Pith. sign in

REVIEW 3 major objections 7 minor 1 cited by

Anomaly Detection and Radio-frequency Interference Classification with Unsupervised Learning in Narrowband Radio Technosignature Searches

T0 review · 3 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read This paper claims that an unsupervised clustering layer, GLOBULAR, can cut false-positive events in narrowband radio technosignature searches by 99.3% while retaining most injected synthetic signals.

desk verdict A genuine and transparent RFI-clustering pipeline whose headline event-reduction numbers are dominated by a trivial baseline; the hit-level enrichment is real, but the evaluation needs a no-seed control and a random-removal comparison. read the letter →

arxiv 2411.16556 v2 pith:BOJWWPRZ submitted 2024-11-25 astro-ph.IM cs.LG

classification astro-ph.IMcs.LG
keywords technosignaturesradio-frequencyinterferenceanomalydetectionunsupervisedlearningHDBSCANnarrowbandsearchSETIoutlier
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper presents GLOBULAR, an unsupervised learning stage that sits between signal detection and spatial filtering in narrowband radio technosignature searches. It claims that running HDBSCAN clustering on a 13-dimensional feature description of detected hits removes common radio-frequency interference before the spatial filter is applied, reducing false-positive hits by 93.1% and false-positive events by 99.3% (from 288 to 2) on the 97-galaxy L-band sample from the earlier turboSETI-only search. The method also retains a substantial fraction of injected technosignature-like signals: 69 of 100 events versus 86 for the pipeline alone. If this holds, large surveys that currently require prohibitive manual follow-up could be winnowed automatically, and signals in RFI-dense spectral regions missed by the standard pipeline might become detectable.

What carries the argument

The load-bearing object is GLOBULAR, a pipeline built on HDBSCAN, a hierarchical density-based clustering algorithm that labels low-density points as noise. Hits are reduced to 13 statistical features (frequency, drift rate, SNR, spectral and temporal shape statistics, signal bandwidth, periodogram redness), then clustered in random batches of about 3,000 with the same 100 synthetic seed signals injected into every batch; pooling and re-clustering over eight epochs cuts the hit set by 93.1%. Cluster centroids are matched across batches using PCA followed by t-SNE and a final HDBSCAN pass, and the retained anomalies are passed through the standard FindEvent spatial filter that requires a signal to appear in on-target pointings but not off-target ones.

What would settle it

Run the same eight-epoch pipeline on the same data with the seed signals omitted, or with seeds of a different morphology (for example, broader bandwidths or different drift-rate signs), and compare the recovered injected-event count; if the false-positive reduction stays near 93% while the recovery rate collapses, the reported true-positive retention is carried by the seed design rather than by the clustering itself.

Watch

Extended reading notes

Core claim

GLOBULAR's central claim is that the false-positive burden of a standard narrowband technosignature pipeline can be reduced by treating detected hits as points in a hand-designed 13-dimensional feature space and clustering them with HDBSCAN after every batch. Because common RFI types form dense clusters while rare or genuinely anomalous signals fall outside them, iterative clustering for eight epochs isolates the anomalies; the surviving hits pass to the FindEvent spatial filter. On the L-band observations of 97 nearby galaxies, this removes 93.1% of the false-positive hits and cuts the false-positive event count from 288 to 2, while recovering 69 of 100 injected synthetic events compared with 86 recovered without the clustering stage. The paper further shows that the clusters correspond to recognizable RFI sources and reports one new event near the 21-cm line, suggesting the method can expose signals hidden by RFI occupancy.

Load-bearing premise

The load-bearing premise is that injecting 100 synthetic signals of exactly the target morphology into every batch fairly measures the method's ability to preserve real technosignatures; if the seeds simply teach the clusterer to keep anything shaped like them, the 69-of-100 recovery rate is partly an artifact of the test design.

Editorial extensions

If this is right

  • Future large technosignature surveys can add GLOBULAR clustering before the spatial filter to cut manual follow-up from hundreds of events to a handful.
  • RFI rejection becomes morphology-agnostic: common interference is identified purely by its prevalence, so new RFI types are grouped without retraining.
  • Injecting different seed morphologies lets the same pipeline be tuned to preserve other signal classes, not just drifting narrowband lines.
  • The method may recover events hidden by dense RFI: trimming RFI from on pointings exposed a new event near the 21-cm line that the standard pipeline did not find.
  • The trade-off is a lower recovery of injected signals (69 vs 86 events), suggesting that additional signal-preservation work, such as refinements in the spatial filter, is needed before full deployment.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the true-positive comparison (69 vs 86) is best read as a test of the seed-preservation mechanism, not of anomaly detection in general; a real technosignature whose morphology differs from the seeds could be clustered away as RFI.
  • Editorial inference: because the pipeline removes hits before the spatial filter, it should also be applicable to non-SETI radio surveys that need to reject human-generated interference, provided the feature space is re-tuned per instrument and band.
  • Editorial inference: the single new event near the 21-cm line raises a testable prediction: in RFI-dense regions, FindEvent may systematically miss faint narrowband signals, and trimming RFI from on pointings should reveal more such events; a future study could map detection completeness against spectral occupancy before and after trimming.
  • Editorial inference: the batching scheme turns quadratic HDBSCAN scaling into near-linear wall-clock behavior, so an all-sky survey with billions of hits appears computationally feasible, though the authors only demonstrate this on the 97-galaxy sample.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. GLOBULAR clustering applies HDBSCAN in an iterative batching scheme to turboSETI detections from Breakthrough Listen L-band observations of 97 galaxies, using a 13-dimensional hand-engineered feature space. Over eight epochs it removes clustered 'non-anomalous' hits, reducing the real-hit sample from 1,917,903 to 132,885 (93.1%) and the number of false-positive events found by the turboSETI FindEvent pipeline from 288 to 2 (99.3%). In an injection test on an M31 cadence, the pipeline recovers 264 of 296 synthetic hits and 69 of 100 synthetic events, compared with 86 events recovered by turboSETI alone. The authors additionally match RFI clusters across batches and report that one of the two surviving events is new relative to the Choza et al. (2024) search, which they interpret as evidence that GLOBULAR may reveal signals hidden by dense RFI.

Significance. If the central claims hold, the method would be practically valuable: reducing the false-positive event rate by two orders of magnitude would substantially ease manual follow-up in current and future SETI surveys, and the pipeline is designed to integrate with the widely used turboSETI framework. The paper has concrete strengths: it benchmarks against a published survey on public data, provides a code repository, defines the features and preprocessing in detail, shows manual inspection of representative clusters, and is transparent about the use of seed signals. However, the significance of the quantitative claims is currently limited by evaluation-design issues: the true-positive recovery is measured on the same morphology that the pipeline is explicitly instructed to keep through per-batch seed injection, and the false-positive event reduction is not compared with a trivial random-hit-removal baseline. Because both issues affect the headline numbers, the method's net benefit over simpler filtering is not yet established.

major comments (3)
  1. [§1.2, §3.1, Table 2] The true-positive recovery numbers (264 hits, 69 events) are confounded by construction because the recovery events and the per-batch seed signals are generated in the same way. Section 1.2 states that the 100 injected events have drift rates between -3 and +3 Hz/s and SNRs between 1,000 and 10,000, while Section 3.1 says that each batch receives an identical set of 100x13 features 'derived from 100 drifting narrowband signals synthesized with setigen' and that these are 'generated in the same way as, but are otherwise distinct from, the 100 events injected for recovery.' Section 3.1 further states that any non-synthetic signals placed in seeded clusters are retained for the next epoch. Thus the recovery events are exactly the morphology the pipeline has been instructed to keep, and the Table 2 comparison (69 vs. 86 events recovered) largely measures seed-template retention rather than the intrinsic ability of an unsupervised anomaly detector to preserve technosignature-like signals. I recommend rerunning GLOBULAR without seed injection and/or with a held-out test morphology (e.g., different drift range, bandwidth, or comb structure) and reporting the recovery rate under those conditions.
  2. [§5, Table 2] The headline event reduction from 288 to 2 is not benchmarked against a trivial baseline. Because an event requires hits in all three on-pointings, random removal of 93.1% of hits would leave roughly (0.069)^3 ≈ 3.3e-4 of the 288 events, i.e., about 0.1 events. A 99.3% event reduction is therefore approximately what any hit-removal filter would produce if event survival is dominated by the presence of three surviving hits. The meaningful claim is that GLOBULAR achieves this reduction while retaining true positives at a much higher rate than random filtering, but that retention is currently established only through the seeded-morphology test described in Major Comment 1. Please add a random-hit-removal baseline at the same 93.1% hit-reduction level and report both false-positive events and true-positive events under that baseline, so the reader can separate the effect of hit removal from the effect of the learned clustering.
  3. [§4, §5] The feature set and HDBSCAN hyperparameters are tuned on the same M31 observation that is later used for the large-scale true-positive injection test. Section 4 says the feature space was developed on a 3,068-hit toy data set from M31 and that the toy-model injections were 'generated using the same parameters as the 100 events for the large-scale model'; Section 5 then measures recovery on the M31 cadence. This creates a risk of overfitting the reported 69-of-100 recovery to one RFI environment and one signal recipe. If the method is intended to generalize to other targets, the recovery test should be performed on at least one cadence not used in any tuning step, or with a proper train/validation split.
minor comments (7)
  1. [Abstract vs. §1] The abstract expands GLOBULAR as 'Grouping Low-frequency Observations By Unsupervised Learning After Reduction' while Section 1 uses 'After Rejection'; please make the expansion consistent.
  2. [§1.2] There is a typo, 'to be an good example', which should read 'to be a good example'; likewise, 'theturboSETI values' should read 'the turboSETI values'.
  3. [§3.1, Figure 2] The text says each batch receives 'an identical set of 100x13 features', while Figure 2 labels this step 'add 100 synthetic hits per batch'; please align the wording because the seeds are feature vectors, not detected hits.
  4. [§2, Feature #8] The treatment of turning-point bandwidths that exceed 100 kHz by setting them to the boundary value could create artificial density at the edge of the feature range; please quantify how many hits are affected and whether the clustering results are sensitive to this censoring.
  5. [§3.2] The cross-batch cluster matching is evaluated only by visual inspection of a 10,000-centroid sample; a quantitative stability measure (e.g., cluster-matching reproducibility across bootstrap samples of centroids) would make the matching step more convincing.
  6. [§5] The new event near the 21-cm line is a false positive and is not a compelling technosignature candidate, as the text itself acknowledges; the later statement that this 'suggests that GLOBULAR clustering may be able to help the turboSETI algorithm find some of the true-positive events it previously missed' goes beyond what a single non-detection can support and should be softened.
  7. [Table 2] The recovery counts 69 vs. 86 are presented without uncertainty estimates; reporting a binomial confidence interval or a simple significance test would help the reader judge whether the 19.8% event-recovery loss is statistically meaningful.

Circularity Check

2 steps flagged · score 6.0 of 10

The claimed true-positive event recovery (69/100) is seeded by construction: the per-batch keeper seeds and the recovery events share the same setigen generation recipe, so the headline TP retention measures template matching rather than independent unsupervised detection.

  1. self definitional [Section 3.1 (Batching and Epochal Iteration), with results in Section 5 and Table 2]
    "To each batch, we inject an identical set of 100×13 features derived from 100 drifting narrowband signals synthesized with setigen. These signals are generated in the same way as, but are otherwise distinct from, the 100 events injected for recovery; the purpose of these is to “seed” clusters with morphologies we want to keep."

    The true-positive benchmark is not an independent probe of the algorithm's intrinsic ability: the 100 recovery events (Section 1.2: drift rates between −3 and 3 Hz/s, SNRs between 1,000 and 10,000) are generated with the same setigen recipe as the 100 seed signals injected into every batch. Since Section 3.1 states that non-synthetic signals placed in seeded clusters are retained 'as though they were outliers,' the recovery events are by construction placed in the keep-class. The reported 69/100 event recovery and the 19.8% loss relative to turboSETI alone therefore largely quantify template retention for a pre-specified morphology, not unsupervised detection of unknown technosignature shapes; a real signal outside the seed family could be discarded as RFI.

  2. fitted input called prediction [Section 4 (Application to Small Data Sets) and Section 1.2 (Data Sample)]
    "The 187.5-MHz M 31 observation used to test the toy model was a portion of the 800-MHz M 31 observation to which we later injected 100 synthetic events for large-scale retrieval testing (Section 1.2), and the 3 injections made for the toy model were generated using the same parameters as the 100 events for the large-scale model."

    The feature list and HDBSCAN hyperparameters (npts, ρpts, ϵm) were selected by iterative experimentation on the M31 toy data set, and the toy injections used the same generation parameters as the large-scale test events. The 100-event recovery benchmark is then run on the same M31 cadence that contains the toy subset used for tuning. This makes the reported true-positive rate a tuned, in-sample result rather than an out-of-sample prediction; the quantitative TP comparison with turboSETI alone is optimistically biased by this overlap.

full rationale

The paper is transparent about its machinery and its benchmark; the Choza et al. (2024) baseline is a published external pipeline result, and the 93.1% hit / 99.3% event false-positive reductions are real outputs of running HDBSCAN on the reduced hit set, not derivable from the seeds alone. The central circularity is confined to the true-positive validation: the paper's own Section 3.1 states that the 100 per-batch seed features are 'generated in the same way as' the 100 recovery events, and that non-synthetic hits in seeded clusters are retained 'as though they were outliers.' This makes the 69/100 event recovery and the 19.8% relative loss a measure of how well the pipeline keeps the seeded morphology, not of its intrinsic ability to preserve diverse technosignature-like signals. The additional M31 toy-tuning/test overlap (Section 4) further reduces the independence of the TP estimate. These are evaluation-design weaknesses that the authors disclose — the paper explicitly notes that 'The signal recovery rate is aided by our use of seed signals' — but they are load-bearing because the paper's headline trade-off (large FP reduction at modest TP cost) depends on the seeded TP number. Score 6: one quantitative prediction (TP event recovery) reduces by construction to template retention, while the FP-reduction claim retains independent content.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

All of the paper's numerical results are conditional on these choices. The most serious is the seed-signal injection, which directly shapes the true-positive retention measurement.

free parameters (7)
  • HDBSCAN minimum cluster size (npts) = 4
    Tuned on the toy M 31/HIP 3223 data set (Section 4); applied to all 639 batches in the benchmark. Cluster assignments depend strongly on this parameter.
  • HDBSCAN neighborhood points (rho_pts) = 2
    Set to 2 by the tuning procedure described in Section 4.
  • HDBSCAN merging threshold (epsilon_m) = 0.18
    Tuned between 0.15 and 0.25 on the toy set; affects how many subclusters are merged and hence the reduction rate.
  • Number of epochs = 8
    Stopping point chosen to reach the desired 93.1% reduction; more epochs would remove more hits at the cost of more false negatives (Section 3.1).
  • Seed signal set = 100 drifting narrowband synthetic signals per batch
    Injected into every batch to seed clusters with the target morphology; directly inflates retention of similarly shaped injected test events (Section 3.1).
  • t-SNE perplexity and early exaggeration = perplexity 40, early exaggeration 4
    Chosen by hand for the cross-batch centroid embedding (Section 3.2); different values change cluster matching.
  • Feature preprocessing choices = quantile transforms, log scales, normalization ranges, 2.7 kHz window, 200 Hz to 100 kHz grid, 100 kHz cap
    Defined in Section 2; the relative importance of features is set by these ranges, and the authors note ranges can be adjusted to change importance.
assumptions (5)
  • domain assumption HDBSCAN cluster labels meaningfully separate RFI morphologies in the 13-dimensional feature space.
    The entire method assumes common RFI types form dense clusters and unusual signals appear as outliers (Section 1.1, Section 3).
  • domain assumption The 13 handcrafted features are sufficient to characterize the structures that distinguish RFI from technosignature candidates.
    Feature list 'developed during preliminary analysis of a small portion of the data set' (Section 2); necessity is assessed only post hoc via random forest and SHAP on the toy set.
  • domain assumption Setigen-generated synthetic signals are a valid proxy for real technosignatures.
    Used as the ground truth for true-positive recovery (Section 1.2) and as seeds (Section 3.1).
  • domain assumption The turboSETI pipeline's hit/event definitions and the ABACAD spatial filter are accepted as the standard baseline.
    The benchmark inherits Choza et al. (2024)'s turboSETI search parameters (SNR cutoff 10, drift +/-4 Hz/s) and event definition without re-validating them (Section 1.2, Section 5).
  • domain assumption The distribution of RFI morphologies is similar across all 97 galaxies in the sample.
    One set of hyperparameters and one batch size is used for all cadences; no per-cadence adaptation or validation that cluster structure is stationary.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Anomaly Detection and Radio-frequency Interference Classification with Unsupervised Learning in Narrowband Radio Technosignature Searches." pith.science (2026). https://pith.science/paper/BOJWWPRZ

@misc{pith2026241116556,
  author       = {Pith},
  title        = {Pith review of: Anomaly Detection and Radio-frequency Interference Classification with Unsupervised Learning in Narrowband Radio Technosignature Searches},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BOJWWPRZ}},
  note         = {Machine review of arXiv:2411.16556}
}
read the original abstract

The search for radio technosignatures is an anomaly detection problem: Candidate signals represent needles of interest in the proverbial haystack of radio-frequency interference (RFI). Current search frameworks find an enormity of false-positive signals, especially in large surveys, requiring manual follow-up to a sometimes prohibitive degree. Unsupervised learning provides an algorithmic way to winnow the most anomalous signals from the chaff, as well as group together RFI signals that bear morphological similarities. We present GLOBULAR (Grouping Low-frequency Observations By Unsupervised Learning After Reduction) clustering, a signal processing method that uses HDBSCAN to reduce the false-positive rate and isolate outlier signals for further analysis. When combined with a standard narrowband signal detection and spatial filtering pipeline, such as turboSETI, GLOBULAR clustering offers significant improvements in the false-positive rate over the standard pipeline alone, suggesting dramatic potential for the amelioration of manual follow-up requirements for future large surveys. By removing RFI signals in regions of high spectral occupancy, GLOBULAR clustering may also enable the detection of signals missed by the standard pipeline. We benchmark our method against the Choza et al. turboSETI-only search of 97 nearby galaxies at the L band, demonstrating a false-positive hit reduction rate of 93.1% and a false-positive event reduction rate of 99.3%.

Figures

Figures reproduced from arXiv: 2411.16556 by the authors.

Figure 1
Figure 1. Distributions for the 13 features in our feature space, showing in light blue the distributions before any cluster-based rejection, in medium blue the distributions after 8 epochs of rejection, and in dark blue the retrieved distributions of the injected synthetic signals. Note the logarithmic scale on all y-axes. These distributions are from before the preprocessing steps outlined in Section 2, and therefore are no… view at source ↗
Figure 2
Figure 2. End-to-end overview of the GLOBULAR clustering method. Numbers and some labels represent our implementation on the Choza et al. (2024) data set and will vary by application. The raw data set also contains 296 synthetic hits corresponding to 98 events for true-positive recovery, which are not reflected in this chart. The three ⟲ indicate the iterations required to choose suitable hyperparameters for HDBSCAN and t-SNE… view at source ↗
Figure 3
Figure 3. Cluster counts for 639 batches, each with n ≈ 3,000. 80.9% of the batches are in the right-hand population and incur data reduction by discarding the non-anomalous classes. The batches in the left-hand population are insensi￾tively clustered and do not incur reduction before the next HDBSCAN iteration. The t-SNE algorithm offers one way to improve the matching. We key each hit to the centroid of the cluster it was f… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Top: 2-dimensional t-SNE embedding of a 10,000-hit sample. The axes are derived from a non-linear mapping and have no intuitive physical meaning. Each hit is passed through t-SNE not with its actual position in our feature space, but with the centroid of the cluster it…
Figure 5
Figure 5. Figure 5: Feature importances from a random forest trained on a 3,387-signal toy data set. Importances are calculated by normalizing the decrease in Gini impurity for each tree to unit sum, then averaging across all 128 trees. Error bars contain 68% of trees. 2001) offer a rough…
Figure 6
Figure 6. Figure 6: SHAP beeswarm plot generated from an HDB￾SCAN model’s cluster predictions of 400 hits. The model was trained on the same 3,387-hit toy data set as the ran￾dom forest in Section 4.1. Each point represents a single feature for a single hit, and its color corresponds to t…
Figure 8
Figure 8. Figure 8: Full-cadence dynamic spectra for the two false-positive events found in the reduced Choza et al. (2024) data set, with time from the beginning of the observation on the y-axis and frequency offset from the starting frequency of the event on the x-axis. Each block in a …
Figure 9
Figure 9. Figure 9: Examples of homogeneous clusters from batches before the cross-batch matching process, showing (top to bottom) low-drift narrowband signals consistent with aeronautical radionavigation, sparse comb structure consistent with GPS L3 signals near 1381 MHz, dense comb stru…
Figure 10
Figure 10. Figure 10: Snippets of dynamic spectra for 50 randomly selected hits from Cluster 0 in the t-SNE/HDBSCAN post-batching RFI analysis (see [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Snippets of dynamic spectra for 50 randomly selected hits from Cluster 17 in the t-SNE/HDBSCAN post-batching RFI analysis (see [PITH_FULL_IMAGE:figures/full_fig_p018_11.png]
Figure 12
Figure 12. Figure 12: Snippets of dynamic spectra for the 45 hits from Cluster 19 in the t-SNE/HDBSCAN post-batching RFI analysis (see [PITH_FULL_IMAGE:figures/full_fig_p018_12.png]
Figure 13
Figure 13. Figure 13: Snippets of dynamic spectra for 50 randomly selected hits from Cluster 21 in the t-SNE/HDBSCAN post-batching RFI analysis (see [PITH_FULL_IMAGE:figures/full_fig_p019_13.png]
Figure 14
Figure 14. Figure 14: Snippets of dynamic spectra for 50 randomly selected hits from Cluster 27 in the t-SNE/HDBSCAN post-batching RFI analysis (see [PITH_FULL_IMAGE:figures/full_fig_p019_14.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Machine Learning and the SKA for Cosmic Dawn and the Epoch of Reionization

    astro-ph.IM 2026-07 accept novelty 2.5 of 10

    A multi-author overview of machine-learning algorithms proposed for instrument modelling, data analysis, simulation and inference in SKA Cosmic Dawn and Epoch of Reionization science.

Reference graph

Works this paper leans on

73 extracted references · 15 canonical work pages · cited by 1 Pith paper

  1. [1]

    f(m0ntbE 3Wcdl *> ,^Jx /!i 3Vď'[ ̲VCIycF *U`

    thebibliography [1] 20pt to REFERENCES 6pt =0pt -12pt 10pt plus 3pt =0pt =0pt =1pt plus 1pt =0pt =0pt -12pt =13pt plus 1pt =20pt =13pt plus 1pt \@M =10000 =-1.0em =0pt =0pt 0pt =0pt =1.0em @enumiv\@empty 10000 10000 `\.\@m \@noitemerr \@latex@warning Empty `thebibliography' environment \@ifnextchar \@reference \@latexerr Missing key on reference command E...

  2. [2]

    2024, arXiv e-prints, arXiv:2410.01034, 10.48550/arXiv.2410.01034

    Andersson , A., Lintott , C., Fender , R., et al. 2024, arXiv e-prints, arXiv:2410.01034, 10.48550/arXiv.2410.01034

  3. [3]

    P., Tollerud , E

    Astropy Collaboration , Robitaille , T. P., Tollerud , E. J., et al. 2013, , 558, A33, 10.1051/0004-6361/201322068

  4. [4]

    M., Sip o cz , B

    Astropy Collaboration , Price-Whelan , A. M., Sip o cz , B. M., et al. 2018, , 156, 123, 10.3847/1538-3881/aabc4f

  5. [5]

    M., Lim , P

    Astropy Collaboration , Price-Whelan , A. M., Lim , P. L., et al. 2022, , 935, 167, 10.3847/1538-4357/ac7c74

  6. [6]

    Biswas , B., Ishida , E. E. O., Peloton , J., et al. 2023, , 677, A77, 10.1051/0004-6361/202245340

  7. [7]

    2001, Machine Learning, 45, 5, 10.1023/A:1010933404324

    Breiman , L. 2001, Machine Learning, 45, 5, 10.1023/A:1010933404324

  8. [8]

    Brzycki, B., Siemion, A. P. V., de Pater, I., et al. 2023, The Astrophysical Journal, 952, 46, 10.3847/1538-4357/acdee0

Show all 73 references
  1. [9]

    2022, The Astronomical Journal, 163, 222, 10.3847/1538-3881/ac5e3d

    ---. 2022, The Astronomical Journal, 163, 222, 10.3847/1538-3881/ac5e3d

  2. [10]

    Brzycki , B., Siemion , A. P. V., de Pater , I., et al. 2024, , 168, 284, 10.3847/1538-3881/ad7e18

  3. [11]

    Campello, R. J. G. B., Moulavi, D., & Sander, J. 2013, in Advances in Knowledge Discovery and Data Mining, ed. J. Pei, V. S. Tseng, L. Cao, H. Motoda, & G. Xu (Berlin, Heidelberg: Springer Berlin Heidelberg), 160--172

  4. [12]

    2024, , 167, 10, 10.3847/1538-3881/acf576

    Choza , C., Bautista , D., Croft , S., et al. 2024, , 167, 10, 10.3847/1538-3881/acf576

  5. [13]

    1959, Nature, 184, 844, 10.1038/184844a0

    Cocconi , G., & Morrison , P. 1959, Nature, 184, 844, 10.1038/184844a0

  6. [14]

    2013, Python and HDF5 (O'Reilly)

    Collette, A. 2013, Python and HDF5 (O'Reilly)

  7. [15]

    A., Syed, T

    dos Santos, J. A., Syed, T. I., Naldi, M. C., Campello, R. J. G. B., & Sander, J. 2019, IEEE Transactions on Big Data, PP, 1, 10.1109/TBDATA.2019.2907624

  8. [16]

    1961, Physics Today, 14, 40, https://doi.org/10.1063/1.3057500

    Drake, F. 1961, Physics Today, 14, 40, https://doi.org/10.1063/1.3057500

  9. [17]

    2019, turboSETI: Python-based SETI search algorithm

    Enriquez , E., & Price , D. 2019, turboSETI: Python-based SETI search algorithm . 1906.006

  10. [18]

    E., Siemion , A., Foster , G., et al

    Enriquez , J. E., Siemion , A., Foster , G., et al. 2017, , 849, 104, 10.3847/1538-4357/aa8d1b

  11. [19]

    1996, in Proceedings of the Second International Conference on Knowledge Discovery and Data Mining, KDD'96 (AAAI Press), 226–231

    Ester, M., Kriegel, H.-P., Sander, J., & Xu, X. 1996, in Proceedings of the Second International Conference on Knowledge Discovery and Data Mining, KDD'96 (AAAI Press), 226–231

  12. [20]

    2024, , 529, 732, 10.1093/mnras/stae496

    Etsebeth , V., Lochner , M., Walmsley , M., & Grespan , M. 2024, , 529, 732, 10.1093/mnras/stae496

  13. [21]

    Franz, N., Croft, S., Siemion, A. P. V., et al. 2022, The Astronomical Journal, 163, 104, 10.3847/1538-3881/ac46c9

  14. [22]

    I., Siemion , A

    Gajjar , V., Perez , K. I., Siemion , A. P. V., et al. 2021, , 162, 33, 10.3847/1538-3881/abfd36

  15. [23]

    2022, , 932, 81, 10.3847/1538-4357/ac6dd5

    Gajjar , V., LeDuc , D., Chen , J., et al. 2022, , 932, 81, 10.3847/1538-4357/ac6dd5

  16. [24]

    C., Fabbro , S., et al

    Gilda , S., Draper , S. C., Fabbro , S., et al. 2022, , 510, 870, 10.1093/mnras/stab3243

  17. [25]

    R., Millman, K

    Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357, 10.1038/s41586-020-2649-2

  18. [26]

    A., & Hartigan, P

    Hartigan, J. A., & Hartigan, P. M. 1985, The Annals of Statistics, 13, 70 , 10.1214/aos/1176346577

  19. [27]

    Hunter, J. D. 2007, Computing in Science & Engineering, 9, 90, 10.1109/MCSE.2007.55

  20. [28]

    Isaacson, H., Siemion, A. P. V., Marcy, G. W., et al. 2017, Publications of the Astronomical Society of the Pacific, 129, 1. https://www.jstor.org/stable/26660103

  21. [29]

    N., & Gill, C

    Joanes, D. N., & Gill, C. A. 1998, Journal of the Royal Statistical Society. Series D (The Statistician), 47, 183. http://www.jstor.org/stable/2988433

  22. [30]

    2007, Journal of Modern Applied Statistical Methods, 6, 8, 10.22237/jmasm/1177992120

    Knapp, T. 2007, Journal of Modern Applied Statistical Methods, 6, 8, 10.22237/jmasm/1177992120

  23. [31]

    Lebofsky , M., Croft , S., Siemion , A. P. V., et al. 2019, , 131, 124505, 10.1088/1538-3873/ab3e82

  24. [32]

    S., & Ip , W.-H

    Lin , C.-L., Apai , D., Giampapa , M. S., & Ip , W.-H. 2024, , 168, 234, 10.3847/1538-3881/ad7956

  25. [33]

    T., Ting, K

    Liu, F. T., Ting, K. M., & Zhou, Z.-H. 2008, in 2008 Eighth IEEE International Conference on Data Mining, 413--422, 10.1109/ICDM.2008.17

  26. [34]

    Lochner , M., & Bassett , B. A. 2021, Astronomy and Computing, 36, 100481, 10.1016/j.ascom.2021.100481

  27. [35]

    Lomb , N. R. 1976, , 39, 447, 10.1007/BF00648343

  28. [36]

    2017, arXiv e-prints, arXiv:1705.07874, 10.48550/arXiv.1705.07874

    Lundberg , S., & Lee , S.-I. 2017, arXiv e-prints, arXiv:1705.07874, 10.48550/arXiv.1705.07874

  29. [37]

    X., Ng, C., Rizk, L., et al

    Ma, P. X., Ng, C., Rizk, L., et al. 2023, Nature Astronomy, 10.1038/s41550-022-01872-z

  30. [38]

    2019, arXiv e-prints, arXiv:1911.02282, 10.48550/arXiv.1911.02282

    Malzer , C., & Baum , M. 2019, arXiv e-prints, arXiv:1911.02282, 10.48550/arXiv.1911.02282

  31. [39]

    2021, Statistical Analysis and Data Mining: An ASA Data Science Journal, 14, 144, https://doi.org/10.1002/sam.11498

    Mantero, A., & Ishwaran, H. 2021, Statistical Analysis and Data Mining: An ASA Data Science Journal, 14, 144, https://doi.org/10.1002/sam.11498

  32. [40]

    2021, The Astronomical Journal, 161, 55, 10.3847/1538-3881/abcc77

    Margot, J.-L., Pinchuk, P., Geil, R., et al. 2021, The Astronomical Journal, 161, 55, 10.3847/1538-3881/abcc77

  33. [41]

    T., Matcheva , K., & Roman , A

    Matchev , K. T., Matcheva , K., & Roman , A. 2022, , 3, 205, 10.3847/PSJ/ac880b

  34. [42]

    2017, in 2017 IEEE International Conference on Data Mining Workshops (ICDMW) (IEEE), 10.1109/icdmw.2017.12

    McInnes, L., & Healy, J. 2017, in 2017 IEEE International Conference on Data Mining Workshops (ICDMW) (IEEE), 10.1109/icdmw.2017.12

  35. [43]

    2017, The Journal of Open Source Software, 2, 10.21105/joss.00205

    McInnes, L., Healy, J., & Astels, S. 2017, The Journal of Open Source Software, 2, 10.21105/joss.00205

  36. [44]

    2010, in P roceedings of the 9th P ython in S cience C onference, ed

    M c K inney, W. 2010, in P roceedings of the 9th P ython in S cience C onference, ed. S t\'efan van der W alt & J arrod M illman, 56 -- 61, 10.25080/Majora-92bf1922-00a

  37. [45]

    J., Iacobelli , M., et al

    Mesarcik , M., Boonstra , A. J., Iacobelli , M., et al. 2023, , 680, A74, 10.1051/0004-6361/202347182

  38. [46]

    Mesarcik , M., Boonstra , A.-J., Ranguelova , E., & van Nieuwpoort , R. V. 2022, , 516, 5367, 10.1093/mnras/stac2503

  39. [47]

    Neto, A. C. A., Naldi, M. C., Campello, R. J. G. B., & Sander, J. 2022, in 2022 IEEE 38th International Conference on Data Engineering (ICDE), 951--964, 10.1109/ICDE53745.2022.00076

  40. [48]

    Neto, A. C. A., Nascimento, M. A., Sander, J., & Campello, R. J. G. B. 2018, Proc. VLDB Endow., 11, 2058–2061, 10.14778/3229863.3236259

  41. [49]

    R., de Bruyn , A

    Offringa , A. R., de Bruyn , A. G., Biehl , M., et al. 2010, , 405, 155, 10.1111/j.1365-2966.2010.16471.x

  42. [50]

    R., van de Gronde , J

    Offringa , A. R., van de Gronde , J. J., & Roerdink , J. B. T. M. 2012, , 539, A95, 10.1051/0004-6361/201118497

  43. [51]

    2024, , in prep.\

    Painter , C., et al. 2024, , in prep.\

  44. [52]

    2011, Journal of Machine Learning Research, 12, 2825

    Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, Journal of Machine Learning Research, 12, 2825

  45. [53]

    A., Janczyk, M., Dale, R., & Freeman, J

    Pfister, R., Schwarz, K. A., Janczyk, M., Dale, R., & Freeman, J. B. 2013, Frontiers in Psychology, 4, 1, 10.3389/fpsyg.2013.00700

  46. [54]

    2022, , 163, 76, 10.3847/1538-3881/ac426f

    Pinchuk , P., & Margot , J.-L. 2022, , 163, 76, 10.3847/1538-3881/ac426f

  47. [55]

    C., Enriquez, J

    Price, D. C., Enriquez, J. E., Chen, Y., & Siebert, M. 2019, Journal of Open Source Software, 4, 1554, 10.21105/joss.01554

  48. [56]

    C., Enriquez, J

    Price, D. C., Enriquez, J. E., Brzycki, B., et al. 2020, The Astronomical Journal, 159, 86, 10.3847/1538-3881/ab65f1

  49. [57]

    A., Forsyth , C., Aruliah , A., et al

    Reddy , S. A., Forsyth , C., Aruliah , A., et al. 2023, Journal of Geophysical Research (Space Physics), 128, e2022JA031183, 10.1029/2022JA031183

  50. [58]

    Scargle , J. D. 1982, , 263, 835, 10.1086/160554

  51. [59]

    Shapley, L. S. 1951, Notes on the N-Person Game --- II: The Value of an N-Person Game (Santa Monica, CA: RAND Corporation), 10.7249/RM0670

  52. [60]

    Siemion, A. P. V., Demorest, P., Korpela, E., et al. 2013, The Astrophysical Journal, 767, 94

  53. [61]

    A., et al

    Sooknunan , K., Lochner , M., Bassett , B. A., et al. 2021, , 502, 206, 10.1093/mnras/staa3873

  54. [62]

    K., et al

    Suazo , M., Zackrisson , E., Mahto , P. K., et al. 2024, , 531, 695, 10.1093/mnras/stae1186

  55. [63]

    2023, , 165, 255, 10.3847/1538-3881/acccf0

    Suresh , A., Gajjar , V., Nagarajan , P., et al. 2023, , 165, 255, 10.3847/1538-3881/acccf0

  56. [64]

    2022, Mathematics, 10, 10.3390/math10071042

    Tarb a , N., Voncil a , M.-L., & Boiangiu, C.-A. 2022, Mathematics, 10, 10.3390/math10071042

  57. [65]

    2021, , 161, 286, 10.3847/1538-3881/abf649

    Traas , R., Croft , S., Gajjar , V., et al. 2021, , 161, 286, 10.3847/1538-3881/abf649

  58. [66]

    J., Ivezi\'c, Z ., & Gray, A

    VanderPlas, J., Connolly, A. J., Ivezi\'c, Z ., & Gray, A. 2012, in 2012 Conference on Intelligent Data Understanding (IEEE), 47–54, 10.1109/cidu.2012.6382200

  59. [67]

    T., & Ivezi\'c, Z

    VanderPlas, J. T., & Ivezi\'c, Z . 2015, The Astrophysical Journal, 812, 18, 10.1088/0004-637x/812/1/18

  60. [68]

    E., et al

    Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261, 10.1038/s41592-019-0686-2

  61. [69]

    2016, Distill, 10.23915/distill.00002

    Wattenberg, M., Vi \' e gas, F., & Johnson, I. 2016, Distill, 10.23915/distill.00002

  62. [70]

    2025, in prep.\

    West , N., et al. 2025, in prep.\

  63. [71]

    P., Drew, J., Siemion, A., et al

    Worden, S. P., Drew, J., Siemion, A., et al. 2017, Acta Astronautica, 139, 98, 10.1016/j.actaastro.2017.06.008

  64. [72]

    T., Griffith , R

    Wright , J. T., Griffith , R. L., Sigurdsson , S., Povich , M. S., & Mullan , B. 2014, , 792, 27, 10.1088/0004-637X/792/1/27

  65. [73]

    2023, , 165, 114, 10.3847/1538-3881/acb342

    Zuckerman , A., Ko , Z., Isaacson , H., et al. 2023, , 165, 114, 10.3847/1538-3881/acb342

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.