Added deepsets (for real now) (documentation is WIP).

deepsets/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2023 Patrick Odagiu
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

deepsets/README.md

@@ -0,0 +1,2 @@
+# deepsets_synth
+Synthesizing the invariant version of the deepsets architecture using hls4ml. This architecture is meant to be used in a fast inference regime.

deepsets/bin/README.md

@@ -0,0 +1,11 @@
+## Deepsets Training Required Packages
+---
+
+Suggested to install the following using conda, with channel conda-forge.
+
+* numpy
+* tensorflow (only available on pip)
+* tensorflow-model-optimization (only available on pip)
+* matplotlib
+* scikit-learn
+* qkeras (only available on pip)

deepsets/bin/deepsets_hyperopt

@@ -0,0 +1,61 @@
+#!/usr/bin/env python
+
+# Run the training of the interaction network..
+import argparse
+import os
+import sys
+sys.path.append("..")
+
+from deepsets.hyperparam_optimisation import main
+parser = argparse.ArgumentParser(formatter_class=argparse.
+                                 ArgumentDefaultsHelpFormatter)
+parser.add_argument("--test_kfold", type=int, default=-1,
+                    help="Which kfold to use for test.")
+parser.add_argument("--gpu", type=str, default="",
+                    help="Sets the number of the GPU to run on.")
+args = parser.parse_args()
+
+if args.gpu:
+    os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
+
+kfolds = 5
+const = 32
+train_kfolds = [kfold for kfold in range(kfolds) if kfold != args.test_kfold]
+data_hyperparams = {
+    "fpath":        f'../../ki_data/jets_{const}constituents_ptetaphi_robust_fast',
+    "fnames_train": [f'jet_images_c{const}_minpt2.0_ptetaphi_robust_fast_{train_kfold}'
+                     for train_kfold in train_kfolds],
+    "fname_test":   f'jet_images_c{const}_minpt2.0_ptetaphi_robust_fast_{args.test_kfold}',
+}
+
+training_hyperparams = {
+    "batch":        [64, 128, 256, 512],
+    "epochs":       150,
+    "lr":           [0.0001, 0.01],
+    "valid_split":  0.2,
+}
+
+compilation_hyperparams = {
+    "optimizer": ['adam'],
+    "loss":      'softmax_with_crossentropy',
+    "metrics":   ['categorical_accuracy'],
+}
+
+model_hyperparams = {
+    "nnodes_phi": [24, 32],
+    "nnodes_rho": [32],
+    "activ": ['relu'],
+}
+
+args = {
+    "study_name":           f'synthable_{const}const_ptetaphi',
+    "storage":              'deepsets_inv',
+    "deepsets_type":        'invariant',
+    "outdir":               'deepsets_hyperopt',
+    "data_hyperparams":     data_hyperparams,
+    "training_hyperparams": training_hyperparams,
+    "model_hyperparams":    model_hyperparams,
+    "compilation":          compilation_hyperparams,
+}
+
+main(args)

deepsets/bin/deepsets_hyperopt_allgpu

@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+nohup python -u deepsets_hyperopt --test_kfold 4 --gpu 0 > logs/synthable_32const_ptetaphi_0.out &
+nohup python -u deepsets_hyperopt --test_kfold 4 --gpu 1 > logs/synthable_32const_ptetaphi_1.out &
+nohup python -u deepsets_hyperopt --test_kfold 4 --gpu 2 > logs/synthable_32const_ptetaphi_2.out &
+nohup python -u deepsets_hyperopt --test_kfold 4 --gpu 3 > logs/synthable_32const_ptetaphi_3.out &

deepsets/bin/deepsets_synth

@@ -0,0 +1,26 @@
+#!/usr/bin/env python
+
+# Run the testing of the interaction network.
+import argparse
+import sys
+import os
+sys.path.append("..")
+
+from deepsets.synthesize import main
+parser = argparse.ArgumentParser(formatter_class=argparse.
+                                 ArgumentDefaultsHelpFormatter)
+parser.add_argument("--model_dir", type=str, default="./trained_deepsets/test",
+                    help="The output directory name.")
+parser.add_argument("--gpu", type=str, default="0",
+                    help="Sets the number of the GPU to run on.")
+args = parser.parse_args()
+
+
+os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
+
+args = {
+    "model_dir":        args.model_dir,
+    "const_seed":       318
+}
+
+main(args)

deepsets/bin/deepsets_test

@@ -0,0 +1,28 @@
+#!/usr/bin/env python
+
+# Run the testing of the interaction network.
+import argparse
+import sys
+import os
+sys.path.append("..")
+
+from deepsets.test import main
+parser = argparse.ArgumentParser(formatter_class=argparse.
+                                 ArgumentDefaultsHelpFormatter)
+parser.add_argument("--model_dir", type=str, default="./trained_deepsets/test",
+                    help="The output directory name.")
+parser.add_argument("--gpu", type=str, default="0",
+                    help="Sets the number of the GPU to run on.")
+args = parser.parse_args()
+
+
+os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
+
+
+args = {
+    "model_dir":        args.model_dir,
+    "kfolds":           True,
+    "const_seed":       318
+}
+
+main(args)

deepsets/bin/deepsets_train

@@ -0,0 +1,70 @@
+#!/usr/bin/env python
+
+# Run the training of the deepsets network given the hyperparameters listed below.
+import argparse
+import os
+import sys
+sys.path.append("..")
+
+from deepsets.train import main
+parser = argparse.ArgumentParser(formatter_class=argparse.
+                                 ArgumentDefaultsHelpFormatter)
+parser.add_argument("--gpu", type=str, default="",
+                    help="Sets the number of the GPU to run on.")
+parser.add_argument("--test_kfold", type=int, default=-1,
+                    help="Which kfold to use for test.")
+parser.add_argument("--nbits", type=int, default=-1,
+                    help="The number of bits for quantisation of the model.")
+parser.add_argument("--outdir", type=str, default="test",
+                    help="The output directory name.")
+args = parser.parse_args()
+
+if args.gpu:
+    os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
+
+# Import the data.
+kfolds = 5
+nconst = 32
+
+data_folder_path = '../../ds_data'
+jets_path = f'jets_{nconst}constituents_ptetaphi_robust_fast'
+
+train_kfolds = [kfold for kfold in range(kfolds) if kfold != args.test_kfold]
+test_kfold = args.test_kfold
+data_hyperparams = {
+    "fpath":        os.path.join(data_folder_path, jets_path),
+    "fnames_train": [f'jet_images_c{nconst}_minpt2.0_ptetaphi_robust_fast_{train_kfold}'
+                     for train_kfold in train_kfolds],
+    "fname_test":   f'jet_images_c{nconst}_minpt2.0_ptetaphi_robust_fast_{test_kfold}',
+}
+
+training_hyperparams = {
+    "epochs":       300,
+    "batch":        256,
+    "lr":           0.0032,
+    "pruning_rate": 0.5
+}
+
+compilation_hyperparams = {
+    "optimizer": "adam",
+    "loss":      "categorical_crossentropy",
+    "metrics":   ["categorical_accuracy"],
+}
+
+model_hyperparams = {
+    "nnodes_phi": 32,
+    "nnodes_rho": 32,
+    "activ": 'relu',
+    "nbits": args.nbits,
+}
+
+args = {
+    "data_hyperparams":     data_hyperparams,
+    "training_hyperparams": training_hyperparams,
+    "model_hyperparams":    model_hyperparams,
+    "compilation":          compilation_hyperparams,
+    "outdir":               args.outdir,
+    "deepsets_type":        "sinvariant",
+}
+
+main(args)

deepsets/bin/deepsets_train_all_kfolds

@@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+
+nohup python -u deepsets_train --test_kfold 0 --gpu 0 --nbits 8 --outdir deepsinv_synthable_8bits_32const_ptetaphi_kfolds/kfold_0 > ./logs/deepsinv_synthable_8bits_32const_ptetaphi_kfold_0 &
+nohup python -u deepsets_train --test_kfold 1 --gpu 1 --nbits 8 --outdir deepsinv_synthable_8bits_32const_ptetaphi_kfolds/kfold_1 > ./logs/deepsinv_synthable_8bits_32const_ptetaphi_kfold_1 &
+nohup python -u deepsets_train --test_kfold 2 --gpu 2 --nbits 8 --outdir deepsinv_synthable_8bits_32const_ptetaphi_kfolds/kfold_2 > ./logs/deepsinv_synthable_8bits_32const_ptetaphi_kfold_2 &
+nohup python -u deepsets_train --test_kfold 3 --gpu 3 --nbits 8 --outdir deepsinv_synthable_8bits_32const_ptetaphi_kfolds/kfold_3 > ./logs/deepsinv_synthable_8bits_32const_ptetaphi_kfold_3 &
+nohup python -u deepsets_train --test_kfold 4 --gpu 2 --nbits 8 --outdir deepsinv_synthable_8bits_32const_ptetaphi_kfolds/kfold_4 > ./logs/deepsinv_synthable_8bits_32const_ptetaphi_kfold_4 &