Table of Contents

Useful links

CP groups


Documents and notes

Presentations and meetings


  • PMG central page (MC20)
  • JETM2 non-allhadronic ttbar mc20 (JIRA)
  • MC20, MC21 & MC23 Sample List from Elisabeth for different physics processes


Ongoing studies

  • Understanding bias in JES in low truth-jet pT
  • Truth-jet matching (Brendon)
    • Check different truth jet matching schemes for response definition
    • Performance as a function of dR quality cut
  • Evaluating the impact of soft lepton information (Prajita)
    • Performance with LeptonID
    • Apply additional selection on soft electron
  • Auxiliary tasks (Brendon)
    • Fixing weights of TruthOrigin classes
    • Add TruthParticleID
  • Model architecture (Brendon)
    • Attention mechanisms
    • Hyperparameter optimization

Samples and models on SDF

TDD output files:

# ttbar non-all hadronic sample in PHYSVAL format; 600 GB

# ttbar samples in FTAG format; 900 GB

The following variables are available in the latest PHYSVAL TDD samples:

Group			 jets
Feature			(D)ata type)
GN120220509_pb:			(float32)
GN120220509_pc:			(float32)
GN120220509_pu:			(float32)
GN1Lep20220509_pb:			(float32)
GN1Lep20220509_pc:			(float32)
GN1Lep20220509_pu:			(float32)
GN2v00_pb:			(float32)
GN2v00_pc:			(float32)
GN2v00_pu:			(float32)
softElectron_ptr:			(float32)
softElectron_dr:			(float32)
softElectron_isoPt:			(float32)
softElectron_abseta:			(float32)
softElectron_d0Signed:			(float32)
softElectron_z0Signed:			(float32)
softElectron_d0SignedSig:			(float32)
softElectron_epht:			(float32)
softElectron_dphires:			(float32)
softElectron_eop:			(float32)
softElectron_rhad:			(float32)
softElectron_rhad1:			(float32)
softElectron_eratio:			(float32)
softElectron_weta2:			(float32)
softElectron_reta:			(float32)
softElectron_f1:			(float32)
softElectron_f3:			(float32)
softElectron_phf:			(float32)
SV1_correctSignificance3d:			(float32)
softMuon_pt:			(float32)
softMuon_dR:			(float32)
softMuon_eta:			(float32)
softMuon_phi:			(float32)
softMuon_qOverPratio:			(float32)
softMuon_momentumBalanceSignificance:			(float32)
softMuon_scatteringNeighbourSignificance:			(float32)
softMuon_pTrel:			(float32)
softMuon_ip3dD0:			(float32)
softMuon_ip3dZ0:			(float32)
softMuon_ip3dD0Significance:			(float32)
softMuon_ip3dZ0Significance:			(float32)
softMuon_ip3dD0Uncertainty:			(float32)
softMuon_ip3dZ0Uncertainty:			(float32)
JetFitterFlip_energyFraction:			(float32)
JetFitterFlip_mass:			(float32)
JetFitterFlip_significance3d:			(float32)
JetFitterFlip_deltaphi:			(float32)
JetFitterFlip_deltaeta:			(float32)
JetFitterFlip_massUncorr:			(float32)
JetFitterFlip_dRFlightDir:			(float32)
pt_btagJes:			(float32)
eta_btagJes:			(float32)
absEta_btagJes:			(float32)
scalarSumTrackPt:			(float32)
JetFitter_energyFraction:			(float32)
JetFitter_mass:			(float32)
JetFitter_significance3d:			(float32)
JetFitter_deltaphi:			(float32)
JetFitter_deltaeta:			(float32)
JetFitter_massUncorr:			(float32)
JetFitter_dRFlightDir:			(float32)
SV1_masssvx:			(float32)
SV1_efracsvx:			(float32)
SV1_significance3d:			(float32)
SV1_dstToMatLay:			(float32)
SV1_deltaR:			(float32)
SV1_Lxy:			(float32)
SV1_L3d:			(float32)
JetFitter_deltaR:			(float32)
JetFitterSecondaryVertex_displacement3d:			(float32)
JetFitterSecondaryVertex_displacement2d:			(float32)
JetFitterSecondaryVertex_mass:			(float32)
JetFitterSecondaryVertex_energy:			(float32)
JetFitterSecondaryVertex_energyFraction:			(float32)
JetFitterSecondaryVertex_minimumTrackRelativeEta:			(float32)
JetFitterSecondaryVertex_maximumTrackRelativeEta:			(float32)
JetFitterSecondaryVertex_averageTrackRelativeEta:			(float32)
JetFitterSecondaryVertex_maximumAllJetTrackRelativeEta:			(float32)
JetFitterSecondaryVertex_minimumAllJetTrackRelativeEta:			(float32)
JetFitterSecondaryVertex_averageAllJetTrackRelativeEta:			(float32)
rnnip_pu:			(float16)
rnnip_pc:			(float16)
rnnip_pb:			(float16)
DL1r_pu:			(float16)
DL1r_pc:			(float16)
DL1r_pb:			(float16)
DL1r20210824r22_pb:			(float16)
DL1r20210824r22_pc:			(float16)
DL1r20210824r22_pu:			(float16)
dipsLoose20210729_pb:			(float16)
dipsLoose20210729_pc:			(float16)
dipsLoose20210729_pu:			(float16)
dipsLoose20220314v2_pb:			(float16)
dipsLoose20220314v2_pc:			(float16)
dipsLoose20220314v2_pu:			(float16)
dipsLoose20210729flip_pb:			(float16)
dipsLoose20210729flip_pc:			(float16)
dipsLoose20210729flip_pu:			(float16)
DL1dv00_pu:			(float16)
DL1dv00_pc:			(float16)
DL1dv00_pb:			(float16)
DL1dv01_pu:			(float16)
DL1dv01_pc:			(float16)
DL1dv01_pb:			(float16)
JetFitterFlip_nVTX:			(int32)
JetFitterFlip_nSingleTracks:			(int32)
JetFitterFlip_nTracksAtVtx:			(int32)
JetFitterFlip_N2Tpair:			(int32)
IP3D_nTrks:			(int32)
JetFitter_nVTX:			(int32)
JetFitter_nSingleTracks:			(int32)
JetFitter_nTracksAtVtx:			(int32)
JetFitter_N2Tpair:			(int32)
SV1_N2Tpair:			(int32)
SV1_NGTinSvx:			(int32)
JetFitterSecondaryVertex_nTracks:			(int32)
softMuon_isDefaults:			(int8)
rnnip_isDefaults:			(int8)
JetFitter_isDefaults:			(int8)
SV1_isDefaults:			(int8)
JetFitterSecondaryVertex_isDefaults:			(int8)
pt:			(float32)
eta:			(float32)
energy:			(float32)
mass:			(float32)
HadronConeExclTruthLabelPt:			(float32)
HadronConeExclTruthLabelLxy:			(float32)
truthJetPt:			(float32)
truthJetEta:			(float32)
truthJetM:			(float32)
n_tracks:			(int32)
GhostBHadronsFinalCount:			(int32)
GhostCHadronsFinalCount:			(int32)
HadronConeExclTruthLabelID:			(int32)
HadronConeExclExtendedTruthLabelID:			(int32)
PartonTruthLabelID:			(int32)
jetPtRank:			(int32)
primaryVertexDetectorZ:			(float32)
mcEventWeight:			(float32)
eventNumber:			(int64)
averageInteractionsPerCrossing:			(float32)
actualInteractionsPerCrossing:			(float32)
nPrimaryVertices:			(int32)
isRun3:			(int8)

Group			 tracks
Feature			(D)ata type)
expectInnermostPixelLayerHit:			(uint8)
expectNextToInnermostPixelLayerHit:			(uint8)
numberOfInnermostPixelLayerHits:			(uint8)
numberOfNextToInnermostPixelLayerHits:			(uint8)
numberOfInnermostPixelLayerSharedHits:			(uint8)
numberOfInnermostPixelLayerSplitHits:			(uint8)
numberOfPixelHits:			(uint8)
numberOfPixelHoles:			(uint8)
numberOfPixelSharedHits:			(uint8)
numberOfPixelSplitHits:			(uint8)
numberOfPixelDeadSensors:			(uint8)
numberOfSCTHits:			(uint8)
numberOfSCTHoles:			(uint8)
numberOfSCTSharedHits:			(uint8)
numberOfSCTDeadSensors:			(uint8)
numberOfTRTHits:			(uint8)
truthOriginLabel:			(int32)
truthVertexIndex:			(int32)
numberDoF:			(float16)
theta:			(float16)
qOverP:			(float32)
chiSquared:			(float32)
radiusOfFirstHit:			(float32)
pixeldEdx:			(float32)
pt:			(float32)
ptfrac:			(float32)
deta:			(float32)
abs_deta:			(float32)
z0RelativeToBeamspot:			(float32)
qOverPUncertainty:			(float32)
d0:			(float16)
z0SinTheta:			(float16)
eta:			(float16)
d0Uncertainty:			(float16)
z0SinThetaUncertainty:			(float16)
phiUncertainty:			(float16)
thetaUncertainty:			(float16)
dphi:			(float16)
dr:			(float16)
z0RelativeToBeamspotUncertainty:			(float16)
IP3D_signed_d0:			(float16)
IP3D_signed_z0:			(float16)
IP3D_signed_d0_significance:			(float16)
IP3D_signed_z0_significance:			(float16)

Umami-preprocessing datasets:

# UPP_1: ttbar non-all had FTAG1, 1.5mil jets; 1 GB

# UPP_2: ttbar non-all had PHYSVAL, 100mil jets w/ 10 < pt_true; 133 GB

# UPP_3: ttbar non-all had PHYSVAL, 49mil jets w/ 10 < pt_true < 50 GeV + 50mil jets w/ pt_true > 50 GeV; 112 GB

Trained models:

### Trained on UPP_1
# No auxiliary tasks

### Trained on UPP_2
# No auxiliary tasks, trained in 2 separate batches

# With auxiliary tasks

### Trained on UPP_3

SDF preliminaries

Migration to S3DF

In Jan/Feb 2024, we migrated from SDF to S3DF. Here are some helpful documentations:

  • S3DF page, including conda installation (link)
  • S3DF in the ATLAS grid (Wei Yang)

Note that our installation of conda is at /sdf/group/atlas/sw/conda.

Compute resources

SDF has a shared queue for submitting jobs via slurm, but this partition has extremely low priority. Instead, use the usatlas partition or request Michael Kagan to join the atlas partition.

You can use the command sacctmgr list qos name=atlas,usatlas format=GrpTRES%125 to check the resources available for each:

atlas has 2 CPU nodes (2x 128 cores) and 1 GPU node (4x A100); 3 TB memory
usatlas has 4 CPU nodes (4x 128 cores) and 1 GPU node (4x A100, 5x GTX 1080. Ti, 20x RTX 2080 Ti); 1.5 TB memory


For sdf

An environment needs to be created to ensure all packages are available. We have explored some options for doing this.

Option 1: stealing instance built for SSI 2023. This installs most useful packages but uses python 3.6, which leads to issues with h5py.
Starting Jupyter sessions via SDF web interface

  1. SDF web interface > My Interactive Sessions > Services > Jupyter (starts a server via SLURM)
  2. Jupyter Instance > slac-ml/SSAI

Option 2 (preferred): create your own conda environment. Follow the SDF docs to use ATLAS group installation of conda.
There is also a slight hiccup with permissions in the folder /sdf/group/atlas/sw/conda/pkgs, which one can sidestep by specifying their own folder for saving packages (in GPFS data space).
The TLDR is:

# Add the following lines to ~/.condarc file (create default file with conda config)
  - /gpfs/slac/atlas/fs1/d/<user>/conda_envs 
  - /gpfs/slac/atlas/fs1/d/<user>/conda_envs
  - defaults
  - conda-forge

This will allow you to also create environments saved in GPFS rather than the login node, which fills up quickly, using –prefix /gpfs/slac/atlas/fs1/d/<user>/conda_envs/env_name .

conda init # the previous will be added to your bashrc file 
conda env create -f bjr_v01.yaml # for example(bjr_v01) conda install jupyter

This env can be activated when starting a kernel in Jupyter by adding the following under Custom Conda Environment:

export CONDA_PREFIX=/sdf/group/atlas/sw/conda
export PATH=${CONDA_PREFIX}/bin/:$PATH
source ${CONDA_PREFIX}/etc/profile.d/
conda env list
conda activate bjr_v01

For the following packages (UPP, SALT, and puma) there are predefined conda environments available (made by Brendon). You can activate them with the usual conda activate command. They are here:

├── puma.yaml
├── salt.yaml
└── upp.yaml

Add the following to your .bashrc script as well

export PATH="/sdf/group/atlas/sw/conda/bin:$PATH"
export TMPDIR="${SCRATCH}"

For S3DF 

Setting up the conda environment has steps similar to those of SDF. To use the atlas group conda, use following steps

1.) Add the following in your .bashrc

export PATH="/sdf/group/atlas/sw/conda/bin:$PATH"

2.) Add the following to your .condarc

  - /fs/ddn/sdf/group/atlas/d/<user>/conda_envs/pkgs
 - /fs/ddn/sdf/group/atlas/d/<user>/conda_envs/envs
  - conda-forge
  - defaults
  - anaconda
  - pytorch

3.) source 

> source /sdf/group/atlas/sw/conda/etc/profile.d/

4.) cross check if conda is set correctly 

> which conda 
#following should appear if things went smoothly
conda ()
   \local cmd="${1-__missing__}";
   case "$cmd" in
       activate | deactivate)
           __conda_activate "$@"
       install | update | upgrade | remove | uninstall)
           __conda_exe "$@" || \return;
           __conda_exe "$@"

5.) Now follow the same steps as SDF to create and activate the conda environment

Producing H5 samples with Training Dataset Dumper


Overview (FTag Workshop Sept. 2023): presentation by Nikita Pond
Tutorial: TDD on FTag docs
General introduction to Athena in Run 3 (component accumulator): ATLAS Software Docs

TDD productions:

v2.1: TDD code, BigPanda
Run with grid-submit -c configs/PHYSVAL_EMPFlow_jer.json -e AntiKt4EMPFlow_PtReco_Correction_MV2c10_DL1r_05032020.root -i BTagTrainingPreprocessing/grid/inputs/physval-jer.txt bjer
NOTE: you will need to copy the PtReco histogram into the top level of training-dataset-dumper


We are using a custom fork of training-dataset-dumper, which is used to extract h5 files from derivations - we are currently using FTAG1 / PHYSVAL / JETM2.
The fork in slac-bjr has our working code to dump datasets based on the AntiKt4TruthDressedWZJets container and EMPFlow as well as UFO jets.

The main pieces of code that we are developing for the dumper are:

Executable/Athena interfaceBTagTrainingProcessing/bin/ca-dump-regression
Top level configurationconfigs/regression_EMPFlow*.jsonconfigs/regression_UFO.json

Class definitions:

  • SingleBTagAlg.*: C++ entry point, called by
    • processSingleBTagEvent.*: templated method to apply tools based on config file
    • SingleBTagTools:
    • SingleBTagConfig:

We are currently using Athena release 23.0.26, this comes before Sam Van Stroud made a series of changes to the SoftElectronTruthDecoratorAlg in FTag tools that mostly broke things (in release 23.0.34 and 23.0.39). Note that Athena needs to be checked out, rather than AnalysisBase.

Component accumulator

The component accumulator currently uses two algorithms for b-jet regression:

  • JetMatcherAlg: truth-reco jet matching

The following aliases can be used for convenient local testing:

alias test_PHYSVAL="ca-dump-regression -c configs/regression_EMPFlow.json -j AntiKt4EMPFlow --soft-lepton -o PHYSVAL.EMPflow.test.100.h5 -m 100 /gpfs/slac/atlas/fs1/d/bbullard/BjetRegression/DAOD/PHYSVAL/mc20_13TeV/DAOD_PHYSVAL.29398576._003055.pool.root.1"
alias test_JETM2_EMPFlow="ca-dump-regression -c configs/regression_EMPFlow.json -j AntiKt4EMPFlow -o JETM2.EMPflow.test.100.h5 -m 100 /gpfs/slac/atlas/fs1/d/bbullard/BjetRegression/DAOD/JETM2/mc20_13TeV/DAOD_JETM2.35974512._003603.pool.root.1"

alias JETM2_test_file="/gpfs/slac/atlas/fs1/d/bbullard/BjetRegression/DAOD/JETM2/mc20_13TeV/DAOD_JETM2.35974512._003603.pool.root.1"
alias PHYSVAL_test_file="/gpfs/slac/atlas/fs1/d/bbullard/BjetRegression/DAOD/PHYSVAL/mc20_13TeV/DAOD_PHYSVAL.29398576._003055.pool.root.1"

Some predefined decorations are in TDD docs.

Local files

A set of test derivation files can be found at:

# FTAG1 mc21 (13.6 TeV)

# PHYSVAL mc20 (13 TeV)

# JETM2 mc20 (13 TeV) [First test sample of new definition]

The current set of available Ntuples is available on:

├── nonallhad_PHYSVAL_mc20_410470 # High stats PHYSVAL derivation 
├── Rel22_ttbar_AllHadronic
├── Rel22_ttbar_DiLep
└── Rel22_ttbar_SingleLep

Adding Lepton & soft lepton to the TDD output

I (Prajita) am using the tools that already exist in the TDD.

Adding Electron and soft Electron

As shown below, two new algorithms are added to the same component accumulator. The first algo adds 10 electrons present near a jet (dR<0.4). The second and third lines add the algorithm that does soft electron tagging. The soft electron is defined as a single electron near a jet with the maximum probability of coming from a heavy flavor decay. This probability is evaluated by a e/gamma DNN tool. 


Adding soft muon

Soft muon information is available as Aux variables in FTAG DAOD, and these can be added just by adding them to configs as "float" to "btag" variables.

example config of adding SoftElectron and Electron information. softmuon info also added in the same way as soft el.

Calibration/Selection to different objects

jets: config to calibration/selection

-Reco Jets

  • rel 22 pre-recommendations
  • baseline kinematic selection pT>20GeV & |eta|<2.5
  • min_jvt 0.5
  • event level jet cleaning

-truth jet selection

  • "pt_minimum": 2e3,
    "abs_eta_maximum": 2.5,
    "dr_maximum": 0.4
  • 20 GeV truth jetpT applied at the umami preprocessing step


Details of electron selection

for soft electron selection, "Loose" identification working point, and all of these selections.

soft muon

Soft muon tagging is done in Athena by the FTAGSoftMuon Tagger tool. combined muon with basic kinematic selection pT > 5 GeV ‣ |η| < 2.5 ‣ d0 < 4 mm 

Samples with electron, soft electron, and soft muon information

base_sdf_dir/nonallhad_PHYSVAL_mc20_410470 This was processed on mc20 all hadronic ttbar samples, to be replaced by mc21 or mc23 13p6 samples. 

Plotting with Umami/Puma (Work In Progress)

NOTE: we have not yet used umami/puma for plotting or preprocessing, but it can in principle be done and we will probably try to get this going soon.

Plotting with umami

Umami (which relies on puma internally) is capable of producing plots based on yaml configuration files. 
The best (read: only) way to use umami out of the box is via a docker container. To configure on SDF following the docs, add the following to your .bashrc:

export SDIR=/scratch/${USER}/.singularity
mkdir $SDIR -p

To install umami, clone the repository (which is forked in the slac-bjr project):
git clone ssh://

Now setup umami and run the singularity image:

singularity shell /cvmfs/

Plotting exercises can be followed in the umami tutorial.

Plotting with puma standalone

Puma can be used to produce plots in a more manual way. To install, there was some difficulty following the nominal instructions in the docs. What I (Brendon) found to work was to do:

conda create -n puma python=3.11
pip install puma-hep
pip install atlas-ftag-tools

This took quite some time to run, so (again) save yourself the effort and use the precompiled environments.


SALT likes to take preprocesses data file formats from Umami (though in principle the format is the same as what's produced by the training dataset dumper).

An alternative (and supposedly better way to go) than using umami for preprocessing is to use the standalone umami-preprocessing (UPP).
Follow the installation instructions on git. This is unfortunately hosted on GitHub, so we cannot add it to the bjr-slac GitLab group (sad) A working fork
is available under Brendon's account and contains UPP configs for each sample configuration (bjr branch). These do not yet implement resampling, but there are 2 methods that can be explored (resampling and reweighting). 

You can cut to the chase by doing conda activate upp, or follow the docs for straightforward installation. 

To do the preprocessing to make the training, validation, and test sets, do:

preprocess --config configs/bjr.yaml --split all

Notes on logic of the preprocessing config files
There are several aspects of the preprocessing config that are not documented (yet). Most things are self explanatory, but note that:

  • within the FTag software, there exist flavor classifications (for example lquarkjets, bjets, cjets, taujets, etc). These can be used to define different sample components. Further selections
    based on kinematic cuts can be made through the region key.
  • It appears only 2 features can be specified for the resampling (nominally pt and eta).
  • The binning also appears not to be respected for the variables used in resampling.

Model development with SALT

The slac-bjr git project contains a fork of SALT. One can follow the SALT documentation for general installation/usage. Some specific notes can be found below:

Creating conda environment

One can use the environment salt that has been setup in /gpfs/slac/atlas/fs1/d/bbullard/conda_envs,
otherwise you may need to use conda install -c conda-forge jsonnet h5utils. Note that this was built using the latest master on Aug 28, 2023.

SALT also uses comet for logging, which has been installed into the salt environment (conda install -c anaconda -c conda-forge -c comet_ml comet_ml).
Follow additional instructions for configuration on the SALT documentation.

If using the latest main branch from Dec 11 2023, after SALT model refactoring has occurred, there are some issues with CUDA version compatibility with SDF.
PyTorch v2.0.1 used in SALT requires newer CUDA version than is available (11.4). One may need to do the following:

conda uninstall pytorch
conda install pytorch torchvision torchaudio cudatoolkit=11.4 -c pytorch

Note that the following error may appear if following the normal installation:

ImportError: cannot import name 'COMMON_SAFE_ASCII_CHARACTERS' from 'charset_normalizer.constant'

To solve it, you can do the following (again, this is already available in the common salt environment):

conda install chardet
conda install --force-reinstall -c conda-forge charset-normalizer=3.2.0

Interactive testing

It is suggested not to use interactive nodes to do training, but instead to open a terminal on an SDF node by:

  • Starting a Jupyter notebook (with a conda environment that has jupyter notebook installed, not necessarily the SALT conda env).
    A singularity container used in the jupyter notebook will propagate to the terminal session, which is incompatible with SALT.
  • New > Terminal. From here, you can test your training configuration.

There are some additional points to be aware of:

  • Running a terminal on an SDF node will reduce your priority when submitting jobs to slurm
  • Be careful about the number of workers you select (in the PyTorch trainer object) which should be <= the number of CPU cores you're using (using more CPU cores parallelizes the data loading,
    which can be the primary bottleneck in training)
  • The number of requested GPUs should match the number of devices used in the training.
  • The number of jets you use in the test training should be larger then the batch size (2x works fine)

The following code can be used to test SALT training (it's basically the same thing you have in the slurm submission script):

salt fit -c configs/<my_config>.yaml --data.num_jets_train <small_number> --data.num_workers <num_workers>

For training on slurm

See the SALT on SDF documentation (also linked at the top of this page) and example configs in the SALT fork in the slac_bjr GitLab project.
For additional clarity, see the following description of the submission scripts. Change the script as follows

#SBATCH --output="<your_output_path>/out/slurm-%j.%x.out"
#SBATCH --error="<your_output_path>/out/slurm-%j.%x.err

export COMET_API_KEY=<your_commet_key>
export COMET_WORKSPACE=<your_commet_workspace>
export COMET_PROJECT_NAME=<your_project_name>

cd <your_path_to_salt_directory>

from top level salt directory can use following command to launch a slurm training job in sdf

sbatch salt/ 

Both slurm-%j.%x.err andslurm-%j.%x.out files will start to fill up 

You can use standard sbatch commands from SDF documentation to understand the state of your job. 

Comet Training Visualization

In your comet profile, you should start seeing the live update for the training which looks as follows. The project name you have specified in the submit script appears under your
workspace which you can click to get the graphs of live training updates.

Training Evaluation

Follow salt documentation to run the evaluation of the trained model in the test dataset. There is a separate batch submission script used for the model evaluation, but is very similar to what is used in the model training batch script.
The main difference is in the salt command that is run (see below). It will produce a log in the same directory as the other log files, and will produce a new output h5 file alongside the one you pass in for evaluation.

The important points are the following:

  • --data.test_suff: Specify a suffix for the sample that is produced from the PredictionWriter callback specified in the model config. There is a separate list of features to be saved to this new file, along with the model output,
    that can be used for studies of the model performance
  • --data.num_workers: you should use the same number of workers for the evaluation as for the training, since both are bottlenecked by the loading of the data
  • --data.test_file: technically can be either the training, testing, or evaluation h5 file. In principle the testing file is created for this purpose. Philosophically, you want to keep this dataset separate so that you don't induce some kind of bias as you manually determine the hyper parameter optimization
  • --ckpt_path: the specific model checkpoint you want to use. Out of the box, SALT should be picking the checkpoint with the lowest validation loss, but this has been found to not be very reliable. So always do it manually to be sure you know what model state you are actually studying. 
srun salt test -c logs/<model>/config.yaml --data.test_file <path_to_file_for_evaluation>.h5 --data.num_jets_train <num_jets> --data.num_workers 20 --trainer.devices 1 --data.test_suff eval --ckpt_path logs/<model>/ckpts/epoch=<epoch>.ckpt

We are developing some baseline evaluation classes in the analysis repository to systematically evaluate model performance. 

To do: develop this, and add some notes on it (Brendon/Prajita)

Miscellaneous tips

You can grant read/write access for GPFS data folder directories to ATLAS group members via the following (note that this does not work for SDF home folder)

groups <username> # To check user groups
cd <your_directory> 
find . -type d|xargs chmod g+rx # Need to make all subdirectories readable *and* executable to the group

You can setup passwordless ssh connection following this page to create an ssh key pair. Then add this block to your ~/.ssh/config file (allows to tunnel directly to iana, but you can adapt to other hosts):

Host sdfiana
  HostName iana
  User <yourusername>
  IdentityFile ~/.ssh/<your private key>
  proxycommand ssh nc %h %p
  ForwardX11 yes

Discussion Summary 

Suggestions from M. Kagan Jan 23, 2023

  • Not to regress the neutrino pT directly, as the pT distribution might not be unimodal. Instead, try two different approaches 
    • Quantile regression with the median
    • Auxilary task based on classification of neutrino pT class prediction, i.e., predict the bin 0-1 for the ratio of neutrino pT & jet pT. Also, add overflow bin 
  • loss function based on median rather than mean square error, ie, MAE rather than MSE

  • No labels