.Vangelis Kourlitisekourlitis@anl.govVangelis Kourlitisekourlitis@anl.govhttps://xgitlab.cels.anl.gov/whopkins/sigclustering/-/issues/1Torch-based clustering algorithm2021-09-13T13:05:26-05:00Walter HopkinsTorch-based clustering algorithmWe need a torch version of a k-means-like algorithm.We need a torch version of a k-means-like algorithm.John ChildersJohn Childers