kmer-learn 0.1.0__tar.gz

kmer_learn-0.1.0/LICENSE

@@ -0,0 +1 @@
+TODO: Add license here.

kmer_learn-0.1.0/MANIFEST.in

@@ -0,0 +1,7 @@
+include README.md
+include LICENSE
+include pyproject.toml
+include setup.py
+recursive-include kmer *.h
+recursive-include kmer *.c
+recursive-include examples *.ipynb

kmer_learn-0.1.0/PKG-INFO

@@ -0,0 +1,187 @@
+Metadata-Version: 2.4
+Name: kmer-learn
+Version: 0.1.0
+Summary: A modern Python toolkit for classical sequence machine learning
+Author: Arsenii Zinkevich
+Project-URL: Homepage, https://github.com/synbioml/kmer-learn
+Project-URL: Repository, https://github.com/synbioml/kmer-learn
+Project-URL: Issues, https://github.com/synbioml/kmer-learn/issues
+Keywords: bioinformatics,machine-learning,sequence-analysis,kmer,svm,kernel,genomics
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: C
+Classifier: Topic :: Scientific/Engineering :: Bio-Informatics
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.20
+Requires-Dist: scipy>=1.7
+Requires-Dist: scikit-learn>=1.0
+Provides-Extra: fast
+Requires-Dist: rapidfuzz>=3.0; extra == "fast"
+Requires-Dist: parasail>=1.3; extra == "fast"
+Provides-Extra: io
+Requires-Dist: pyfastx>=0.8; extra == "io"
+Provides-Extra: test
+Requires-Dist: pytest>=7.0; extra == "test"
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Requires-Dist: rapidfuzz>=3.0; extra == "dev"
+Requires-Dist: parasail>=1.3; extra == "dev"
+Requires-Dist: jupyter; extra == "dev"
+Requires-Dist: matplotlib; extra == "dev"
+Requires-Dist: nbformat; extra == "dev"
+Requires-Dist: nbclient; extra == "dev"
+Dynamic: license-file
+
+# kmer-learn
+
+Classical machine learning primitives for nucleotide sequences — kernels, encoders, distances, models, and sequence perturbation, all in one composable Python package.
+
+> **⚠ API stability:** This package is in early development (v0.x). The public API is **not yet stable** — breaking changes may be introduced between minor versions until v1.0. Pin your dependency to an exact version (e.g. `kmer-learn==0.1.0`) if reproducibility matters.
+
+## Install
+
+```bash
+pip install kmer-learn
+```
+
+From source (requires a C compiler):
+
+```bash
+git clone https://github.com/synbioml/kmer-learn.git
+cd kmer-learn
+pip install -e .
+```
+
+Optional backends (strongly recommended for speed):
+
+```bash
+pip install rapidfuzz parasail   # 10-100x faster edit distances and alignments
+pip install pyfastx              # fast FASTA parsing for examples
+```
+
+## Features
+
+### Kernels
+- **GKMKernel** family — gapped k-mer kernels from gkmSVM / LS-GKM (Ghandi et al. 2014; Lee 2016). C-backed, OpenMP-parallel, with full/estimated/truncated mismatch schemes, 6 post-transforms (RBF, poly, sigmoid, exponential, …), 5 positional weighting kernels (triangular, Epanechnikov, Gaussian, Laplacian, Cauchy), reverse-complement indexing, sliding-window scan, and 3D windowed tensors.
+- **DistanceKernel** — turns any distance into a kernel via a post-transform.
+
+### Encoders (CSR output)
+- **SpectrumEncoder** — plain k-mer counts via rolling hash (k ≤ 12).
+- **GappyEncoder** — gappy k-mer counts with explicit masks (`"*--*"`) or gap ranges (`L=6, g_min=2, g_max=3`).
+- **MismatchEncoder** — mismatch-tolerant k-mer counts (Leslie, Eskin, Noble 2004).
+
+All encoders support `canonical_rc=True` for reverse-complement collapsing.
+
+### Distances
+- **Hamming**, **Levenshtein** (rapidfuzz backend + Python fallback).
+- **NeedlemanWunsch**, **SmithWaterman** (parasail backend + Python fallback), with custom substitution matrices (NUC4.4, BLOSUM62, …).
+
+### Models
+- **DifferentialKmerScorer** — Multinomial Naive Bayes on k-mer features, with auto-generated negatives via Shuffler/Chunker.
+- **KernelSVM** — SVM with a precomputed kernel (works with GKMKernel and DistanceKernel).
+- **LinearSVM** — Linear SVM on encoder features.
+- **KNNClassifier** — k-Nearest Neighbors with a sequence distance.
+
+### Sequence perturbation
+- **KmerShuffler** — k-mer-preserving shuffle via random Eulerian paths in the De Bruijn graph. Three endpoint modes (preserve / free / crop). Philox4×32-10 RNG, reproducible across `n_jobs`.
+- **Chunker** — block-level perturbation: split into chunks of size `[min, max]`, optionally reverse-complement each, shuffle, concatenate. Five residual-handling modes, two algorithms (random / backtrack).
+- **BaseBackgroundModel** — ABC for custom background models.
+
+### Utilities
+- `kmer.utils` — bit-packed k-mer helpers (`kmer_to_code`, `code_to_kmer`, `reverse_complement`, `canonical_code`).
+
+## Quick start
+
+```python
+from kmer.kernels import GKMKernel
+from kmer.models import KernelSVM
+
+# Train a gkm-SVM
+clf = KernelSVM(GKMKernel(L=10, k=6, d=3, kernel_type="truncated", use_rc=True), C=1.0)
+clf.fit(positives + negatives, [1]*len(positives) + [0]*len(negatives))
+preds = clf.predict(test_seqs)
+```
+
+```python
+from kmer.encoders import SpectrumEncoder
+from kmer.models import DifferentialKmerScorer
+from kmer.perturb import KmerShuffler
+
+# Differential k-mer scoring with dinucleotide-shuffled background
+scorer = DifferentialKmerScorer(
+    featurizer=SpectrumEncoder(k=6, canonical_rc=True),
+    background=KmerShuffler(k=2, seed=42),
+)
+scorer.fit(positives)
+top_motifs = scorer.kmer_scores_.sort_values(ascending=False).head(20)
+```
+
+```python
+from kmer.distance import Levenshtein, DistanceKernel
+from kmer.models import KNNClassifier
+
+# KNN with edit distance
+clf = KNNClassifier(Levenshtein(), n_neighbors=5)
+clf.fit(train_seqs, y_train)
+```
+
+## Examples (Vignettes)
+
+The `examples/` directory (top-level, next to `kmer/`) contains a series of cross-linked Jupyter notebooks. Each notebook starts with a vignette index linking to all others.
+
+| # | Notebook | Topic |
+|---|----------|-------|
+| 01 | `01_basic_kernel_matrix.ipynb` | GKMKernel: build, inspect, verify invariants |
+| 02 | `02_distance_metrics_and_kernels.ipynb` | Distance metrics (Hamming, Levenshtein, NW, SW) + DistanceKernel (RBF, PSD, KernelSVM) |
+| 03 | `03_svc_with_kernel.ipynb` | Train a gkm-SVM with KernelSVM |
+| 04 | `04_clustering_sequences.ipynb` | Hierarchical clustering with kernel distances |
+| 05 | `05_score_long_sequence.ipynb` | Sliding-window scan of a long sequence |
+| 06 | `06_weighted_kernel.ipynb` | WGKMKernel positional weighting (centered motif) |
+| 07 | `07_transform_and_comparison.ipynb` | All 3 schemes × 6 transforms, GKM vs WGKM |
+| 08 | `08_windowed_3d_tensors.ipynb` | WindowedGKMKernel 3D output (line plot) |
+| 09 | `09_spectrum_encoder_and_differential.ipynb` | SpectrumEncoder + DifferentialKmerScorer |
+| 10 | `10_gappy_encoder.ipynb` | GappyEncoder with masks, gap ranges, RC collapse |
+| 11 | `11_mismatch_encoder.ipynb` | MismatchEncoder and comparison to spectrum |
+| 12 | `12_shuffler_and_chunker.ipynb` | KmerShuffler + Chunker for negative-set generation |
+
+## Citation
+
+An article describing this package is in preparation. Until it is published, please cite the package as:
+
+> *kmer-learn: Classical machine learning primitives for nucleotide sequences.* (in preparation).
+
+For the mean time, if you use the package in your research, please cite the relevant foundational works listed below.
+
+## References
+
+The package builds on the following foundational works:
+
+- **gkmSVM** — Ghandi M, Lee D, Mohammad-Noori M, Beer MA. *Enhanced regulatory sequence prediction using gapped k-mer features.* PLoS Comput Biol. 2014;10(7):e1003711.
+- **LS-GKM** — Lee D. *LS-GKM: a new gkm-SVM for large-scale datasets.* Bioinformatics. 2016;32(14):2196–8.
+- **Mismatch kernel** — Leslie CS, Eskin E, Cohen A, Weston J, Noble WS. *Mismatch string kernels for discriminative protein classification.* Bioinformatics. 2004;20 Suppl 1:i467–76.
+- **Spectrum / gappy kernel** — Leslie CS, Eskin E, Weston J, Noble WS. *The spectrum kernel: a string kernel for SVM protein classification.* Pacific Symposium on Biocomputing. 2002:564–75.
+- **Dinucleotide shuffle** — Clote P. *Efficient calculation of the number of native states of a protein.* (2003, unpublished note); Altschul SF, Erickson BW. *Significance of nucleotide sequence alignments: a method for random sequence permutation.* Bull Math Biol. 1985;47(4):541–51.
+- **Philox4×32 RNG** — Salmon JK, Moraes MA, Dror RO, Shaw DE. *Parallel random numbers: as easy as 1, 2, 3.* SC '11.
+- **Multinomial Naive Bayes** — Manning CD, Raghavan P, Schütze H. *Introduction to Information Retrieval.* Cambridge University Press, 2008.
+- **Needleman-Wunsch** — Needleman SB, Wunsch CD. *A general method applicable to the search for similarities in the amino acid sequence of two proteins.* J Mol Biol. 1970;48(3):443–53.
+- **Smith-Waterman** — Smith TF, Waterman MS. *Identification of common molecular subsequences.* J Mol Biol. 1981;147(1):195–7.
+- **NUC4.4 matrix** — NCBI standard DNA scoring matrix.
+
+Third-party libraries used as optional backends:
+
+- **[rapidfuzz](https://github.com/maxbachmann/RapidFuzz)** — fast Levenshtein and Hamming distances.
+- **[parasail](https://github.com/jeffdaily/parasail)** — SIMD-accelerated sequence alignment (Daily, 2016).
+- **[scikit-learn](https://scikit-learn.org)** — SVM, Naive Bayes, KNN.
+- **[NumPy](https://numpy.org) / [SciPy](https://scipy.org)** — array and sparse-matrix infrastructure.
+
+## License
+
+To be specified.

kmer_learn-0.1.0/README.md

@@ -0,0 +1,145 @@
+# kmer-learn
+
+Classical machine learning primitives for nucleotide sequences — kernels, encoders, distances, models, and sequence perturbation, all in one composable Python package.
+
+> **⚠ API stability:** This package is in early development (v0.x). The public API is **not yet stable** — breaking changes may be introduced between minor versions until v1.0. Pin your dependency to an exact version (e.g. `kmer-learn==0.1.0`) if reproducibility matters.
+
+## Install
+
+```bash
+pip install kmer-learn
+```
+
+From source (requires a C compiler):
+
+```bash
+git clone https://github.com/synbioml/kmer-learn.git
+cd kmer-learn
+pip install -e .
+```
+
+Optional backends (strongly recommended for speed):
+
+```bash
+pip install rapidfuzz parasail   # 10-100x faster edit distances and alignments
+pip install pyfastx              # fast FASTA parsing for examples
+```
+
+## Features
+
+### Kernels
+- **GKMKernel** family — gapped k-mer kernels from gkmSVM / LS-GKM (Ghandi et al. 2014; Lee 2016). C-backed, OpenMP-parallel, with full/estimated/truncated mismatch schemes, 6 post-transforms (RBF, poly, sigmoid, exponential, …), 5 positional weighting kernels (triangular, Epanechnikov, Gaussian, Laplacian, Cauchy), reverse-complement indexing, sliding-window scan, and 3D windowed tensors.
+- **DistanceKernel** — turns any distance into a kernel via a post-transform.
+
+### Encoders (CSR output)
+- **SpectrumEncoder** — plain k-mer counts via rolling hash (k ≤ 12).
+- **GappyEncoder** — gappy k-mer counts with explicit masks (`"*--*"`) or gap ranges (`L=6, g_min=2, g_max=3`).
+- **MismatchEncoder** — mismatch-tolerant k-mer counts (Leslie, Eskin, Noble 2004).
+
+All encoders support `canonical_rc=True` for reverse-complement collapsing.
+
+### Distances
+- **Hamming**, **Levenshtein** (rapidfuzz backend + Python fallback).
+- **NeedlemanWunsch**, **SmithWaterman** (parasail backend + Python fallback), with custom substitution matrices (NUC4.4, BLOSUM62, …).
+
+### Models
+- **DifferentialKmerScorer** — Multinomial Naive Bayes on k-mer features, with auto-generated negatives via Shuffler/Chunker.
+- **KernelSVM** — SVM with a precomputed kernel (works with GKMKernel and DistanceKernel).
+- **LinearSVM** — Linear SVM on encoder features.
+- **KNNClassifier** — k-Nearest Neighbors with a sequence distance.
+
+### Sequence perturbation
+- **KmerShuffler** — k-mer-preserving shuffle via random Eulerian paths in the De Bruijn graph. Three endpoint modes (preserve / free / crop). Philox4×32-10 RNG, reproducible across `n_jobs`.
+- **Chunker** — block-level perturbation: split into chunks of size `[min, max]`, optionally reverse-complement each, shuffle, concatenate. Five residual-handling modes, two algorithms (random / backtrack).
+- **BaseBackgroundModel** — ABC for custom background models.
+
+### Utilities
+- `kmer.utils` — bit-packed k-mer helpers (`kmer_to_code`, `code_to_kmer`, `reverse_complement`, `canonical_code`).
+
+## Quick start
+
+```python
+from kmer.kernels import GKMKernel
+from kmer.models import KernelSVM
+
+# Train a gkm-SVM
+clf = KernelSVM(GKMKernel(L=10, k=6, d=3, kernel_type="truncated", use_rc=True), C=1.0)
+clf.fit(positives + negatives, [1]*len(positives) + [0]*len(negatives))
+preds = clf.predict(test_seqs)
+```
+
+```python
+from kmer.encoders import SpectrumEncoder
+from kmer.models import DifferentialKmerScorer
+from kmer.perturb import KmerShuffler
+
+# Differential k-mer scoring with dinucleotide-shuffled background
+scorer = DifferentialKmerScorer(
+    featurizer=SpectrumEncoder(k=6, canonical_rc=True),
+    background=KmerShuffler(k=2, seed=42),
+)
+scorer.fit(positives)
+top_motifs = scorer.kmer_scores_.sort_values(ascending=False).head(20)
+```
+
+```python
+from kmer.distance import Levenshtein, DistanceKernel
+from kmer.models import KNNClassifier
+
+# KNN with edit distance
+clf = KNNClassifier(Levenshtein(), n_neighbors=5)
+clf.fit(train_seqs, y_train)
+```
+
+## Examples (Vignettes)
+
+The `examples/` directory (top-level, next to `kmer/`) contains a series of cross-linked Jupyter notebooks. Each notebook starts with a vignette index linking to all others.
+
+| # | Notebook | Topic |
+|---|----------|-------|
+| 01 | `01_basic_kernel_matrix.ipynb` | GKMKernel: build, inspect, verify invariants |
+| 02 | `02_distance_metrics_and_kernels.ipynb` | Distance metrics (Hamming, Levenshtein, NW, SW) + DistanceKernel (RBF, PSD, KernelSVM) |
+| 03 | `03_svc_with_kernel.ipynb` | Train a gkm-SVM with KernelSVM |
+| 04 | `04_clustering_sequences.ipynb` | Hierarchical clustering with kernel distances |
+| 05 | `05_score_long_sequence.ipynb` | Sliding-window scan of a long sequence |
+| 06 | `06_weighted_kernel.ipynb` | WGKMKernel positional weighting (centered motif) |
+| 07 | `07_transform_and_comparison.ipynb` | All 3 schemes × 6 transforms, GKM vs WGKM |
+| 08 | `08_windowed_3d_tensors.ipynb` | WindowedGKMKernel 3D output (line plot) |
+| 09 | `09_spectrum_encoder_and_differential.ipynb` | SpectrumEncoder + DifferentialKmerScorer |
+| 10 | `10_gappy_encoder.ipynb` | GappyEncoder with masks, gap ranges, RC collapse |
+| 11 | `11_mismatch_encoder.ipynb` | MismatchEncoder and comparison to spectrum |
+| 12 | `12_shuffler_and_chunker.ipynb` | KmerShuffler + Chunker for negative-set generation |
+
+## Citation
+
+An article describing this package is in preparation. Until it is published, please cite the package as:
+
+> *kmer-learn: Classical machine learning primitives for nucleotide sequences.* (in preparation).
+
+For the mean time, if you use the package in your research, please cite the relevant foundational works listed below.
+
+## References
+
+The package builds on the following foundational works:
+
+- **gkmSVM** — Ghandi M, Lee D, Mohammad-Noori M, Beer MA. *Enhanced regulatory sequence prediction using gapped k-mer features.* PLoS Comput Biol. 2014;10(7):e1003711.
+- **LS-GKM** — Lee D. *LS-GKM: a new gkm-SVM for large-scale datasets.* Bioinformatics. 2016;32(14):2196–8.
+- **Mismatch kernel** — Leslie CS, Eskin E, Cohen A, Weston J, Noble WS. *Mismatch string kernels for discriminative protein classification.* Bioinformatics. 2004;20 Suppl 1:i467–76.
+- **Spectrum / gappy kernel** — Leslie CS, Eskin E, Weston J, Noble WS. *The spectrum kernel: a string kernel for SVM protein classification.* Pacific Symposium on Biocomputing. 2002:564–75.
+- **Dinucleotide shuffle** — Clote P. *Efficient calculation of the number of native states of a protein.* (2003, unpublished note); Altschul SF, Erickson BW. *Significance of nucleotide sequence alignments: a method for random sequence permutation.* Bull Math Biol. 1985;47(4):541–51.
+- **Philox4×32 RNG** — Salmon JK, Moraes MA, Dror RO, Shaw DE. *Parallel random numbers: as easy as 1, 2, 3.* SC '11.
+- **Multinomial Naive Bayes** — Manning CD, Raghavan P, Schütze H. *Introduction to Information Retrieval.* Cambridge University Press, 2008.
+- **Needleman-Wunsch** — Needleman SB, Wunsch CD. *A general method applicable to the search for similarities in the amino acid sequence of two proteins.* J Mol Biol. 1970;48(3):443–53.
+- **Smith-Waterman** — Smith TF, Waterman MS. *Identification of common molecular subsequences.* J Mol Biol. 1981;147(1):195–7.
+- **NUC4.4 matrix** — NCBI standard DNA scoring matrix.
+
+Third-party libraries used as optional backends:
+
+- **[rapidfuzz](https://github.com/maxbachmann/RapidFuzz)** — fast Levenshtein and Hamming distances.
+- **[parasail](https://github.com/jeffdaily/parasail)** — SIMD-accelerated sequence alignment (Daily, 2016).
+- **[scikit-learn](https://scikit-learn.org)** — SVM, Naive Bayes, KNN.
+- **[NumPy](https://numpy.org) / [SciPy](https://scipy.org)** — array and sparse-matrix infrastructure.
+
+## License
+
+To be specified.

kmer_learn-0.1.0/examples/01_basic_kernel_matrix.ipynb

@@ -0,0 +1,73 @@
+{
+ "nbformat": 4,
+ "nbformat_minor": 5,
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python",
+   "version": "3.13.5",
+   "mimetype": "text/x-python",
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "pygments_lexer": "ipython3",
+   "nbconvert_exporter": "python",
+   "file_extension": ".py"
+  }
+ },
+ "cells": [
+  {
+   "id": "40a8967d",
+   "cell_type": "markdown",
+   "source": "# 01 \u2014 GKMKernel: Basic Kernel Matrix\n\n**Vignette index:** | `**01**` GKMKernel basics | [`02` Distance metrics & kernels](02_distance_metrics_and_kernels.ipynb) | [`03` SVM with kernel](03_svc_with_kernel.ipynb) | [`04` Clustering](04_clustering_sequences.ipynb) | [`05` Long sequence scoring](05_score_long_sequence.ipynb) | [`06` Weighted (WGKM) kernel](06_weighted_kernel.ipynb) | [`07` Transforms & comparison](07_transform_and_comparison.ipynb) | [`08` Windowed 3D tensors](08_windowed_3d_tensors.ipynb) | [`09` Spectrum encoder & NB](09_spectrum_encoder_and_differential.ipynb) | [`10` Gappy encoder](10_gappy_encoder.ipynb) | [`11` Mismatch encoder](11_mismatch_encoder.ipynb) | [`12` Shuffler & chunker](12_shuffler_and_chunker.ipynb)\n\nThis vignette shows how to build a gkm kernel matrix and verify its basic invariants (symmetry, unit diagonal, PSD).",
+   "metadata": {}
+  },
+  {
+   "id": "0401a422",
+   "cell_type": "code",
+   "metadata": {
+    "execution": {
+     "iopub.status.busy": "2026-06-21T16:14:02.435770Z",
+     "iopub.execute_input": "2026-06-21T16:14:02.435946Z",
+     "shell.execute_reply": "2026-06-21T16:14:02.509236Z",
+     "iopub.status.idle": "2026-06-21T16:14:02.510335Z"
+    }
+   },
+   "execution_count": 1,
+   "source": "import numpy as np\nfrom kmer.kernels import GKMKernel\n\nseqs = [\n    \"ACGTACGTACGTACGTACGT\",\n    \"TTTTAAAAGGGGCCCCAAAA\",\n    \"ACGTTGCATGCATGCATGCA\",\n    \"CCCCGGGGTTTTAAAACCCC\",\n    \"ATATGCGCATATGCGCATAT\",\n    \"GAATTCGAATTCGAATTCGA\",\n]\n\nkern = GKMKernel(L=10, k=6, d=3, kernel_type=\"truncated\", use_rc=True)\nkern.set_references(seqs)\nK = np.asarray(kern.kernel())\nprint(\"Shape:\", K.shape)\nprint(\"Symmetric:\", np.allclose(K, K.T))\nprint(\"Unit diagonal:\", np.allclose(np.diag(K), 1.0))\nprint(\"Min eigenvalue:\", np.linalg.eigvalsh(K).min(), \"(PSD if >= 0)\")",
+   "outputs": [
+    {
+     "output_type": "stream",
+     "name": "stdout",
+     "text": "Shape: (6, 6)\nSymmetric: True\nUnit diagonal: True\nMin eigenvalue: 0.98676708653555 (PSD if >= 0)\n"
+    }
+   ]
+  },
+  {
+   "id": "81e9494f",
+   "cell_type": "code",
+   "metadata": {
+    "execution": {
+     "iopub.status.busy": "2026-06-21T16:14:02.511866Z",
+     "iopub.execute_input": "2026-06-21T16:14:02.512072Z",
+     "iopub.status.idle": "2026-06-21T16:14:02.529055Z",
+     "shell.execute_reply": "2026-06-21T16:14:02.528680Z"
+    }
+   },
+   "execution_count": 2,
+   "source": "# Cross-kernel: query sequences vs reference set\nqueries = [\"ACGTACGTACGTACGTACGTAC\", \"CCCCGGGGTTTTAAAACCCCAG\"]\nkern.set_references(seqs)\nKq = np.asarray(kern.kernel(X_query=queries))\nprint(\"Cross-kernel shape:\", Kq.shape)",
+   "outputs": [
+    {
+     "output_type": "stream",
+     "name": "stdout",
+     "text": "Cross-kernel shape: (2, 6)\n"
+    }
+   ]
+  }
+ ]
+}