Understanding molecular ratios in the carbon and oxygen poor outer Milky Way with interpretable machine learning

Francesco Fontani; Gijs Vermari\"en; Johannes Heyl; Serena Viti

REVIEW 4 major objections 4 minor 1 cited by

Understanding molecular ratios in the carbon and oxygen poor outer Milky Way with interpretable machine learning

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The CN/HCN and HNC/HCN molecular ratios respond mainly to the initial carbon abundance, making them practical probes of metallicity in the carbon- and oxygen-poor outer Milky Way.

desk verdict A clean, useful parameter survey showing CN/HCN and HNC/HCN respond to carbon abundance in low-metallicity clouds, but the 'excellent probes' claim outruns what the unreported surrogate fit metrics can support. read the letter →

arxiv 2505.08410 v1 pith:PMMU2GQ6 submitted 2025-05-13 astro-ph.GA cs.LG

Gijs Vermariën , Serena Viti , Johannes Heyl , Francesco Fontani This is my paper · ORCID

classification astro-ph.GAcs.LG

keywords astrochemicalmodelingmolecularlineratiosouterMilkyWaylowmetallicityinitialcarbonabundanceSHAPUMAPgas-grainchemistry

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

This paper asks which physical conditions can be read off from molecular line ratios in the carbon- and oxygen-poor outer Milky Way. Using roughly 65,000 gas-grain models with varying density, temperature, cosmic-ray ionisation, UV field, and initial carbon and oxygen abundances, it trains a boosted regression forest for each of nine ratios and explains the forest with SHAP values. The paper claims that temperature and density dominate most ratios, but that CN/HCN and HNC/HCN respond strongly to the initial carbon abundance, making them useful observational probes of that parameter. If true, measuring these two ratios in outer-Galaxy clouds beyond about 16 kpc could test whether the extrapolated radial metallicity gradient is correct. The paper also introduces UMAP embeddings of the SHAP vectors as a way to see higher-order parameter dependencies that classical two-dimensional plots miss.

What carries the argument

The load-bearing mechanism is the pairing of a large gas-grain chemistry grid with an interpretable surrogate. UCLCHEM supplies time-dependent abundances for each parameter set sampled with Sobol sequences; for each of the nine ratios an XGBoost regression forest is trained on the physical parameters, and TreeSHAP decomposes every prediction into six additive per-feature contributions that sum to the predicted ratio. Those six-dimensional SHAP vectors, with the ratio itself attached, are then embedded with UMAP into two dimensions so that regions of parameter space with similar chemistry appear as clustered regions on a manifold. The SHAP decomposition is what lets the paper claim that carbon abundance, not just temperature or density, drives CN/HCN and HNC/HCN; UMAP is what lets it see where in parameter space that carbon sensitivity lives.

What would settle it

A concrete check: report the held-out test error or $R^2$ of each of the nine regression forests and recompute the SHAP rankings after resampling or densifying the high-density, low-temperature part of the grid; the paper's carbon-abundance ranking for CN/HCN and HNC/HCN should survive. Observationally, measure CN/HCN and HNC/HCN in outer-Galaxy clouds beyond $16$ kpc where the metallicity gradient is only extrapolated: the models predict these ratios should shift with the initial carbon abundance in a specific direction.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is a ranking: among the nine ratios, CN/HCN and HNC/HCN are the two whose behaviour across the model grid is most controlled by the initial carbon abundance after density, with carbon ahead of temperature as the second most important feature. Only CS/SO among the nine shows measurable sensitivity to the initial oxygen abundance. The models deplete carbon and oxygen independently down to one-twentieth of their solar values, in cold clouds of $10^3$ to $10^7$ cm$^{-3}$ and $10$ to $100$ K, take ratios at $10^5$ years, and discard any ratio whose molecule falls below an abundance of $10^{-12}$. From that grid the paper argues that temperature and density dominate most ratios, that the cosmic-ray ionisation rate cannot be constrained by these ratios in the studied range, and that the carbon-sensitive pair can therefore serve as an observational probe of the carbon abundance, and with it the metallicity gradient, in the outer Galaxy.

Load-bearing premise

The entire sensitivity ranking depends on the boosted forest accurately reproducing the chemistry grid in every region of parameter space, including sparse high-density, low-temperature regions where some UCLCHEM models did not converge; if the surrogate is wrong there, the SHAP attributions could assign ratio changes to the wrong physical parameter.

Editorial extensions

If this is right

If the carbon-abundance ranking is right, CN/HCN and HNC/HCN become practical metallicity probes: observers can test the extrapolated radial metallicity gradient beyond $16$ kpc by measuring these two ratios in outer-Galaxy clouds.
Because temperature and density dominate most of the nine ratios, observational use of the carbon-sensitive pair must first constrain the local temperature and density, or the abundance signature will be masked.
The cosmic-ray ionisation rate cannot be pinned down by these nine ratios over the studied range of $\zeta = 10^{-17}$ to $10^{-14}$ s$^{-1}$, so constraining it will require other tracers or a grid with coupled thermal balance.
The SHAP-plus-UMAP workflow identifies which parameter subspace each ratio responds to, which can be used to set better priors in Bayesian backwards modelling of observed cores.

Reading between the lines

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

Beyond the paper: because CN/HCN and HNC/HCN are routinely observed in starbursts and active galaxies, the carbon-abundance sensitivity found here suggests those extragalactic systems could in principle be probed for their carbon-to-oxygen abundance with the same two ratios, not just outer-Galaxy clouds.
Beyond the paper: the low importance assigned to the cosmic-ray ionisation rate may be partly inherited from the model setup, in which the temperature is fixed and therefore decoupled from cosmic-ray heating; a coupled thermal-chemical grid could change the SHAP ranking.
Beyond the paper: a direct observational test would measure CN/HCN and HNC/HCN in clouds with independently known metallicities beyond $16$ kpc; the paper's models predict a systematic shift with the initial carbon abundance that an abundance-blind calibration could falsify.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Here's my take on arXiv:2505.08410. It is a well-executed forward-model study: a 65k-model UCLCHEM grid over a relevant low-metallicity outer-Galaxy parameter space, interpreted with the authors' established XGBoost+SHAP+UMAP pipeline. The physical result—CN/HCN and HNC/HCN are sensitive to the initial carbon abundance, temperature, and density—is plausible and consistent with older literature (e.g., Milam et al. 2005). What the paper does well: the grid design is thoughtful (Sobol sampling, wide ranges, independent C and O depletion), the classical temperature-density maps (Fig. 3) are genuinely informative, and the paper is transparent about what it cannot constrain (cosmic rays, because T is fixed). It also honestly notes the non-convergent high-density/low-temperature region. As a mapping of ratio behavior in a previously under-explored parameter space, this is a useful contribution.

The soft spots are real but not fatal. The main one: the SHAP importances that drive every sensitivity ranking come from XGBoost surrogates, and the paper never reports the held-out accuracy of those surrogates. The text says test error was the Optuna target, but no R², RMSE, or MAE appears in Table B.1 or anywhere else. Given that the same test split was used for hyperparameter selection, there is no clean estimate of generalization, and the problem is most acute in sparse regions—precisely where the "either molecule detected" filter produces extreme ratios (one species orders of magnitude below the 1e-12 threshold). If the surrogate is inaccurate there, the SHAP rankings could be artifacts. I don't think they are—the rankings line up with physical expectations—but the claim "excellent probes" needs that metric to be credible.

Second, the "either molecule" filter is a deliberate choice, but the headline claim should be restricted to the both-detected subset (or at least checked there). Many of the extreme ratios in Table 2 are not directly observable. Third, the abstract's "excellent probes" is a forward-model statement; there is no test against observations (e.g., CHEMOUT sources). That's acceptable for a model characterization paper, but the language should not imply empirical validation.

Overall: this is a solid, incremental paper for astrochemistry readers who want a ratio sensitivity map for low-metallicity clouds. It does not need a desk rejection. It needs a referee who asks for (1) surrogate fit metrics on a fresh holdout, (2) sensitivity of the SHAP ranking to the detection filter, and (3) a softened or qualified headline claim. With those changes, it would be publishable. I'd send it to review, and I'd cite it in my own work if I were mapping out carbon probes in the outer Galaxy.

Referee Report

4 major / 4 minor

Summary. The paper presents a modeling study of nine molecular abundance ratios in dark-cloud conditions representative of the carbon- and oxygen-poor outer Milky Way. A six-dimensional parameter grid (n_H, T, ζ, F_UV, f_C, f_O) is sampled with 65,536 Sobol points, run with UCLCHEM to 10^7 yr, and ratio values are evaluated at 10^5 yr after applying a 10^-12 abundance detection threshold. For each ratio, an XGBoost regression forest is trained on the surviving grid points and TreeSHAP is used to rank the physical parameters by importance; SHAP vectors and ratio values are then embedded with UMAP to identify regimes of similar chemical behavior. The authors conclude that temperature and density are generally the most important parameters, that CN/HCN and HNC/HCN are sensitive to the initial carbon abundance and are therefore 'excellent probes' for it, that CS/SO is the only ratio with oxygen sensitivity, and that the cosmic-ray ionization rate cannot be constrained by these ratios.

Significance. If the carbon-sensitivity claim holds, the paper offers a practical observational route to test extrapolated metallicity gradients beyond ~16 kpc with CHEMOUT-type observations, which would be a valuable contribution. The pipeline is transparent in several respects: the UCLCHEM grid, Sobol sampling, detection threshold, Optuna hyperparameter search, and the TreeSHAP/UMAP choices are all documented, and the inclusion of nine ratios spanning different chemical families is a useful feature. The main limitation is that all sensitivity statements pass through nine XGBoost surrogates whose held-out accuracy is never reported; until that is fixed, the 'excellent probes' conclusion is not yet supported. With the requested fit metrics and a few clarifying corrections, the paper would be a solid methods-focused contribution to astrochemical interpretation.

major comments (4)

[§2.5, Table B.1] The central conclusion that CN/HCN and HNC/HCN are 'excellent probes' for the initial carbon abundance is derived from TreeSHAP values of nine XGBoost surrogates, yet the manuscript never reports any generalization metric (e.g., R², RMSE, or MAE) for these surrogates. Table B.1 lists only the Optuna-selected hyperparameters, and Section 2.5 states that the test error was used as the optimisation target, so the same 30% split is not an independent test set for the final retrained models. Because TreeSHAP explanations inherit the errors of the fitted model, the sensitivity rankings in Figure 4 and the abstract's conclusion are unverified until held-out fit metrics are provided. Please report per-ratio fit statistics and, ideally, the fit quality in the sparse high-density/low-temperature and single-species-detection regimes.
[§3.1, Figure 2] The dataset used for SHAP includes samples in which only one of the two molecules is above the 10^-12 threshold. As shown by the minima in Table 2 (e.g., log10(HCO+/HCN) = -23.14), many ratios are then extreme values that do not correspond to directly observable line ratios. These unobservable points can dominate the training distribution and may drive the SHAP attribution of f_C for CN/HCN and HNC/HCN. The paper should quantify how much of the carbon sensitivity is present in the both-detected subset alone and should report surrogate accuracy on that subset; otherwise the 'excellent probe' claim is not established for the regime in which the ratios can actually be observed.
[Abstract vs §4 item 7 and Figure 4] The abstract states that 'only CS/SO shows a sensitivity to the oxygen abundance,' but Figure 4 gives normalized f_O importances of 0.15 for HNC/HCN and CS/CN and 0.14 for CS/SO, and Section 4 item 7 states that 'CS/CN' has the largest oxygen dependence. These statements are mutually inconsistent. Please correct the summary and clarify whether the criterion is importance magnitude, monotonic direction, or something else, and reconcile the abstract with the quantitative importances.
[§2.1, Table A.1] The models vary only the initial carbon and oxygen abundances while keeping nitrogen, sulfur, and all other heavy elements at their solar values (Table A.1). The paper frames the study as representative of low-metallicity outer-Galaxy gas; if the real low-metallicity gas is depleted in N and S as well, the CN/HCN and CS/SO chemistries could behave differently. Please justify this assumption or test the sensitivity of the main conclusions to scaling the other heavy-element abundances down with metallicity.

minor comments (4)

[Table 1] The symbols and names in the last two rows appear interchanged: the row labeled 'Initial abundance of carbon' uses f_O/f_O,⊙, and the row labeled 'Initial abundance of oxygen' uses f_C/f_C,⊙. Please correct the symbols so that f_C is associated with the carbon abundance and f_O with oxygen.
[§2.4] The text says 'our dataset is sampled on a regular grid,' but Section 2.1 describes Sobol sequence sampling, which is a low-discrepancy quasi-random sampling instead of a regular grid; the wording should be changed to avoid this inconsistency.
[§2.1, throughout] The manuscript repeatedly refers to 'molecular line ratios,' but only abundance ratios are computed from UCLCHEM; no radiative transfer or excitation calculation is performed. The assumed proportionality between abundance ratios and observable line ratios should be stated explicitly as a caveat.
[§2.4] The UMAP hyperparameters (k, d_min, w_ratio) were chosen by manual tuning with the stated goal of obtaining a 'smooth manifold'; a brief robustness check, such as varying k and d_min and describing how the grouping changes, would strengthen the interpretation of the UMAP-based claims.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's sensitivity rankings are derived from UCLCHEM forward models through a surrogate SHAP explanation, and prior self-citations are corroborative rather than load-bearing.

full rationale

The central claim that CN/HCN and HNC/HCN are sensitive to the initial carbon abundance is a forward-model statement about the UCLCHEM grid, in which f_C is varied from 0.05 to 1.0 times solar and the molecular ratios are simulated outputs. The XGBoost surrogate is a fitted interpolation of that grid, and TreeSHAP decomposes the surrogate's predictions; this is a standard machine-learning explanation pipeline, not a case where a parameter is fitted to a subset and then the same subset is 'predicted' as independent confirmation. No equation in the paper defines the carbon abundance in terms of the ratios, and the paper performs no inverse retrieval of f_C from the ratios, so the 'excellent probes' conclusion is not forced by construction. The self-citations (Fontani et al. 2024, Heyl et al. 2023a,b, and the UCLCHEM code paper) are used for contextual abundance ranges, methodological precedent, or corroboration of a secondary result about cosmic-ray ionisation sensitivity; none is the sole justification for the carbon-sensitivity ranking, and the paper even notes a discrepancy with Heyl et al. 2023a on temperature importance, showing the comparison is not an imported conclusion. The absence of reported held-out R², RMSE, or MAE for the nine surrogates is a genuine validation gap: if the boosted forests predict poorly in sparse regions such as non-convergent high-density/low-temperature cells or 'either molecule detected' regimes, SHAP importance rankings could inherit that error. However, surrogate fidelity is a robustness concern about whether the explanation reflects the underlying grid, not a circularity in which the output is equivalent to the input by definition. UMAP is used only as a dimensionality-reduction and visualization step on the SHAP vectors and ratios and cannot by itself create a circular dependency. Overall, the derivation chain is self-contained against the UCLCHEM forward models, and no load-bearing step reduces to its own inputs.

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

The paper introduces no new physical entities or fitted physical constants. Its load-bearing choices are the UCLCHEM model itself, the grid ranges, the single evaluation time, the 1e-12 threshold, and the ML surrogate whose SHAP values are interpreted as physical sensitivities. These choices are documented but not independently benchmarked.

free parameters (5)

Grid ranges for n_H, T, zeta, F_UV, f_C, f_O = n_H 1e3 to 1e7 cm-3; T 10 to 100 K; zeta 1e-17 to 1e-14 s-1; F_UV 0.1 to 100 Habing; f_C 0.05 to 1.0 times 1.77e-4…
The manually chosen sampling ranges define the parameter space over which sensitivities are ranked; conclusions could change outside these ranges.
Ratio evaluation time = 1e5 yr
Choosing t=1e5 yr avoids full freeze-out but is a significant modeling choice; ratios at other times could give different sensitivities.
Observational abundance threshold = x_i >= 1e-12
Adopted as the conservatively lowest observable abundance; filtering on 'either molecule detected' later includes unobservable extreme ratios.
UMAP hyperparameters = k=100, d_min=0.1 or 0.5, w_ratio=0.1
Manually tuned to produce smooth clustered manifolds; embeddings are stochastic and no seed is reported.
XGBoost hyperparameters = Table B.1, e.g., max_depth 7 to 13, n_estimators 142 to 981
Optimized via Optuna against test error on the synthetic grid; SHAP values depend on this fitted surrogate, yet final fit quality metrics are not reported.

assumptions (5)

domain assumption UCLCHEM's gas-grain reaction network accurately describes the chemistry of low-metallicity dark clouds in the outer Milky Way.
All ratio values are generated by UCLCHEM; the conclusions about real probes inherit this assumption (Section 2.1).
domain assumption Clouds are isothermal, constant-density spheres of radius 0.5 pc with no dynamics or internal radiation sources.
Stated in Section 2.1; the modeled parameter space is a simplification of real outer Galaxy clouds.
domain assumption Lower metallicity is captured solely by scaling initial carbon and oxygen abundances down by factors up to 20 while keeping all other elemental abundances fixed.
Section 2.1 and Appendix A; real metallicity gradients also affect nitrogen, sulfur, and dust properties.
ad hoc to paper SHAP values from the XGBoost surrogate faithfully represent the sensitivity of the UCLCHEM ratio surface.
The paper optimizes the surrogate on test error but never reports fit quality (Section 2.5), so local faithfulness is assumed.
ad hoc to paper Including samples where only one molecule is above the detection threshold still yields meaningful ratio distributions for SHAP analysis.
Section 3.1; these samples use very low model abundances for the undetected partner, creating extreme ratios not directly observable.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Understanding molecular ratios in the carbon and oxygen poor outer Milky Way with interpretable machine learning." pith.science (2026). https://pith.science/paper/PMMU2GQ6

@misc{pith2026250508410,
  author       = {Pith},
  title        = {Pith review of: Understanding molecular ratios in the carbon and oxygen poor outer Milky Way with interpretable machine learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PMMU2GQ6}},
  note         = {Machine review of arXiv:2505.08410}
}

read the original abstract

Context. The outer Milky Way has a lower metallicity than our solar neighbourhood, but still many molecules are detected in the region. Molecular line ratios can serve as probes to better understand the chemistry and physics in these regions. Aims. We use interpretable machine learning to study 9 different molecular ratios, helping us understand the forward connection between the physics of these environments and the carbon and oxygen chemistries. Methods. Using a large grid of astrochemical models generated using UCLCHEM, we study the properties of molecular clouds of low oxygen and carbon initial abundance. We first try to understand the line ratios using a classical analysis. We then move on to using interpretable machine learning, namely Shapley Additive Explanations (SHAP), to understand the higher order dependencies of the ratios over the entire parameter grid. Lastly we use the Uniform Manifold Approximation and Projection technique (UMAP) as a reduction method to create intuitive groupings of models. Results. We find that the parameter space is well covered by the line ratios, allowing us to investigate all input parameters. SHAP analysis shows that the temperature and density are the most important features, but the carbon and oxygen abundances are important in parts of the parameter space. Lastly, we find that we can group different types of ratios using UMAP. Conclusions. We show the chosen ratios are mostly sensitive to changes in the carbon initial abundance, together with the temperature and density. Especially the CN/HCN and HNC/HCN ratio are shown to be sensitive to the initial carbon abundance, making them excellent probes for this parameter. Out of the ratios, only CS/SO shows a sensitivity to the oxygen abundance.

Figures

Figures reproduced from arXiv: 2505.08410 by the authors.

**Figure 1.** Plots showing fractional abundances for each of the ratios discussed in this paper. Contour levels of a kernel [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗

**Figure 2.** The distribution of each of the ratios that exceed the [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

Figures from the paper (4 more)

**Figure 3.** Figure 3: The distribution of the log of the ratios as a function of both density and temperature. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png]

**Figure 4.** Figure 4: The relative importances of each physical parameter for all of the ratios. [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]

**Figure 5.** Figure 5: All of the SHAP values for each of the ratios. The features are sorted by mean absolute impact as shown in [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

**Figure 6.** Figure 6: The H2CO/CH3OH ratio, feature and SHAP values plotted on 2-dimensional manifold using the ratio and the SHAP values. The manifold consists of a broad left region with two separate right lobes. 4. Discussion and conclusions Overall, the results of this study point to th…

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

End-to-end differentiable retrieval of molecular spectra using hydrodynamics, chemistry, and radiative transfer
astro-ph.IM 2026-07 conditional novelty 6.0 of 10

An end-to-end differentiable JAX pipeline couples 1D hydrodynamics, time-dependent chemistry, and radiative transfer, and recovers shock and rate parameters from synthetic HCO+ spectra.

Reference graph

Works this paper leans on

69 extracted references · 58 canonical work pages · cited by 1 Pith paper

[1]

2019, Optuna: A Next-generation Hyperparameter Optimization Framework

Akiba, T., Sano, S., Yanase, T., Ohta, T., & Koyama, M. 2019, Optuna: A Next-generation Hyperparameter Optimization Framework

work page 2019
[2]

& Faure, A

Bacmann, A. & Faure, A. 2016, A&A, 587, A130

work page 2016
[3]

W., Williams, D

Bayet, E., Hartquist, T. W., Williams, D. A., et al. 2011, Memorie della Societa Astronomica Italiana, 82, 893

work page 2011
[4]

A., & Rawlings, J

Bayet, E., Viti, S., Williams, D. A., & Rawlings, J. M. C. 2008, ApJ, 676, 978

work page 2008
[5]

A., Rawlings, J

Bayet, E., Viti, S., Williams, D. A., Rawlings, J. M. C., & Bell, T. 2009, ApJ, 696, 1466

work page 2009
[6]

G., Holdship, J., et al

Behrens, E., Mangum, J. G., Holdship, J., et al. 2022, ApJ, 939, 119

2022
[7]

J., Sephus, C

Bernal, J. J., Sephus, C. D., & Ziurys, L. M. 2021, ApJ, 922, 106

work page 2021
[8]

D., Charnley, S

Brown, P. D., Charnley, S. B., & Millar, T. J. 1988, MNRAS, 231, 409

work page 1988

Show all 69 references

[9]

2022, A&A, 667, A131

Butterworth, J., Holdship, J., Viti, S., & Garc \'i a-Burillo , S. 2022, A&A, 667, A131

2022
[10]

H., Hashimoto, T., Goto, T., et al

Chen, B. H., Hashimoto, T., Goto, T., et al. 2022, MNRAS, 509, 1227

2022
[11]

2019, Explaining Models by Propagating Shapley Values of Local Components

Chen, H., Lundberg, S., & Lee, S.-I. 2019, Explaining Models by Propagating Shapley Values of Local Components

2019
[12]

& Guestrin, C

Chen, T. & Guestrin, C. 2016, in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , 785--794

2016
[13]

& Bachiller, R

Codella, C. & Bachiller, R. 1999, A&A, 350, 659

1999
[14]

2022, A&A, 667, A151

Colzi, L., Romano, D., Fontani, F., et al. 2022, A&A, 667, A151

2022
[15]

2017, MNRAS, 471, 987

Esteban, C., Fang, X., Garc \'i a-Rojas , J., & Toribio San Cipriano, L. 2017, MNRAS, 471, 987

2017
[16]

2022 a , A&A, 660, A76

Fontani, F., Colzi, L., Bizzocchi, L., et al. 2022 a , A&A, 660, A76

2022
[17]

2022 b , A&A, 664, A154

Fontani, F., Schmiedeke, A., S \'a nchez-Monge , A., et al. 2022 b , A&A, 664, A154

2022
[18]

2024, A&A, 691, A180

Fontani, F., Vermari \"e n, G., Viti, S., et al. 2024, A&A, 691, A180

2024
[19]

L., \"O berg, K

Gal, R. L., \"O berg, K. I., Teague, R., et al. 2021, ApJ Supplement Series, 257, 12

2021
[20]

2010, A&A, 519, A2

Garc \'i a-Burillo , S., Usero, A., Fuente, A., et al. 2010, A&A, 519, A2

2010
[21]

2025, Mapping Synthetic Observations to Prestellar Core Models : An Interpretable Machine Learning Approach

Grassi, T., Padovani, M., Galli, D., et al. 2025, Mapping Synthetic Observations to Prestellar Core Models : An Interpretable Machine Learning Approach

2025
[22]

D., & Van Dishoeck, E

Hacar, A., Bosman, A. D., & Van Dishoeck, E. F. 2020, A&A, 635, A4

2020
[23]

S., Mart \'i n, S., et al

Harada, N., Meier, D. S., Mart \'i n, S., et al. 2024 a , The ALCHEMI Atlas: Principal Component Analysis Reveals Starburst Evolution in NGC 253

2024
[24]

2024 b , A Temperature or FUV Tracer? The HNC / HCN Ratio in M83 on the GMC Scale

Harada, N., Saito, T., Nishimura, Y., Watanabe, Y., & Sakamoto, K. 2024 b , A Temperature or FUV Tracer? The HNC / HCN Ratio in M83 on the GMC Scale

2024
[25]

2009, The Elements of Statistical Learning , Springer Series in Statistics (New York, NY: Springer)

Hastie, T., Tibshirani, R., & Friedman, J. 2009, The Elements of Statistical Learning , Springer Series in Statistics (New York, NY: Springer)

2009
[26]

& van Dishoeck, E

Herbst, E. & van Dishoeck, E. F. 2009, Annual Review of A&A, 47, 427

2009
[27]

2023 a , MNRAS, 526, 404

Heyl, J., Butterworth, J., & Viti, S. 2023 a , MNRAS, 526, 404

2023
[28]

2023 b , Faraday Discussions, 245, 569

Heyl, J., Viti, S., & Vermari \"e n, G. 2023 b , Faraday Discussions, 245, 569

2023
[29]

2017, The Astronomical Journal, 154, 38

Holdship, J., Viti, S., Jim \'e nez-Serra , I., Makrymallis, A., & Priestley, F. 2017, The Astronomical Journal, 154, 38

2017
[30]

R., Oka, T., & McCall, B

Indriolo, N., Geballe, T. R., Oka, T., & McCall, B. J. 2007, ApJ, 671, 1736

2007
[31]

A., Viti, S., Yusef-Zadeh , F., Royster, M., & Wardle, M

James, T. A., Viti, S., Yusef-Zadeh , F., Royster, M., & Wardle, M. 2021, ApJ, 916, 69

2021
[32]

2023, 241, 208.11

Kane, S., Hawkins, K., & Maas, Z. 2023, 241, 208.11

2023
[33]

2018, A&A, 615, A122

K \"o nig, S., Aalto, S., Muller, S., et al. 2018, A&A, 615, A122

2018
[34]

2019, Correlation of Auroral Dynamics and GNSS Scintillation with an Autoencoder

Lamb, K., Malhotra, G., Vlontzos, A., et al. 2019, Correlation of Auroral Dynamics and GNSS Scintillation with an Autoencoder

2019
[35]

2015, ApJ, 802, 40

Li, J., Wang, J., Zhu, Q., Zhang, J., & Li, D. 2015, ApJ, 802, 40

2015
[36]

& Lee, S.-I

Lundberg, S. & Lee, S.-I. 2017, A Unified Approach to Interpreting Model Predictions

2017
[37]

M., Erion, G., Chen, H., et al

Lundberg, S. M., Erion, G., Chen, H., et al. 2020, Nature Machine Intelligence, 2, 56

2020
[38]

2020, UMAP : Uniform Manifold Approximation and Projection for Dimension Reduction

McInnes, L., Healy, J., & Melville, J. 2020, UMAP : Uniform Manifold Approximation and Projection for Dimension Reduction

2020
[39]

E., Amayo, A., Arellano-C \'o rdova , K

M \'e ndez-Delgado , J. E., Amayo, A., Arellano-C \'o rdova , K. Z., et al. 2022, MNRAS, 510, 4436

2022
[40]

N., Savage, C., Brewster, M

Milam, S. N., Savage, C., Brewster, M. A., Ziurys, L. M., & Wyckoff, S. 2005, ApJ, 634, 1126

2005
[41]

J., Bennett, A., Rawlings, J

Millar, T. J., Bennett, A., Rawlings, J. M. C., Brown, P. D., & Charnley, S. B. 1991, A&A Supplement Series, 87, 585

1991
[42]

2022, Interpretable Machine Learning: A Guide for Making Black Box Models Explainable, second edition edn

Molnar, C. 2022, Interpretable Machine Learning: A Guide for Making Black Box Models Explainable, second edition edn. (Munich, Germany: Christoph Molnar)

2022
[43]

1901, The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 2, 559

Pearson, K. 1901, The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 2, 559

1901
[44]

H., Clark, P

Pe \ n aloza, C. H., Clark, P. C., Glover, S. C. O., & Klessen, R. S. 2018, MNRAS, 475, 1508

2018
[45]

A., Plaza, C

Ramos, A. A., Plaza, C. W., Navarro-Almaida , D., et al. 2024, MNRAS, 531, 4930

2024
[46]

Why Should I Trust You ?

Ribeiro, M. T., Singh, S., & Guestrin, C. 2016, " Why Should I Trust You ?": Explaining the Predictions of Any Classifier

2016
[47]

P., Bell, T., et al

Rollig, M., Abel, N. P., Bell, T., et al. 2007

2007
[48]

2016, MNRAS, 459, 3756

Ruaud, M., Wakelam, V., & Hersant, F. 2016, MNRAS, 459, 3756

2016
[49]

2021, A&A, 652, A71

Sabatini, G., Bovino, S., Giannetti, A., et al. 2021, A&A, 652, A71

2021
[50]

A., Molnar, C., Heumann, C., Bischl, B., & Casalicchio, G

Scholbeck, C. A., Molnar, C., Heumann, C., Bischl, B., & Casalicchio, G. 2020, 1167, 205

2020
[51]

2018, A&A, 617, A28

Semenov, D., Favre, C., Fedele, D., et al. 2018, A&A, 617, A28

2018
[52]

B., et al

Sewi o, M., Indebetouw, R., Charnley, S. B., et al. 2018, ApJ Letters, 853, L19

2018
[53]

E., et al

Sewi o, M., Karska, A., Kristensen, L. E., et al. 2022, ApJ, 933, 64

2022
[54]

Shapley, L. S. & Shubik, M. 1971, International Journal of Game Theory, 1, 111

1971
[55]

2021, ApJ, 922, 206

Shimonishi, T., Izumi, N., Furuya, K., & Yasui, C. 2021, ApJ, 922, 206

2021
[56]

Shimonishi, T., Tanaka, K. E. I., Zhang, Y., & Furuya, K. 2023, ApJ Letters, 946, L41

2023
[57]

Sobol', I. M. 1967, USSR Computational Mathematics and Mathematical Physics, 7, 86

1967
[58]

E., et al

Spezzano, S., Caselli, P., Pineda, J. E., et al. 2020, A&A, 643, A60

2020
[59]

2021, A&A, 646, A97

Tafalla, M., Usero, A., & Hacar, A. 2021, A&A, 646, A97

2021
[60]

Usero, A., Garc \'i a-Burillo , S., Fuente, A., Mart \'i n-Pintado , J., & Rodr \'i guez-Fern \'a ndez , N. J. 2004, A&A, 419, 897

2004
[61]

& Hinton, G

van der Maaten , L. & Hinton, G. 2008, Journal of Machine Learning Research, 9, 2579

2008
[62]

& Williams, D

Viti, S. & Williams, D. A. 1999, MNRAS, 305, 755

1999
[63]

2010, A&A, 517, A21

Wakelam, V., Herbst, E., Le Bourlot, J., et al. 2010, A&A, 517, A21

2010
[64]

2022, ApJ, 937, 120

Wang, J., Qi, C., Li, S., & Wu, J. 2022, ApJ, 937, 120

2022
[65]

Williams, D. A. 1998, Faraday Discussions, 109, 1

1998
[66]

D., Bemis, A., Ledger, B., & Klimi, O

Wilson, C. D., Bemis, A., Ledger, B., & Klimi, O. 2023, MNRAS, 521, 717

2023
[67]

M., Kelly, G., Viti, S., et al

Woods, P. M., Kelly, G., Viti, S., et al. 2012, ApJ, 750, 19

2012
[68]

, " * write output.state after.block = add.period write newline

ENTRY address archiveprefix author booktitle chapter edition editor howpublished institution eprint journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sent...
[69]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

2019, Optuna: A Next-generation Hyperparameter Optimization Framework

Akiba, T., Sano, S., Yanase, T., Ohta, T., & Koyama, M. 2019, Optuna: A Next-generation Hyperparameter Optimization Framework

work page 2019

[2] [2]

& Faure, A

Bacmann, A. & Faure, A. 2016, A&A, 587, A130

work page 2016

[3] [3]

W., Williams, D

Bayet, E., Hartquist, T. W., Williams, D. A., et al. 2011, Memorie della Societa Astronomica Italiana, 82, 893

work page 2011

[4] [4]

A., & Rawlings, J

Bayet, E., Viti, S., Williams, D. A., & Rawlings, J. M. C. 2008, ApJ, 676, 978

work page 2008

[5] [5]

A., Rawlings, J

Bayet, E., Viti, S., Williams, D. A., Rawlings, J. M. C., & Bell, T. 2009, ApJ, 696, 1466

work page 2009

[6] [6]

G., Holdship, J., et al

Behrens, E., Mangum, J. G., Holdship, J., et al. 2022, ApJ, 939, 119

2022

[7] [7]

J., Sephus, C

Bernal, J. J., Sephus, C. D., & Ziurys, L. M. 2021, ApJ, 922, 106

work page 2021

[8] [8]

D., Charnley, S

Brown, P. D., Charnley, S. B., & Millar, T. J. 1988, MNRAS, 231, 409

work page 1988