feat: add ttH CP parity benchmark configuration

docs/quickstart.md

@@ -57,6 +57,22 @@ uv run gnn4colliders prepare --config-name config_hf_delphes
 The data source, revision, checksum, tree, feature branches, and split rules
 are documented in [configuration.md](configuration.md).
 
+The inclusive no-selection ttH CP benchmark has a dedicated configuration. It
+uses `ttH_NLO.root` as label 0 (CP-even) and `ttH_CPodd.root` as label 1
+(CP-odd), with all configured object branches passed through without event or
+object selections:
+
+```bash
+uv run gnn4colliders prepare \
+  --config-name config_tth_cp_even_odd \
+  data.num_workers=8
+```
+
+This task configuration enables absolute event weights for optimization
+because the NLO files contain signed weights. The original signed weights are
+still retained in event metadata and used by the configured evaluation
+semantics.
+
 ## Use your own ROOT file
 
 Preparation accepts Hydra overrides. Supply the ROOT file, tree, cache path,

src/gnn4colliders/config/application.py

@@ -7,6 +7,7 @@ import multiprocessing
 import os
 import shutil
 import tempfile
+from collections.abc import Sequence
 from concurrent.futures import ProcessPoolExecutor
 from pathlib import Path
 from typing import Any
@@ -187,10 +188,12 @@ def prepare(config: DictConfig) -> Path:
     labels_per_file = (
         list(labels) if isinstance(labels, (list, tuple)) else [labels] * len(files)
     )
+    if not isinstance(labels, (list, tuple)) and isinstance(labels, Sequence):
+        labels_per_file = list(labels)
     source = RootEventDataset(
         files,
         tree_name=str(data.tree_name),
-        label=labels,
+        label=labels_per_file,
         feature_branches=plain(data.get("feature_branches")),
         global_features=plain(data.get("global_features", [])),
         fold_var=str(data.get("fold_var", "eventNumber")),

src/gnn4colliders/configs/config_tth_cp_even_odd.yaml

@@ -0,0 +1,16 @@
+defaults:
+  - data: tth_cp_even_odd
+  - model: root_gnn/edge_network_binary
+  - task: tth_cp_binary
+  - trainer: default
+  - checkpoint: default
+  - inference: default
+  - environment: local
+  - distributed: single
+  - export: onnx
+  - _self_
+
+experiment:
+  name: tth_cp_even_odd
+logging:
+  level: INFO

src/gnn4colliders/configs/data/tth_cp_even_odd.yaml

@@ -0,0 +1,33 @@
+defaults:
+  - _self_
+
+# The two files are the inclusive no-selection ttH H->gamma gamma benchmark.
+# Override data.files/data.label when using a different local data layout.
+files:
+  - data/raw/ttH_NLO_download/samples/higgs/top-associated/tth/ttH_NLO.root
+  - data/raw/ttH_cp_even_odd/samples/higgs/top-associated/tth/ttH_CPodd.root
+label: [0, 1]
+tree_name: output
+feature_branches:
+  - [jet_pt, ele_pt, mu_pt, ph_pt, MET_met]
+  - [jet_eta, ele_eta, mu_eta, ph_eta, 0]
+  - [jet_phi, ele_phi, mu_phi, ph_phi, MET_phi]
+  - CALC_E
+  - [jet_btag, 0, 0, 0, 0]
+  - [0, ele_charge, mu_charge, 0, 0]
+  - NODE_TYPE
+object_types: [vector, vector, vector, vector, single]
+scales: [0.1, 1, 1, 0.1, 1, 1, 1]
+global_features: []
+fold_var: Number
+weight_var: weight
+batch_size: 1024
+num_workers: 0
+shuffle: true
+seed: 42
+cache:
+  path: outputs/ttH_cp_even_odd/graphs.pt
+splits:
+  train_folds: [0, 1, 2, 3, 4, 5, 6, 7]
+  validation_folds: [8]
+  test_folds: [9]

src/gnn4colliders/configs/model/root_gnn/edge_network_binary.yaml

@@ -1,5 +1,8 @@
-defaults:
-  - /model/root_gnn/edge_network
-  - _self_
-
+family: root_gnn
+backend: torch
+name: edge_network
+hid_size: 128
 out_size: 1
+n_layers: 2
+n_proc_steps: 4
+dropout: 0.1

src/gnn4colliders/configs/task/tth_cp_binary.yaml

@@ -0,0 +1,7 @@
+defaults:
+  - binary_classification
+  - _self_
+
+# NLO samples contain signed weights. Absolute weights make the optimization
+# denominator stable while the original signed weights remain in metadata.
+absolute_weights: true