tabpfn-graph 0.2.0__tar.gz

tabpfn_graph-0.2.0/.gitignore

@@ -0,0 +1,26 @@
+__pycache__/
+*.py[cod]
+.mypy_cache/
+.pytest_cache/
+.ruff_cache/
+.coverage
+htmlcov/
+build/
+dist/
+*.egg-info/
+.venv*/
+.benchmark-deps/
+.env
+.env.*
+!.env.example
+*.key
+*.pem
+credentials.json
+.DS_Store
+.idea/
+.vscode/
+data/
+benchmark-results/
+logs/
+*.whl
+*.tar.gz

tabpfn_graph-0.2.0/CHANGELOG.md

@@ -0,0 +1,79 @@
+# Changelog
+
+## 0.2.0
+
+Breaking changes are listed first. The package had not been published, so no
+deprecation cycle is provided.
+
+### Breaking
+
+- `approximation_nodes` is replaced by `approximation_pivots`. Approximate
+  descriptors no longer build a random induced subgraph; they estimate the
+  whole-graph quantity (pivot-sampled betweenness, sampled-source distances,
+  rescaled triangle counts, truncated-spectrum heat traces).
+- WL initial labels default to the categorical node/edge attributes discovered
+  during `fit` (`wl_node_attributes="auto"`). Pass `None` for the previous,
+  purely structural behavior.
+- The WL vocabulary is pruned by document frequency and capped
+  (`wl_min_graph_count`, `wl_max_features`), so the table width no longer grows
+  with the training set.
+- Spectral descriptors default to the normalized Laplacian (`laplacian=`), so
+  the NetLSD time grid is comparable across graph sizes and
+  `spectral__algebraic_connectivity` is the normalized spectral gap.
+- `motifs__average_clustering` and `motifs__core_number__*` are removed; those
+  quantities were exact duplicates of `local_profile` columns.
+- Histogram boundaries are deduplicated, so discrete attributes get fewer bins
+  instead of structurally empty ones, and bin column counts change.
+- `networkx>=3.5` is required, and `transform` refuses to run under a different
+  NetworkX feature release than the one used at `fit`.
+- Python 3.11 is the minimum supported version. NetworkX 3.5 is 3.11+ only, so
+  the 3.10 support claimed by earlier metadata was not installable.
+
+### Added
+
+- Edge weight support: `edge_weight`, `edge_weight_agg`, and
+  `edge_weight_semantics`. Weighted strength profiles, weighted clustering,
+  weighted PageRank/betweenness/assortativity/shortest paths, and a weighted
+  Laplacian spectrum.
+- Direction-aware columns whenever any training graph is directed: reciprocity,
+  in/out degree summaries and correlation, strongly connected components,
+  largest-SCC fraction, acyclicity, and PageRank on a direction-preserving
+  projection of the native arcs.
+- `motifs__square_clustering` and `basic__is_bipartite`, so bipartite datasets
+  are not described entirely by identically-zero triangle columns.
+- Largest-component path descriptors (`paths__lcc_*`) and component size
+  summaries, which stay defined for disconnected graphs.
+- `undefined="nan"` to stop conflating an undefined descriptor with a genuine
+  zero.
+- `diagnostics_` plus `fit`-time warnings for ignored edge weights, mixed
+  directedness, extreme size heterogeneity, all-bipartite datasets, and
+  aggressive WL pruning.
+- `tabpfn_graph.column_report` and `prune_uninformative=True`, which drops
+  constant and exactly duplicated columns as a fit-learned schema decision.
+- `docs/graph-types.md`, covering which graph families need a non-default
+  configuration.
+
+### Fixed
+
+- Mixed directed/undirected batches produced silent NaN columns, or dropped the
+  in/out degree columns entirely, depending on which graph came first. The
+  schema now depends only on fitted state.
+- Directed PageRank read the native graph with the literal key `"weight"`, so a
+  configured `edge_weight` under any other name silently fell back to the
+  unweighted computation. It now runs on a direction-preserving projection that
+  normalizes the attribute onto the canonical key and aggregates parallel arcs.
+- Negative edge weights under `edge_weight_semantics="distance"` were converted
+  into infinite traversal costs, which silently removed the edge from path and
+  betweenness descriptors, leaked `inf` into the feature table with a `__valid`
+  flag of 1, and left the edge present in the degree and clustering
+  descriptors. Negative weights are now rejected under both semantics, during
+  `fit` and `transform`. A zero distance is treated as free traversal rather
+  than as an unreachable edge.
+- `build-system.requires` allowed hatchling 1.25/1.26, which cannot build the
+  PEP 639 license metadata this project declares.
+- Removed the deprecated `License ::` classifier, which PyPI rejects alongside
+  a license expression, and added `[project.urls]`.
+
+## 0.1.0
+
+Initial internal release.

tabpfn_graph-0.2.0/LICENSE

@@ -0,0 +1,202 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+

tabpfn_graph-0.2.0/NOTICE

@@ -0,0 +1,7 @@
+tabpfn-graph
+Copyright 2026 tabpfn-graph contributors
+
+This product is licensed under the Apache License, Version 2.0.
+TabPFN model checkpoints and third-party datasets are not part of this work and
+are subject to their respective terms.
+

tabpfn_graph-0.2.0/PKG-INFO

@@ -0,0 +1,281 @@
+Metadata-Version: 2.5
+Name: tabpfn-graph
+Version: 0.2.0
+Summary: Stable graph-to-table features for TabPFN and scikit-learn estimators
+Project-URL: Homepage, https://github.com/m-herre/tabpfn-graph
+Project-URL: Repository, https://github.com/m-herre/tabpfn-graph
+Project-URL: Issues, https://github.com/m-herre/tabpfn-graph/issues
+Project-URL: Changelog, https://github.com/m-herre/tabpfn-graph/blob/main/CHANGELOG.md
+Author: tabpfn-graph contributors
+License-Expression: Apache-2.0
+License-File: LICENSE
+License-File: NOTICE
+Keywords: graphs,networkx,scikit-learn,tabpfn
+Classifier: Development Status :: 3 - Alpha
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: <3.15,>=3.11
+Requires-Dist: joblib>=1.3
+Requires-Dist: networkx>=3.5
+Requires-Dist: numpy>=1.24
+Requires-Dist: pandas>=2.0
+Requires-Dist: scikit-learn>=1.4
+Requires-Dist: scipy>=1.10
+Requires-Dist: tabpfn>=2.0
+Provides-Extra: bench
+Requires-Dist: catboost>=1.2; extra == 'bench'
+Requires-Dist: ogb>=1.3; extra == 'bench'
+Requires-Dist: torch-geometric>=2.5; extra == 'bench'
+Requires-Dist: xgboost>=2; extra == 'bench'
+Provides-Extra: client
+Requires-Dist: tabpfn-client>=0.1; extra == 'client'
+Provides-Extra: dev
+Requires-Dist: build>=1.2; extra == 'dev'
+Requires-Dist: hatch>=1.12; extra == 'dev'
+Requires-Dist: mypy>=1.10; extra == 'dev'
+Requires-Dist: pandas-stubs>=2.0; extra == 'dev'
+Requires-Dist: pytest-cov>=5; extra == 'dev'
+Requires-Dist: pytest>=8; extra == 'dev'
+Requires-Dist: ruff>=0.7; extra == 'dev'
+Requires-Dist: twine>=5; extra == 'dev'
+Requires-Dist: types-networkx>=3.5; extra == 'dev'
+Provides-Extra: fast
+Requires-Dist: networkit>=11; extra == 'fast'
+Description-Content-Type: text/markdown
+
+# tabpfn-graph
+
+`tabpfn-graph` turns a collection of graphs into a schema-stable pandas table and applies
+TabPFN or any scikit-learn-compatible estimator. It targets single-target, graph-level
+classification and regression.
+
+```python
+from tabpfn_graph import GraphClassifier, GraphRegressor
+
+clf = GraphClassifier().fit(graphs_train, y_train)
+predictions = clf.predict(graphs_test)
+
+reg = GraphRegressor(estimator=my_pipeline).fit(graphs_train, y_train)
+```
+
+The representation uses established graph descriptors: Local Degree/Topological Profile
+statistics, Weisfeiler–Lehman subtree hashes over categorical node/edge labels, attribute
+distribution aggregation, and optional centrality, paths, motifs, Laplacian, and NetLSD
+summaries. This follows the graph-level descriptor evidence from
+[LTP](https://arxiv.org/abs/2305.00724) and [MOLTOP](https://arxiv.org/abs/2407.12136). The
+broader graph-to-table foundation-model pattern has also been explored for node tasks by
+[G2T-FM](https://arxiv.org/abs/2508.20906) and [TabPFN-GN](https://arxiv.org/abs/2512.08798);
+those papers do not imply that this package implements their node-level methods.
+
+Weighted, directed, bipartite, disconnected, and size-heterogeneous graphs each need a
+different configuration, and `fit` warns when the training data looks like a mismatch.
+`docs/graph-types.md` in the source repository states what each family needs.
+
+## Install
+
+```bash
+pip install tabpfn-graph
+```
+
+Python 3.11–3.14 is supported. The standard installation includes TabPFN, NetworkX (3.5 or
+newer, because WL subtree hashes changed in that release), NumPy, pandas, SciPy,
+scikit-learn, and joblib. Optional extras are:
+
+```bash
+pip install 'tabpfn-graph[fast]'    # Networkit for compatible primitives
+pip install 'tabpfn-graph[client]'  # use hosted clients as user-supplied estimators
+pip install 'tabpfn-graph[dev]'     # tests, lint, typing, packaging
+```
+
+The first local TabPFN fit may require accepting the checkpoint terms and downloading model
+weights. Package source code is Apache-2.0; TabPFN checkpoints have separate terms. More detail
+is available in `docs/model-access.md` in the source repository.
+
+## NetworkX quickstarts
+
+Classification with the default local TabPFN:
+
+```python
+import networkx as nx
+from tabpfn_graph import GraphClassifier
+
+graphs = [nx.path_graph(5), nx.cycle_graph(5), nx.star_graph(4), nx.complete_graph(5)]
+y = [0, 1, 0, 1]
+
+clf = GraphClassifier(random_state=0).fit(graphs, y)
+labels = clf.predict([nx.path_graph(7), nx.cycle_graph(7)])
+probabilities = clf.predict_proba([nx.path_graph(7), nx.cycle_graph(7)])
+```
+
+Regression uses the same extraction contract:
+
+```python
+from tabpfn_graph import GraphRegressor
+
+reg = GraphRegressor(random_state=0).fit(graphs, [1.2, 2.5, 0.8, 4.1])
+values = reg.predict(graphs)
+```
+
+The default local estimators are created lazily during `fit`, with TabPFN's local text
+transformation enabled. Importing or constructing `GraphClassifier()` does not access model
+weights.
+
+## PyTorch Geometric datasets
+
+PyG is intentionally optional. Pass a homogeneous iterable of `torch_geometric.data.Data`
+objects when it is installed:
+
+```python
+from torch_geometric.datasets import TUDataset
+from tabpfn_graph import GraphClassifier
+
+dataset = TUDataset(root="data/TU", name="MUTAG")
+graphs = [data for data in dataset]
+y = [int(data.y.item()) for data in dataset]
+model = GraphClassifier().fit(graphs[:150], y[:150])
+prediction = model.predict(graphs[150:])
+```
+
+PyG's common doubled-edge representation is recognized: if every non-loop arc has a reciprocal
+arc with matching multiplicity, each pair is collapsed into one undirected logical edge.
+
+## Standalone feature extraction
+
+```python
+from tabpfn_graph import GraphFeatureExtractor
+
+extractor = GraphFeatureExtractor(features="balanced", n_jobs=-1, random_state=0)
+X_train = extractor.fit_transform(graphs_train)
+X_test = extractor.transform(graphs_test)
+
+assert list(X_train.columns) == list(X_test.columns)
+```
+
+`X_train` and `X_test` are pandas DataFrames. Numeric values remain numeric, graph-level
+categoricals use pandas categorical dtype, and semantic documents use pandas string dtype.
+
+## Custom estimators
+
+No text encoding is inserted for custom estimators. For numeric-only feature tables:
+
+```python
+from sklearn.ensemble import RandomForestClassifier
+from tabpfn_graph import GraphClassifier
+
+model = GraphClassifier(
+    features="balanced",
+    estimator=RandomForestClassifier(n_estimators=500, random_state=0),
+).fit(graphs_train, y_train)
+```
+
+XGBoost and CatBoost-style estimators work the same way:
+
+```python
+from xgboost import XGBClassifier
+
+model = GraphClassifier(
+    features="balanced",
+    estimator=XGBClassifier(n_estimators=500, random_state=0),
+).fit(graphs_train, y_train)
+```
+
+CatBoost can consume categorical/text columns when configured with their column names or
+indices. Hosted TabPFN clients can be passed through `estimator=`; the package does not require a
+particular client API beyond sklearn-style `fit` and `predict`.
+
+For a text-aware sklearn pipeline, explicitly select and transform the document column:
+
+```python
+from sklearn.compose import ColumnTransformer
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.linear_model import LogisticRegression
+from sklearn.pipeline import make_pipeline
+from sklearn.preprocessing import StandardScaler
+from tabpfn_graph import GraphClassifier, GraphFeatureExtractor
+
+extractor = GraphFeatureExtractor(
+    features=("basic", "text"),
+    text_attributes=("node.description",),
+)
+preprocess = ColumnTransformer([
+    ("text", TfidfVectorizer(), "node_attr__description__document"),
+    ("numeric", StandardScaler(), ["basic__n_nodes", "basic__n_edges_native"]),
+])
+estimator = make_pipeline(preprocess, LogisticRegression(random_state=0))
+model = GraphClassifier(feature_extractor=extractor, estimator=estimator)
+```
+
+## Weighted and directed graphs
+
+Structural descriptors ignore edge weights unless you name the attribute, because a weight has
+no universal meaning. Path and betweenness descriptors need a traversal cost, so the semantics
+are explicit rather than guessed:
+
+```python
+extractor = GraphFeatureExtractor(
+    edge_weight="weight",
+    edge_weight_semantics="similarity",  # cost = 1 / w; use "distance" when w is a length
+)
+```
+
+That switches strength profiles, weighted clustering, PageRank, betweenness, assortativity,
+shortest paths, and the Laplacian spectrum to their weighted definitions. When `edge_weight` is
+left unset but the training graphs carry numeric edge attributes, `fit` warns.
+
+Directed input additionally gets reciprocity, in/out degree summaries and their correlation,
+strongly connected components, acyclicity, and PageRank on a direction-preserving projection of
+the native arcs; PageRank on the
+undirected projection is close to a rescaled degree. Mixed batches are supported and yield one
+schema, with undirected graphs described as their own symmetrization.
+
+## Diagnostics
+
+```python
+extractor = GraphFeatureExtractor().fit(graphs_train)
+extractor.diagnostics_        # sizes, directedness, bipartiteness, WL vocabulary, versions
+```
+
+`fit` warns about ignored edge weights, mixed directedness, extreme size heterogeneity,
+all-bipartite datasets, and aggressive WL pruning. `fit_transform` also records how many
+columns are constant or exactly duplicated on the training data, available standalone as
+`tabpfn_graph.column_report(frame)`. Setting `prune_uninformative=True` turns that into a
+fit-learned schema decision and drops those columns.
+
+## Feature selection
+
+Presets are `fast`, `balanced` (default), and `comprehensive`:
+
+- `fast`: native/basic topology and node, edge, graph metadata, and text aggregation.
+- `balanced`: fast plus clustering/core/LDP profiles and two-iteration hashed WL counts.
+- `comprehensive`: balanced plus centrality, paths, motifs, Laplacian, and NetLSD summaries.
+
+Or select groups directly:
+
+```python
+extractor = GraphFeatureExtractor(features=("basic", "wl", "spectral"))
+```
+
+Valid groups are `basic`, `local_profile`, `attributes`, `text`, `wl`, `centrality`, `paths`,
+`motifs`, and `spectral`. Expensive groups preflight against `max_exact_nodes=2000`. To make the
+change in semantics explicit, larger graphs require either a raised exact limit or
+`allow_approximate=True`, which switches to pivot-sampled betweenness, sampled-source
+distances, rescaled triangle counts, and a truncated-spectrum heat trace — estimators of the
+whole-graph quantity, recorded per graph in `approx__active`.
+
+## Scope
+
+Single-target graph-level classification and regression on homogeneous NetworkX or PyG batches.
+Heterogeneous graphs (`HeteroData`), temporal graphs, and multilabel targets are out of scope.
+
+This is an alpha release. The descriptors are established ones and the schema contract is
+tested, but the package is not backed by a broad benchmark study; treat it as a descriptor
+baseline rather than a method shown to beat tuned GBDT-on-descriptors or GNNs.
+
+Detailed feature semantics, graph-family guidance, evaluation guidance, and reproducible
+benchmarks are kept in the source repository under `docs/` and `benchmarks/`. Benchmark code is
+not part of the published wheel or source distribution.