Pith. sign in

REVIEW 3 major objections 5 minor 51 references

Identifying Causal Direction via Dense Functional Classes

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A minimum description length ratio over dense functional classes asymptotically identifies causal direction in low-noise bivariate additive noise models; the cubic-spline instantiation LCUBE reports high precision on benchmarks.

desk verdict Plausible method, strong empirical numbers, but the identifiability proof does not go through as written: the score is not an IRSF because λ depends on n and knot spacings, not just parameter count. read the letter →

arxiv 2509.00538 v1 pith:3W6G7KYH submitted 2025-08-30 stat.ML cs.LGstat.CO

Katerina Hlavackova-Schindler , Suzana Marsela This is my paper · ORCID
classification stat.MLcs.LGstat.CO
keywords causaldirectionminimumdescriptionlengthdensefunctionalclasscubicsplinesidentifiabilityadditivenoisemodellowLCUBE
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 aims to prove that a Minimum Description Length causal score, defined through any functional class dense in the continuous functions on a compact interval, can separate cause from effect in bivariate additive noise models with low noise and no hidden confounders. Its main theorem states that the MDL score in the true causal direction is asymptotically at most the score in the anti-causal direction, with equality only for linear functions. The paper instantiates this with cubic regression splines in a method called LCUBE, which it shows is identifiable, has one hyperparameter, and runs quickly; it reports an AUDRC of 87 on the Tuebingen pairs and perfect AUDRC on most low-noise synthetic benchmarks. If correct, this gives practitioners a fast, principled way to decide causal direction under testable conditions and without assuming Gaussian noise.

What carries the argument

The load-bearing object is the score ratio L(Y|X, θ_X) / L(X|Y, θ_Y), where each code is a two-part MDL description: a parameter part log m + Σ log u_j + ((m+4)/2) log n and a goodness-of-fit part (n/2) log(RSS/n). The proof works by showing this score is an IRSF—a scoring function that adds a strictly increasing penalty for error to a strictly increasing penalty for parameter count—so the ratio inherits the regression-error ratio of Assumptions 1–3. The density property of the function class lets the approximation error be absorbed into the low-noise term, and Assumption 4 keeps the anti-causal model from being simpler than the causal one. For LCUBE, cubic splines supply the dense class and

What would settle it

Simulate X → Y with a nonlinear spline function f and low noise, then choose a function whose inverse is a spline with fewer knots; compute the LCUBE ratio for increasing n. If the ratio eventually exceeds 1, Assumption 4 is violated and Theorem 1's conclusion fails; a direct linearity check of f should likewise predict whether the method goes undecided.

Watch

Extended reading notes

Core claim

The paper proves that for a bivariate additive noise model Y = f(X) + (1/n)N, with compactly supported cause and low noise, the direction X → Y is identifiable from a ratio of two-part MDL code lengths. Theorem 1 states that if the regression functions come from any dense functional class and Assumptions 1–4 hold, then L(Y|X, θ_X) / L(X|Y, θ_Y) converges to at most 1 as n grows, with equality only when f is linear. The instantiation LCUBE uses cubic regression splines, which are dense in C([0,1]); its score is an IRSF, so the theorem applies. The paper also reports that LCUBE achieves AUDRC 87 on the Tuebingen pairs and perfect AUDRC on several low-noise synthetic benchmarks, with no Gaussia

Load-bearing premise

The load-bearing premise is Assumption 4: the best-fitting anti-causal function must need at least as many parameters as the best-fitting causal function; if a more compact inverse representation exists, the inequality behind the theorem can reverse.

Editorial extensions

If this is right

  • Under the theorem, any dense functional class satisfying Assumptions 1–4 inherits the same asymptotic identifiability result, so the method is a template rather than a spline-only trick.
  • The score remains valid for non-Gaussian low-noise errors, extending regression-error causal inference beyond Gaussian likelihood settings.
  • For linear f, the score ratio converges to 1, so the algorithm should output 'undecided' rather than a direction; users should expect ties exactly in the linear case.
  • LCUBE's one hyperparameter and low computational cost make it practical for large collections of bivariate pairs, where neural approaches need hours.
  • Because the assumptions are testable, practitioners can check a given dataset for compact support and low noise before trusting the direction output.

Reading between the lines

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

  • Editorial inference: Assumption 4 is checkable on each dataset by comparing fitted parameter counts in both directions; if the anti-causal spline is systematically more compact, the claimed limit can fail and the score should not be trusted.
  • Editorial inference: because the theorem is stated for any dense class, the same proof strategy could be instantiated with other dense bases, but the paper only works out cubic splines; extending to another class would require a new practical MDL encoding and a new check of Assumption 4.
  • Editorial inference: the strong empirical numbers come mostly from low-noise synthetic benchmarks and one real-world collection; a wider set of real cause-effect pairs with known ground truth would test how the asymptotic guarantee holds at finite sample sizes.
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.

Referee Report

3 major / 5 minor

Summary. The paper proposes an MDL-based bivariate causal score for low-noise additive noise models, using function classes that are dense in C([0,1]). Theorem 1 claims that, under Assumptions 1-4, the ratio of conditional code lengths L(Yn|X,thetaX)/L(X|Yn,thetaY) converges to at most 1, with equality iff the regression function is linear. The method LCUBE instantiates the score with cubic regression splines. Corollary 1 asserts that the LCUBE score is identifiable by showing that the code length in Eq. (24) is an IRSF. The paper reports experiments on synthetic benchmarks and the Tuebingen cause-effect pairs dataset, with competitive or superior AUDRC/accuracy compared to several state-of-the-art methods.

Significance. If the identifiability theorem were correct, the paper would give a clean theoretical justification for a simple, fast, one-hyperparameter MDL score over dense function classes, going beyond the AIC/BIC instantiations of the IRSF framework. The empirical contribution is real: the authors provide public code, evaluate on a wide range of benchmarks, and report strong AUDRC on the Tuebingen pairs. However, the central proof connecting LCUBE to the IRSF framework is invalid as written: the score in Eq. (24) is not an IRSF in the sense of Definition 1, and Assumption 4 is asserted rather than derived for cubic splines. The theoretical claim is therefore not currently established, although it may be repairable with a substantial revision.

major comments (3)
  1. [Section V-H, Eq. (25)] The proof of Corollary 1 defines lambda(||beta_phi||_0) := log m + sum_j log u_j + ((m+4)/2) log n. This depends on the data-derived knot spacings u_j defined in Eq. (22) and on the sample size n, not solely on the parameter count ||beta_phi||_0 as Definition 1 requires. Two spline models with the same number of parameters and the same RSS can therefore receive different values of L(Y|X,theta). Consequently Eq. (24) is not an IRSF, and Lemma 1 cannot be invoked for LCUBE. This is load-bearing for both Theorem 1 and Corollary 1.
  2. [Section IV, proof of Theorem 1] The verification of the gamma condition is also not in IRSF form. The authors write gamma := (n/2) log(RSS(theta)/n) and say it is strictly increasing 'for n'. Definition 1 requires gamma: R_{>=0} -> R to be strictly increasing in the expected squared error only. The factor n and the use of the finite-sample RSS mean that the score is a function of (n, RSS, knot positions), not of (ELSE, parameter count). The asymptotic convergence RSS/n -> ELSE does not make the finite-sample score an IRSF.
  3. [Section V-H and Section VI-A, Assumption 4] Assumption 4 is load-bearing for Lemma 1 and Corollary 1. For cubic splines it is asserted rather than derived. The text says 'we can therefore assume' and cites Kilbertus et al. [29] for low-degree polynomials, but no argument is given for splines with variable or equidistant knots. The intuitive smoothness discussion in Section VI-A does not establish ||beta_phi||_0 <= ||beta_psi||_0. If the anti-causal spline can be represented with fewer effective parameters, the ratio inequality in Lemma 1 can fail.
minor comments (5)
  1. [Section IV, proof of Theorem 1] Typo: 'numerator and denominator' is written as 'nomonator an denominator'. Also, the sentence claiming that fixed parameterizations 'do not affect the limit' is unclear and should be expanded or removed.
  2. [Section III-A, Eq. (3)] The transition from Eq. (2) to Eq. (3) is inconsistent: the text mentions approximation error 1/(2n)N and then writes Y_n ≈ h(X) + (1/n)N. Please clarify the noise scaling.
  3. [Section V-H, Eq. (25)] The claim that the product m * prod_j u_j * n^{(m+4)/2} is a natural number for even m does not address the more serious issue that lambda must depend only on the parameter count. Also, if any u_j = 0, log u_j is undefined; the Schoenberg-Whitney conditions should be explicitly imposed in Algorithm 1 and checked in practice.
  4. [References] References [8] and [43] are the same paper (Galkin), and [2] and [25] are the same paper (Janzing and Schoelkopf). Please consolidate.
  5. [Table I] The column headings are confusing: '1. column', '2. column', and '2. column' are repeated. Clarify which columns correspond to the 10-dataset versus 13-dataset averages.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation is conditional on external, independently stated assumptions, and the causal score is not fitted to the target direction.

full rationale

The paper's central claim is conditional on Assumptions 1-4 and is justified by importing external results: the IRSF framework and its identifiability theorem from Marx and Vreeken [30], and the RECI error-ratio inequality from Blobaum et al. [31]. These are not the authors' own prior claims, and they are not verified by fitting to the target. The MDL score L(Y|X,theta) is computed from the data by Algorithm 1, and the causal decision is the argmin of two independently computed scores; the true direction is not used to construct the score. There is no step where a fitted parameter is renamed as a prediction, and no load-bearing self-citation. The proof of Corollary 1 does contain a formal gap: Eq. (25) defines lambda to include knot spacings u_j and sample size n, so L in Eq. (24) is not literally an IRSF in the sense of Definition 1 (which requires lambda to depend only on the parameter count), and the sentence 'Function lambda is strictly increasing function for n' changes the argument from parameter count to n. This is a correctness risk in the identifiability proof, but it is not circularity: the theorem's conclusion does not reduce to its own assumption, and the cited IRSF/RECI theorems are independent external results. Minor self-citations ([14], [15]) appear only in auxiliary approximation-rate discussions and are not load-bearing.

Assumptions & free parameters 1 free parameters · 7 assumptions · 0 invented entities

The central theorem imports the ANM and noise assumptions from [31] and the IRSF and simplicity conditions from [30]. On top of those, it adds the density property of cubic splines and a specific MDL code. The main additional burden is Assumption 4, which is not proved for splines, plus the claim that the MDL code satisfies the IRSF definition despite depending on n and on knot-spacing data.

free parameters (1)
  • mmax, the maximum number of knots = not reported in the paper
    This is the single hyperparameter of LCUBE. It bounds the number of equidistant knots in Algorithm 1, and its value for the reported experiments is not stated, nor is its sensitivity analyzed.
assumptions (7)
  • domain assumption Assumption 1: the effect can be written as Y = f(X) + alpha N, an additive noise model.
    Imported directly from Blobaum et al. [31], Eq. (1). This is the modeling assumption that makes regression-based causal scoring meaningful.
  • domain assumption Assumption 2: the noise N is unbiased with unit variance.
    Imported from [31], used in the proof of Lemma 1 and Theorem 1. The paper does not justify this property; it merely states it.
  • domain assumption Assumption 3: X and N have compact supports, and Y is rescaled to [0,1].
    Imported from [31]. The paper says compact supports are assumed for numerical stability but does not show the rescaling is performed in the algorithm.
  • domain assumption Assumption 4: the anti-causal regression psi has at least as many parameters as the causal regression phi, ||beta_phi||_0 <= ||beta_psi||_0.
    Imported from [30] and asserted for cubic splines in Section V-H. This is a load-bearing premise for the ratio inequality and is not actually derived for spline models.
  • standard math The class of cubic splines with increasing knots is dense in C([0,1]).
    Used to justify replacing f with a spline approximation; the paper cites Subbotin [42] and Wahba [17]. This is a standard approximation-theory result.
  • domain assumption The knots used in the code length are a subsequence of the observed x_i and satisfy the Schoenberg-Whitney conditions.
    The code length for k in Eq. (21) relies on the u_j being positive integers derived from gaps between data points, while the algorithm uses equidistant knots. This mismatch is an unstated additional assumption.
  • domain assumption MRSS converges to ELSE for large n.
    Used in the proof of Theorem 1 to pass from residual sum of squares to expected least-squared error. The paper cites Vapnik [18] but does not state the uniformity conditions needed for the spline model class.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Identifying Causal Direction via Dense Functional Classes." pith.science (2026). https://pith.science/paper/3W6G7KYH

@misc{pith2026250900538,
  author       = {Pith},
  title        = {Pith review of: Identifying Causal Direction via Dense Functional Classes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3W6G7KYH}},
  note         = {Machine review of arXiv:2509.00538}
}
read the original abstract

We address the problem of determining the causal direction between two univariate, continuous-valued variables, X and Y, under the assumption of no hidden confounders. In general, it is not possible to make definitive statements about causality without some assumptions on the underlying model. To distinguish between cause and effect, we propose a bivariate causal score based on the Minimum Description Length (MDL) principle, using functions that possess the density property on a compact real interval. We prove the identifiability of these causal scores under specific conditions. These conditions can be easily tested. Gaussianity of the noise in the causal model equations is not assumed, only that the noise is low. The well-studied class of cubic splines possesses the density property on a compact real interval. We propose LCUBE as an instantiation of the MDL-based causal score utilizing cubic regression splines. LCUBE is an identifiable method that is also interpretable, simple, and very fast. It has only one hyperparameter. Empirical evaluations compared to state-of-the-art methods demonstrate that LCUBE achieves superior precision in terms of AUDRC on the real-world Tuebingen cause-effect pairs dataset. It also shows superior average precision across common 10 benchmark datasets and achieves above average precision on 13 datasets.

Figures

Figures reproduced from arXiv: 2509.00538 by the authors.

Figure 1
Figure 1. AUDRC of four methods The ACC and AUDRC values for CAM, HECI, QCCD, RESIT, IGCI, and LOCI are adopted from [23] ( [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 48 canonical work pages

  1. [30]

    A. Marx, J. Vreeken. Identifiability of cause and effect using regularized regression. 25th ACM SIGKDD KDD. 2019

    work page 2019

  2. [29]

    Kilbertus, G

    N. Kilbertus, G. Parascandolo, B. Sch ¨olkopf. Generalization in anti-causal learning. arXiv:1812.00524 (2018)

    arXiv 2018

  3. [1]

    B ¨uhlmann, J

    P. B ¨uhlmann, J. Peters, J. Ernest. CAM: Causal additive models, high-dimensional order search and penalized regression. Ann. Statist. 42 (6), pp. 2526 - 2556, 2014

    work page 2014

  4. [2]

    Janzing, B

    D. Janzing, B. Sch ¨olkopf. Causal Inference Using the Algorithmic Markov Condition. IEEE Trans on Inform. Tech. 56/10, 5168–94, 2010

    work page 2010

  5. [3]

    Stegle, D

    O. Stegle, D. Janzing, K. Zhang, J.M. Mooij, B. Sch ¨olkopf. Probabilistic latent variable models for distinguishing between cause and effect. Advances in Neurips 23 (2010)

    work page 2010

  6. [4]

    M. Li, P. Vit ´anyi, An Introduction to Kolmogorov Complexity and its Applications. Springer, 2019, vol. 4

    work page 2019

  7. [5]

    Budhathoki, J

    K. Budhathoki, J. Vreeken. MDL for causal inference on discrete data. IEEE ICDM 2017

    work page 2017

  8. [6]

    C. De Boor. A Practical Guide to Splines, V olume 27. Springer-Verlag New York, 1978

    work page 1978

Show all 51 references
  1. [7]

    A. Marx, J. Vreeken. Formally justifying MDL-based inference of cause and effect. In AAAI 2022

    2022

  2. [8]

    P.V . Galkin. The uniqueness of the element of best mean approximation to a continuous function using splines with fixed nodes. Mathematical notes of the Academy of Sciences of the USSR, 15 (1):3–8, 1974

    1974

  3. [9]

    Goudet, D

    O. Goudet, D. Kalainathan, P. Caillou, I. Guyon, D. Lopez-Paz, M. Sebag. Learning functional causal models with generative neural networks. Explainable and Interpr. Models in Comp. Vis. and ML, pp. 39–80, 2018

    2018

  4. [10]

    Rissanen

    J. Rissanen. Modeling by shortest data description. Automatica 14.5 (1978): 465-471

    1978

  5. [11]

    Gr ¨unwald

    P. Gr ¨unwald. The minimum description length principle and reasoning under uncertainty. University of Amsterdam, 1998

    1998

  6. [12]

    Guyon, A

    I. Guyon, A. Statnikov, B.B. Batu. Cause Effect Pairs in Machine Learning. Springer, 2019

    2019

  7. [13]

    Hall, W.W

    C.A. Hall, W.W. Meyer. Optimal error bounds for cubic spline interpolation. Journal of Approximation Theory 16.2 (1976): 105-122

    1976

  8. [14]

    Hlav ´aˇckov´a-Schindler, M

    K. Hlav ´aˇckov´a-Schindler, M. Sanguineti. Bounds on the complexity of neural-network models and comparison with linear methods. Intern. Journal of Adapt. Control and Sign. Process. 17.2 (2003): 179-194

    2003

  9. [15]

    Hlav ´aˇckov´a-Schindler

    K. Hlav ´aˇckov´a-Schindler. Rates of approximation in a feedforward network depend on the types of computational unit. In book: Dealing with Complexity, Springer, pp 205–219, 1998

    1998

  10. [16]

    Kovachki, S

    N. Kovachki, S. Lanthaler, S. Mishra. On universal approximation and error bounds for Fourier neural operators. JMLR 22.290 (2021): 1-76

    2021

  11. [17]

    G. Wahba. Interpolating spline methods for density estimation I. Equi-spaced knots. Annals of Stat. (1975): 30-48

    1975

  12. [18]

    V . Vapnik. The nature of statistical learning theory. Springer Science and Business Media, 2013

    2013

  13. [19]

    Hornik, M

    K. Hornik, M. Stinchcombe, H. White. Multilayer feedforward networks are universal approximators. Neural Networks, 2(5):359-366, 1989

    1989

  14. [20]

    Hoyer, D

    P. Hoyer, D. Janzing, J.M. Mooij, J. Peters, B. Sch ¨olkopf. Nonlinear causal discovery with additive noise models. Advances in Neurips, 21, 2008

    2008

  15. [21]

    Schultheiss, P

    C. Schultheiss, P. B ¨uhlmann. On the pitfalls of Gaussian likelihood scoring for causal discovery. Journal of Causal Inference 11.1 (2023): 20220068

    2023

  16. [22]

    Nowzohour, P

    C. Nowzohour, P. B ¨uhlmann. Score-based causal learning in additive noise models. Statistics. 2016;50(3):471–85

    2016

  17. [23]

    Immer, C

    A. Immer, C. Schultheiss, J.E. V ogt, B. Sch ¨olkopf, P. B¨uhlmann, A. Marx. On the identifiability and estimation of causal location-scale noise models. ICDM, pp. 14316–14332. PMLR, 2023

    2023

  18. [24]

    Kainen, V

    P.C. Kainen, V . K ˚urkov´a. An integral upper bound for neural network approximation. Neural Computation 21.10 (2009): 2970-2989

    2009

  19. [25]

    Janzing, B

    D. Janzing, B. Sch ¨olkopf. Causal inference using the algorithmic Markov condition. IEEE Trans. on Information Theory, 56(10):5168–5194, 2010

    2010

  20. [26]

    Janzing, J.M

    D. Janzing, J.M. Mooij, K. Zhang, J. Lemeire, J. Zscheischler, P. Daniu ˇsis, B. Steudel, B. Sch ¨olkopf. Information-geometric approach to inferring causal directions. Artificial Intelligence, 182:1–31, 2012

    2012

  21. [27]

    T.C.M. Lee. Regression spline smoothing using the minimum description length principle. Statistics and Probability Letters, 48(1):71–82, 2000

    2000

  22. [28]

    A. Marx, J. Vreeken. Telling cause from effect using MDL-based local and global regression. 2017 IEEE ICDM, pp. 307–316. IEEE, 2017

    2017

  23. [31]

    Bl ¨obaum, D

    P. Bl ¨obaum, D. Janzing, T. Washio, S.Shimizu, B. Sch ¨olkopf. Cause-Effect Inference by Comparing Regression Errors. AISTATS 2018

    2018

  24. [32]

    Mooij, J

    J.M. Mooij, J. Peters, D. Janzing, J. Zscheischler, B. Sch ¨olkopf. Distinguishing cause from effect using observational data: methods and benchmarks. JMLR, 17(32):1–102, 2016

    2016

  25. [33]

    J. Pearl. Causality. Cambridge University Press, 2009

    2009

  26. [34]

    Peters, J.M

    J. Peters, J.M. Mooij, D. Janzing, B. Sch ¨olkopf. Identifiability of causal graphs using functional models. arXiv preprint arXiv:1202.3757, 2012

    2012 arXiv

  27. [35]

    Peters, J.M

    J. Peters, J.M. Mooij, D. Janzing, B. Sch ¨olkopf. Causal discovery with continuous additive noise models. JMLR, 2014

    2014

  28. [36]

    Zhang, A

    K. Zhang, A. Hyv ¨arinen. 2009. On the Identifiability of the Post-nonlinear Causal Model. UAI 2009, 647–655

    2009

  29. [37]

    Peters, D

    J. Peters, D. Janzing, B. Sch ¨olkopf. Elements of causal inference: Foundations and learning algorithms. The MIT Press, 2017

    2017

  30. [38]

    Rissanen

    J. Rissanen. Stochastic complexity in statistical inquiry, V ol. 15. World Scientific, 1998

    1998

  31. [39]

    Rissanen

    J. Rissanen. Stochastic complexity. Information and Complexity in Statistical Modeling, pages 57–77, 2007

    2007

  32. [40]

    M.H. Schultz. Spline analysis. Prentice Hall, 1973

    1973

  33. [41]

    Shimizu, P.O

    S. Shimizu, P.O. Hoyer, A. Hyv ¨arinen, A. Kerminen, M. Jordan. A linear non-Gaussian acyclic model for causal discovery. JMLR, 7(10), 2006

    2006

  34. [42]

    Subbotin, N.I

    Y .N. Subbotin, N.I. Chernykh. Order of the best spline approximations of some classes of functions. Math. Notes of the Academy of Sciences of the USSR, 7(1):20–26, 1970

    1970

  35. [43]

    P.V . Galkin. The uniqueness of the element of best mean approximation to a continuous function using splines with fixed nodes. Math. Notes of the Academy of Sciences of the USSR 15.1 (1974): 3-8

    1974

  36. [44]

    Tagasovska, V

    N. Tagasovska, V . Chavez-Demoulin, T. Vatter. Distinguishing cause from effect using quantiles: Bivariate quantile causal discovery. ICML, pp. 9311–9323. PMLR, 2020

    2020

  37. [45]

    Q.-D. Tran, B. Duong, P. Nguyen, T. Nguyen. Robust estimation of causal heteroscedastic noise models. SIAM SDM, pp. 788–796, 2024

    2024

  38. [46]

    S. Xu, O.A. Mian, A. Marx, J. Vreeken. Inferring cause and effect in the presence of heteroscedastic noise. ICML, pp. 24615-24630. 2022. APPENDIX A. Proof of Lemma 1: Under Assumptions 1-3 from Bl ¨obaum et al. in [31], it holds lim α→0 E[( ˜Yα − φ(X))2] E[(X − ψ( ˜Yα))2] ≤ 1 ...

    2022

  39. [47]

    Thus based on this, L(m) ≈ log2 m

    Code L(m): L(m) can be approximated by log m when m is reasonably large, see paper [38], Section 2.2.4. Thus based on this, L(m) ≈ log2 m

  40. [48]

    Since k = {k1,

    Code L(k|m): First we will compute the code L(k|m). Since k = {k1, . . . , km} ⊂X, the sequence of indices k can be specified by the indices of those xi’s where a knot is placed. This set of sorted indices can be compactly specified by their successive differences. Define k0 =...

  41. [49]

    Given m, k, the values b, β can be computed from Eq

    Code for L(b, β|m, k): Now we compute L(b, β|m, k). Given m, k, the values b, β can be computed from Eq. (16) and the result is the conditional maximum likelihood estimates of b and β (or the least square estimates if the assumption of normal errors is not given). Based on [38...

  42. [50]

    AUDRC: The area under the decision rate curve (AUDRC). [23] justifies the selection of AUDRC over AUROC, since it weights correctly identified X → Y pairs in the same way as correctly identified Y → X pairs and thus avoids an arbitrary selection of true positives and true nega...

  43. [51]

    ← or → or ? = undecided) and wq is the weight of the pair, see [32]

    Forced decision: Evaluation of accuracy ACC: Given (X, Y), the method has to decide either X → Y or Y → X; We evaluate the accuracy of these decisions as accuracy = PQ q=1 wqδ ˆdq,dq PQ q=1 wq , where dq is the true causal direction for the q’th T ¨ubingen pair (either ← or →)...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.