Pith. sign in

REVIEW 2 major objections 4 minor 52 references

How to set up your first machine learning project in astronomy

T0 review · 2 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read Machine learning astronomy projects fail less when they start from a specific objective, trivial baselines, uncertainty handling, and a test harness.

desk verdict Useful best-practices guide for astronomy ML, but a wrong definition of precision and a purity-based model-selection example undermine its own reliability and need correction. read the letter →

arxiv 2502.08222 v1 pith:YHSEA4W7 submitted 2025-02-12 astro-ph.IM

Johannes Buchner , Sotiria Fotopoulou This is my paper · ORCID
classification astro-ph.IM
keywords machinelearningastronomyprojectworkflowbaselinestestharnessuncertaintyquantificationcovariateshiftreproducibility
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

Astronomy's large public datasets have made it a popular playground for machine learning, but many projects end without a model that is actually used or trusted. The paper argues that the usual cause is not a weak algorithm but an underspecified project. It lays out a workflow: state a specific, measurable objective that implies the required error rates; define a finish line; establish dummy, historical, and simple baselines; validate the pipeline on a test harness with known truth; ingest and report uncertainties; and correct for covariate shift between training and application data. If followed, the authors claim, projects are more likely to produce robust scientific insight and less likely to consume months on dead ends.

What carries the argument

The load-bearing mechanism is the test harness together with the baseline ladder. A test harness is an external validation setup, such as simulated data, hand-crafted inputs like a two-dimensional Gaussian on noise, or a small independent ground-truth sample, that the pipeline must pass before being trusted on real data. The baseline ladder runs from the dummy predictor that always returns the most common class, through the historical method the project aims to replace, to simple standard models with default settings. The paper uses these two devices to turn machine learning from open-ended tinkering into a measurable engineering process: the baselines define what counts as improvement, and the test harness defines what counts as understanding. Calibrated uncertainty output, judged by the distribution of true values on the cumulative predicted distribution, is the third pillar that makes the results usable in downstream science.

What would settle it

A systematic audit of published astronomy ML projects would falsify the paper's causal claim if it found no association between using a test harness, baselines, and uncertainty calibration and later independent verification or adoption of the model; the check would be whether models from workflow-following projects are more often confirmed by new observations.

Watch

Extended reading notes

Core claim

The paper's central claim is that the success of a machine-learning project in astronomy is largely decided before any model is trained, by the clarity of the objective and the honesty of the validation setup. It is a recommendation document, not a new algorithm. The authors contend that an objective should be specific enough to determine the required true-positive and false-positive rates, that trivial baselines must be reported beside any sophisticated method, that the full pipeline should be exercised on simulated or independently known data before real application, and that a scientific model is incomplete unless it produces uncertainties that have been checked for calibration. The intended result is a literature in which successes, failures, and equally-good variants are reported, so the community can learn which methods suit which problems.

Load-bearing premise

The workflow assumes that a high-quality simulation or independent ground-truth data set exists to serve as the test harness; for rare or poorly understood phenomena, no such external check may be available, and the validation step cannot be fully applied.

Editorial extensions

If this is right

  • Projects with an explicit finish line can stop when the target is reached instead of tuning indefinitely.
  • Every model comparison becomes more meaningful when dummy and historical baselines appear in the same plot.
  • Calibrated per-object uncertainties let downstream Bayesian analyses ingest ML predictions as probability distributions rather than point estimates.
  • Reporting failed variations and equally-good hyperparameters would save the community from repeating the same searches.
  • Covariate-shift corrections, such as weighting training samples by their probability of belonging to the training set, are needed whenever labeled sources are brighter or nearer than the application sample.

Reading between the lines

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

  • The test-harness step could be operationalized as a mandatory submission check for astronomy ML papers: a rank-histogram plot and baseline table would let reviewers see calibration at a glance.
  • The same workflow transfers to other sciences with expensive labels, where the labeled sample is systematically different from the unlabeled population.
  • A quantitative prediction of the paper is that teams following the workflow should have their models independently verified or adopted more often; this could be tested by auditing whether published models are later checked against new observations.
  • Framing the test harness as an engineering gate suggests a sharper criterion than 'the model works': a model should not be deployed until it passes adversarial hybrid inputs, such as two galaxy images summed and still classified as a galaxy.
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

2 major / 4 minor

Summary. The paper is a perspective/recommendation document for astronomers and data scientists starting machine learning projects. It argues that ML projects should begin with a specific, measurable, and useful objective; establish dummy, historical, and initial baselines; create a test harness with simulated or independent ground-truth data; ingest and produce uncertainties; address covariate shift; and report ablation studies and failed trials. It also includes a dictionary that maps ML terms to astronomy terms, discusses non-technical aspects of the work environment, and provides workflow diagrams. The paper contains no new empirical results or derivations; its claims are heuristic and grounded in external references and well-known ML methodology.

Significance. If followed, the proposed workflow could reduce common sources of error and frustration in astronomy ML projects, particularly through its emphasis on explicit objectives, baselines, uncertainty handling, and out-of-distribution validation. The paper is clearly written, well-structured, and uses concrete examples (e.g., the eFEDS covariate-shift illustration) to make its points accessible. Its strengths include an unusually explicit glossary, a strong recommendation to report failures as well as successes, and practical advice on K-fold cross-validation and ablation studies. However, a formal error in the definition of precision—repeated in the dictionary and in two key sections—undermines the paper's central model-selection recommendation and must be corrected before the paper can serve as a reliable reference.

major comments (2)
  1. [Dictionary; §2.1; §2.3.1] The Dictionary defines precision as '1 - false positive rate'. This is incorrect: precision = TP/(TP+FP), whereas 1 - FPR = TN/(FP+TN), which is specificity (true negative rate). The same conflation appears in §2.1, where the paper suggests that reporting 'true positive rate (completeness) and false positive rate (purity)' allows the user to assess the objective, and in §2.3.1, where calibrating both classifiers to a 1% false positive rate is described as encoding 'pure samples'. Purity is precision, not 1 - FPR. Under class imbalance, a fixed FPR can correspond to very different precision values, so an FPR-calibrated comparison does not generally select the classifier that yields the purest sample. This is load-bearing because the paper's recommended model-selection criterion (calibrate to a desired FPR then compare TPR) is presented as equivalent to choosing a pure sample. Please correct the glossary definition and revise the relevant passages to use precision (or explicitly discuss the connection between FPR and precision under the assumed operating point).
  2. [§2.4 (Create a test harness)] The test-harness recommendation assumes that high-quality simulations or independent ground-truth datasets are available for the problem under study. The paper does not address the common case where no such external validation exists, such as classification of rare transient phenomena whose physical classes are not yet known, or problems where simulations are unavailable or not faithful. Without an alternative strategy (e.g., internal consistency checks, researcher-declared diagnostic experiments, or semi-supervised agreement metrics), the universal claim that a test harness 'should' be created may not be actionable for a substantial fraction of astronomy projects. I recommend adding a brief discussion of fallback validation strategies when no simulation or independent ground truth is available.
minor comments (4)
  1. [§5.3] The sentence 'Similarly, in classification balanced training samples or imbalance corrections are needed' is grammatically incomplete; it appears to intend 'in classification, either balanced training samples or imbalance corrections are needed'.
  2. [Dictionary] The entry for 'unbiased (regression)' contains a typo: 'on on the training sample' should read 'on the training sample'.
  3. [§3.3 (Unsupervised learning)] The text reads 'also known aspareidolia' with a missing space; it should be 'also known as pareidolia'.
  4. [Figure 1 caption] The caption states that 'Any confusion matrix corresponds to a position on the receiver operator curve (ROC)'; this is true only for a confusion matrix at a specified threshold, and the phrasing could be tightened to avoid implying that every confusion matrix lies on the ROC curve.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper is a guidelines document with no derivation chain whose conclusions reduce to its inputs.

full rationale

This paper is a recommendations-style guide for setting up machine learning projects in astronomy. It contains no fitted parameters, no predictive model, and no derivation that could reduce to its own inputs. The central claims—that projects should have measurable objectives, baselines, uncertainty handling, and test-harness validation—are supported by external references to general ML practice, methodology literature, and standard statistical concepts. Self-references, such as 'unsupervised learning (Fotopoulou, sub.)' and '(Singhal, Fotopoulou, et al., in prep)', are mentions of ongoing or related work, not load-bearing evidence for any derived result. The only notable issue is a definitional error in the Dictionary ('precision: 1 - false positive rate'), which is a correctness concern, not a circularity concern: it does not make any conclusion equivalent to its premises. No step in the paper is self-definitional, no prediction is fitted and then reported as validated, and no uniqueness claim is imported from the authors' prior work. The circularity burden is therefore zero.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

No free parameters or invented entities appear. The paper relies on established ML methodology from the literature and the unproven premise that these practices apply to astronomy problems.

assumptions (2)
  • domain assumption Standard machine learning evaluation practices, including train/test splits, K-fold cross-validation, and dummy baselines, are applicable to astronomy datasets without modification.
    Invoked in Sections 4.1-4.4 and 5.6; no astronomy-specific demonstration that these practices transfer is provided.
  • domain assumption The target audience has sufficient ML background to implement the recommendations, as stated in Section 1 ('It is assumed you are familiar with ML fundamentals').
    The guidelines depend on this prerequisite knowledge but do not provide training in it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of How to set up your first machine learning project in astronomy." pith.science (2026). https://pith.science/paper/YHSEA4W7

@misc{pith2026250208222,
  author       = {Pith},
  title        = {Pith review of: How to set up your first machine learning project in astronomy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YHSEA4W7}},
  note         = {Machine review of arXiv:2502.08222}
}
read the original abstract

Large, freely available, well-maintained data sets have made astronomy a popular playground for machine learning projects. Nevertheless, robust insights gained to both machine learning and physics could be improved by clarity in problem definition and establishing workflows that critically verify, characterize and calibrate machine learning models. We provide a collection of guidelines to setting up machine learning projects to make them likely useful for science, less frustrating and time-intensive for the scientist and their computers, and more likely to lead to robust insights. We draw examples and experience from astronomy, but the advice is potentially applicable to other areas in science.

Figures

Figures reproduced from arXiv: 2502.08222 by the authors.

Figure 1
Figure 1. The confusion matrix (left) reports the classification numbers needed to compute the false positive rate (FPR) and the false negative rate (FNR) of a classifier at a certain threshold X, for example the classification probability threshold. If computed with k-fold cross-validation, these rates can be reported with uncertainties. Any confusion matrix corresponds to a position on the receiver operator curve (ROC), and… view at source ↗
Figure 2
Figure 2. From the predicted probability density function (blue curve in the top panel) of one validation sample, the probability of the true value (purple) can be read off (ptruth). The higher the product of these probabilities over the entire validation sample, the more informative the prediction. The middle panel shows the corresponding cumulative probability function, from which the cumulative probability of the true valu… view at source ↗
Figure 3
Figure 3. At distribution tails, regression trends towards the bulk of the training sample’s target. A normal distributed sample was generated (blue), and K-folded out-of-sample prediction of a neural network is shown (red). 10 15 20 25 Magnitude (r band) 0 1000 2000 3000 Number class=AGN (ground truth, from spectrum) class=Star (ground truth, from spectrum) probably class=AGN (application domain, no spectrum) probably class=… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Illustration covariate shift from the eROSITA Full-depth equatorial survey (eFEDS;34). From expensive ground truth (spectroscopy), some of the sample was classified into AGN (blue) and stars (yellow), and can be used as a training sample for classification of the remai…
Figure 5
Figure 5. Figure 5: Typical workflow in introductory machine learning tutorials and data challenges. The data set is split into a training set used to optimize hyper-parameters θ, yielding a best method. The performance of this method is characterized with an unseen validation dataset. So…
Figure 6
Figure 6. Figure 6: A workflow for predicting with trustworthy uncertainties. A small portion of the training data is held out to calibrate the uncertainties. This can be considered part of the training. Instead of training once, K-folding (here K = 4) trains with different seeds and data…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 36 canonical work pages

  1. [1]

    Mandt, S., Hoffman, M. D. & Blei, D. M. Stochastic Gradient Descent as Approximate Bayesian Inference. arXiv e-prints arXiv:1704.04289, DOI: 10.48550/arXiv.1704.04289 (2017). 1704.04289

    arXiv doi:10.48550/arxiv.1704.04289 2017

  2. [2]

    The crisp-dm model: the new blueprint for data mining

    Shearer, C. The crisp-dm model: the new blueprint for data mining. J. data warehousing 5, 13–22 (2000)

    work page 2000

  3. [3]

    Saltz, J. S. The need for new processes, methodologies and tools to support big data teams and improve big data project effectiveness. In 2015 IEEE International Conference on Big Data (Big Data), 2066–2071 (IEEE, 2015)

    work page 2015

  4. [4]

    Data Science Methodologies: Current Challenges and Future Approaches

    Martinez, I., Viles, E. & Olaizola, I. G. Data Science Methodologies: Current Challenges and Future Approaches. arXiv e-prints arXiv:2106.07287, DOI: 10.48550/arXiv.2106.07287 (2021). 2106.07287

    work page Pith review arXiv doi:10.48550/arxiv.2106.07287 2021

  5. [5]

    Artrith, N. et al. Best practices in machine learning for chemistry. Nat. Chem. 13, 505–508, DOI: 10.1038/ s41557-021-00716-z (2021)

    work page 2021

  6. [6]

    & Ventre, G

    Garofalo, M., Botta, A. & Ventre, G. Astrophysics and big data: Challenges, methods, and tools. Proc. Int. Astron. Union 12, 345–348 (2016). 17/20

    work page 2016

  7. [7]

    & Zhao, Y

    Zhang, Y . & Zhao, Y . Astronomy in the Big Data Era.Data Sci. J. 14, 11, DOI: 10.5334/dsj-2015-011 (2015)

    doi:10.5334/dsj-2015-011 2015

  8. [8]

    Deep Machine Learning in Cosmology: Evolution or Revolution?

    Lahav, O. Deep Machine Learning in Cosmology: Evolution or Revolution? arXiv e-prints arXiv:2302.04324, DOI: 10.48550/arXiv.2302.04324 (2023). 2302.04324

    work page Pith review arXiv doi:10.48550/arxiv.2302.04324 2023

Show all 52 references
  1. [9]

    Scientific Data Mining in Astronomy

    Borne, K. Scientific Data Mining in Astronomy. arXiv e-prints arXiv:0911.0505, DOI: 10.48550/arXiv.0911. 0505 (2009). 0911.0505

    2009 doi arXiv

  2. [10]

    G., Mahabal, A

    Djorgovski, S. G., Mahabal, A. A., Graham, M. J., Polsterer, K. & Krone-Martins, A. Applications of AI in Astronomy. arXiv e-prints arXiv:2212.01493, DOI: 10.48550/arXiv.2212.01493 (2022). 2212.01493

    2022 doi arXiv

  3. [11]

    Fluke, C. J. & Jacobs, C. Surveying the reach and maturity of machine learning and artificial intelligence in astronomy. WIREs Data Min. Knowl. Discov. 10, e1349, DOI: 10.1002/widm.1349 (2020). 1912.02934

    2020 arXiv

  4. [12]

    J., VanderPlas, J

    Ivezi´c, Ž., Connolly, A. J., VanderPlas, J. T. & Gray, A. Statistics, Data Mining, and Machine Learning in Astronomy: A Practical Python Guide for the Analysis of Survey Data, Updated Edition (2019)

    2019

  5. [13]

    Machine Learning in Astronomy: a practical overview

    Baron, D. Machine Learning in Astronomy: a practical overview. arXiv e-prints arXiv:1904.07248, DOI: 10.48550/arXiv.1904.07248 (2019). 1904.07248

    2019 doi arXiv

  6. [14]

    Mastering Machine Learning with scikit-learn (Packt, 2017)

    Hackeling, G. Mastering Machine Learning with scikit-learn (Packt, 2017)

    2017

  7. [15]

    G., Mahabal, A

    Graham, M., Drake, A., Djorgovski, S. G., Mahabal, A. & Donalek, C. Challenges in the automated classification of variable stars in large databases. In European Physical Journal Web of Conferences, vol. 152 of European Physical Journal Web of Conferences, 03001, DOI: 10.1051/e...

    2017

  8. [16]

    Yang, H. et al. Data mining techniques on astronomical spectra data - II. Classification analysis. MNRAS 518, 5904–5928, DOI: 10.1093/mnras/stac3292 (2023). 2212.09286

    2023 arXiv

  9. [17]

    Yang, H. et al. Data mining techniques on astronomical spectra data - I. Clustering analysis. MNRAS 517, 5496–5523, DOI: 10.1093/mnras/stac2975 (2022). 2212.08419

    2022 arXiv

  10. [18]

    Active learning literature survey

    Settles, B. Active learning literature survey. Computer Sciences Technical Report 1648, University of Wisconsin– Madison (2009)

    2009

  11. [19]

    & Bassett, B

    Lochner, M. & Bassett, B. A. ASTRONOMALY: Personalised active anomaly detection in astronomical data. Astron. Comput. 36, 100481, DOI: 10.1016/j.ascom.2021.100481 (2021). 2010.11202

    2021

  12. [20]

    The mnist database of handwritten digit images for machine learning research

    Deng, L. The mnist database of handwritten digit images for machine learning research. IEEE Signal Process. Mag. 29, 141–142 (2012)

    2012

  13. [21]

    Doran, G. T. et al. There’sa smart way to write management’s goals and objectives. Manag. review 70, 35–36 (1981)

    1981

  14. [22]

    Bausell, R. B. & Li, Y .-F. Power analysis for experimental research: a practical guide for the biological, medical and social sciences (Cambridge University Press, 2002)

    2002

  15. [23]

    operational definition, DOI: 10.1093/oi/authority.20110803100251503

    doi

  16. [24]

    Spectra of Supernovae

    Minkowski, R. Spectra of Supernovae. PASP 53, 224, DOI: 10.1086/125315 (1941)

    1941 doi

  17. [25]

    Thornton, C., Hutter, F., Hoos, H. H. & Leyton-Brown, K. Auto-weka: Combined selection and hyperparameter optimization of classification algorithms. In Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’13, 847–855, DOI: 10...

  18. [26]

    Erickson, N. et al. AutoGluon-Tabular: Robust and Accurate AutoML for Structured Data. arXiv e-prints arXiv:2003.06505, DOI: 10.48550/arXiv.2003.06505 (2020). 2003.06505

    2020 doi arXiv

  19. [27]

    Lieu, M. et al. Deep Learning of Astronomical Features with Big Data. In Teuben, P. J., Pound, M. W., Thomas, B. A. & Warner, E. M. (eds.)Astronomical Data Analysis Software and Systems XXVII, vol. 523 of Astronomical Society of the Pacific Conference Series, 49 (2019). 18/20

    2019

  20. [28]

    Interpretable Machine Learning (2022), 2 edn

    Molnar, C. Interpretable Machine Learning (2022), 2 edn

    2022

  21. [29]

    Rudin, C. et al. Interpretable machine learning: Fundamental principles and 10 grand challenges. Stat. Surv. 16, 1–85 (2022)

    2022

  22. [30]

    & Galstyan, A

    Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K. & Galstyan, A. A survey on bias and fairness in machine learning. ACM computing surveys (CSUR) 54, 1–35 (2021)

    2021

  23. [31]

    & Salakhutdinov, R

    Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I. & Salakhutdinov, R. Dropout: A simple way to prevent neural networks from overfitting. J. Mach. Learn. Res. 15, 1929–1958 (2014)

    2014

  24. [32]

    Hamill, T. M. Interpretation of Rank Histograms for Verifying Ensemble Forecasts. Mon. Weather. Rev. 129, 550, DOI: 10.1175/1520-0493(2001)129<0550:IORHFV>2.0.CO;2 (2001)

    2001 doi

  25. [33]

    Ghosh, A. et al. GaMPEN: A Machine-learning Framework for Estimating Bayesian Posteriors of Galaxy Morphological Parameters. ApJ 935, 138, DOI: 10.3847/1538-4357/ac7f9e (2022). 2207.05107

    2022 arXiv

  26. [34]

    Salvato, M. et al. The eROSITA Final Equatorial-Depth Survey (eFEDS). Identification and characterization of the counterparts to point-like sources. Astron. Astrophys. 661, A3, DOI: 10.1051/0004-6361/202141631 (2022). 2106.14520

    2022 arXiv

  27. [35]

    Rosenbaum, P. R. & Rubin, D. B. Reducing bias in observational studies using subclassification on the propensity score. J. Am. statistical Assoc. 79, 516–524 (1984)

    1984

  28. [36]

    A., Trotta, R

    Revsbech, E. A., Trotta, R. & van Dyk, D. A. STACCATO: a novel solution to supernova photometric classification with biased training sets. MNRAS 473, 3969–3986, DOI: 10.1093/mnras/stx2570 (2018). 1706. 03811

    2018 doi

  29. [37]

    Domain-adversarial training of neural networks

    Ganin, Y .et al. Domain-adversarial training of neural networks. The journal machine learning research 17, 2096–2030 (2016)

    2016

  30. [38]

    Perdue, G. N. et al. Reducing model bias in a deep learning classifier using domain adversarial neural networks in the MINERvA experiment. J. Instrumentation 13, P11020, DOI: 10.1088/1748-0221/13/11/P11020 (2018). 1808.08332

    2018 arXiv

  31. [39]

    & Stitelman, O

    Kaufman, S., Rosset, S., Perlich, C. & Stitelman, O. Leakage in data mining: Formulation, detection, and avoidance. ACM Trans. Knowl. Discov. Data6, DOI: 10.1145/2382577.2382579 (2012)

    2012

  32. [40]

    & Narayanan, A

    Kapoor, S. & Narayanan, A. Leakage and the reproducibility crisis in machine-learning-based science. Patterns DOI: 10.1016/j.patter.2023.100804 (2023). Publisher: Elsevier

    2023

  33. [41]

    The logic of scientific discovery (Routledge, 2005)

    Popper, K. The logic of scientific discovery (Routledge, 2005)

    2005

  34. [42]

    Hopkins, P. F. A new class of accurate, mesh-free hydrodynamic simulation methods. MNRAS 450, 53–110, DOI: 10.1093/mnras/stv195 (2015). 1409.7395

    2015 arXiv

  35. [43]

    Carleo, G. et al. Machine learning and the physical sciences. Rev. Mod. Phys. 91, 045002, DOI: 10.1103/ RevModPhys.91.045002 (2019). 1903.10563

    2019 arXiv

  36. [44]

    & Tang, K

    Zhang, Y ., Tiˇno, P., Leonardis, A. & Tang, K. A survey on neural network interpretability. IEEE Transactions on Emerg. Top. Comput. Intell.5, 726–742, DOI: 10.1109/TETCI.2021.3100641 (2021)

    2021

  37. [45]

    & Wang, G

    Fan, F.-L., Xiong, J., Li, M. & Wang, G. On interpretability of artificial neural networks: A survey. IEEE Transactions on Radiat. Plasma Med. Sci. 5, 741–760, DOI: 10.1109/TRPMS.2021.3066428 (2021)

    2021

  38. [46]

    J., Shlens, J

    Goodfellow, I. J., Shlens, J. & Szegedy, C. Explaining and harnessing adversarial examples. arXiv preprint arXiv:1412.6572 (2014)

    2014 arXiv

  39. [47]

    Wilkinson, M. D. et al. The FAIR Guiding Principles for scientific data management and stewardship. Sci. Data 3, 160018, DOI: 10.1038/sdata.2016.18 (2016)

    2016 doi

  40. [48]

    Chamba, N., Knapen, J. H. & Black, D. How to plan your astronomy research paper in ten steps. Nat. Astron. 6, 1015–1020, DOI: 10.1038/s41550-022-01757-1 (2022). 2207.12959. 19/20

    2022 arXiv

  41. [49]

    H., Chamba, N

    Knapen, J. H., Chamba, N. & Black, D. How to write and develop your astronomy research paper. Nat. Astron. 6, 1021–1026, DOI: 10.1038/s41550-022-01759-z (2022). 2110.05503

    2022 arXiv

  42. [50]

    Writing a Scientific Paper I

    Sterken, C. Writing a Scientific Paper I. The writing process. In Sterken, C. (ed.) EAS Publications Series, vol. 50 of EAS Publications Series, 1–63, DOI: 10.1051/eas/1150001 (2011)

    2011

  43. [51]

    Writing a Scientific Paper II

    Sterken, C. Writing a Scientific Paper II. Communication by Graphics. In Sterken, C. (ed.) EAS Publications Series, vol. 50 of EAS Publications Series, 65–170, DOI: 10.1051/eas/1150002 (2011)

    2011

  44. [52]

    Writing a Scientific Paper III

    Sterken, C. Writing a Scientific Paper III. Ethical Aspects. In Sterken, C. (ed.) EAS Publications Series, vol. 50 of EAS Publications Series, 173–282, DOI: 10.1051/eas/1150003 (2011). Acknowledgements The thoughts laid out here have been heavily influenced by conversations wi...

    2011

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.