Pith. sign in

REVIEW 2 major objections 8 minor 38 references

Sled dogs switch galloping stride patterns every few steps

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · glm-5.2

2026-07-10 03:43 UTC pith:7EGSIQZT

load-bearing objection Solid field pipeline for stride-state discovery; segmentation convention may inflate state count the 2 major comments →

arxiv 2607.08644 v1 pith:7EGSIQZT submitted 2026-07-09 physics.bio-ph

Acceleration-based clustering reveals frequent gait switching in sprint sled dogs

Benjamin Seleb , Saad Bhamla This is my paper
classification physics.bio-ph
keywords sled doggait transitionaccelerometrybiologgingdynamic time warpingunsupervised clusteringgalloplocomotion
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Sprint sled dogs do not settle into a single, stereotyped gallop. By attaching tri-axial accelerometers to the harnesses of ten dogs and analyzing over 20,000 strides with unsupervised clustering—no manual gait labels required—this paper shows that each dog cycles through four to seven recurrent stride states, switching between them far more often than expected. Most dogs dwelled in a given state for only two to four strides before transitioning, a rate dramatically higher than the lead-change frequency documented in galloping horses (roughly every 48 strides). The authors then tested whether these transitions are driven by mechanical perturbations from the tugline connecting each dog to the sled. A logistic model using peak tugline-force timing and magnitude predicted transitions only marginally better than chance (AUC 0.52–0.62). Team-wide coordination was also weak: the Fano factor of population switching activity sat near 1, consistent with largely independent transitions across dogs. The central claim is that sprint sled dog gallop is inherently variable—a repertoire of nearby stride states with frequent, apparently spontaneous switching that is not well explained by local external forcing.

Core claim

Using harness-mounted accelerometers and unsupervised clustering (multivariate dynamic time warping, UMAP embedding, and DBSCAN), the authors identified recurrent stride states in galloping sled dogs and found that dogs switch among them every few strides—far more frequently than lead changes reported in other galloping quadrupeds—and that these transitions are poorly predicted by local tugline-force timing or magnitude and show only weak team-wide coordination.

What carries the argument

The pipeline segments strides using zero-crossings in dorsoventral acceleration, computes pairwise stride dissimilarity via multivariate dynamic time warping (MDTW), embeds the resulting distance matrix into two dimensions with UMAP, and clusters strides with DBSCAN. Each stride receives a discrete state label, and transitions between labels define a binary switch train analyzed with dwell-time statistics, Fano factors, and logistic regression against tugline-force features.

Load-bearing premise

The stride-segmentation method uses an arbitrary feature of the acceleration signal—steep positive zero-crossings in dorsoventral acceleration—as stride boundaries, and this marker may not correspond to the same biomechanical event across all galloping modes, meaning some identified stride states could be artifacts of the segmentation convention rather than genuine locomotor patterns.

What would settle it

If synchronized high-speed video showed that the acceleration-derived stride states do not correspond to distinguishable footfall patterns, or if sub-stride-resolution analysis revealed that the identified states are artifacts of the zero-crossing segmentation convention splitting or merging true gait cycles, the central claim of frequent switching between genuine stride states would be undermined.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If frequent stride-state switching is a general feature of high-speed canine gallop rather than a response to load pulling, comparative studies of tethered and untethered dogs should reveal similar switching rates.
  • The weak predictive power of single-stride tugline-force features suggests that multi-stride force history, neckline tension between adjacent dogs, or internally generated neuromechanical adjustments may be the true drivers of transitions.
  • The acceleration-based clustering pipeline could be applied to other species and field contexts where continuous video is impractical, extending within-gait variation analysis to free-ranging or working animals.
  • If some identified states are segmentation artifacts rather than genuine gait patterns (as the authors acknowledge), sub-stride-resolution analysis and synchronized video validation would clarify how many distinct stride states actually exist.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 8 minor

Summary. This manuscript presents an unsupervised pipeline for identifying recurrent stride states from harness-mounted tri-axial accelerometers in sprint sled dogs. The pipeline combines multivariate dynamic time warping (MDTW), UMAP embedding, and DBSCAN clustering to identify stride states without manual gait labels. Applied to over 20,000 strides from a 10-dog team, the authors find that dogs occupy 4–7 recurrent stride states and switch between them frequently (mean dwell 2–12 strides for most dogs), with weak team-level coordination (Fano factor ~1.17) and weak predictive power of local tugline-force features for transition events (AUC 0.52–0.62). The pipeline is validated against manually labeled video data from one individual (96% cluster purity), and parameter robustness is systematically assessed (Appendix A, median ARI=0.95 across parameter sweeps, median ARI=0.997 across random seeds).

Significance. The paper addresses a genuine methodological gap: continuous video is impractical for field studies of fast-moving sled dog teams, and the authors demonstrate that wearable accelerometry can recover within-gait structure that would otherwise be inaccessible. The validation against labeled video data, the systematic parameter sweep in Appendix A, and the seed reproducibility analysis are commendable methodological strengths. The finding that tugline-force peak timing and magnitude weakly predict switching is a useful negative result that narrows the space of plausible drivers. The code and data availability statements are appropriate.

major comments (2)
  1. §III.A.1, Eq. (4): The cluster purity metric validates that each of the four manually labeled gait patterns maps to a dominant cluster, but it does not assess the 2–3 extra clusters per dog that exceed the four canonical gallop patterns. These extra states are central to the claim that gallop 'occupies a variable set of nearby stride states' (Abstract, §IV, §V), yet they remain unvalidated. The paper acknowledges in §IV that several identified states 'lack strong self-transitions and appear primarily as intermediate states,' which is consistent with the segmentation artifact concern raised below, but no quantitative analysis of these extra clusters is provided. The authors should either (a) explicitly characterize the extra clusters (e.g., are they transitional/mixed strides?) or (b) demonstrate that the core finding of frequent switching holds even when intermediate states are merged or
  2. §II.C and §IV: The segmentation convention uses steep positive zero-crossings in dorsoventral acceleration (a_z) as stride boundaries, which the paper acknowledges is 'an arbitrary signal feature closest to the touchdown of the trailing forelimb' (§IV). In rotary vs. transverse gallop, limb contact timing relative to the a_z waveform differs, so the zero-crossing marker may correspond to different biomechanical events across gait patterns. This could phase-shift strides from different gaits in MDTW comparison, inflating inter-gait distances and fragmenting transition strides into apparent intermediate states. The parameter robustness analysis (Appendix A) addresses sensitivity to UMAP/DBSCAN parameters but does not address sensitivity to the segmentation convention itself. A sensitivity analysis using an alternative segmentation marker (or, where available, hind-limb touchdown from the 2
minor comments (8)
  1. §III.C: The text states 'a total of n events stride-boundary events were retained for analysis of each individual.' The variable 'n_events' appears to be a placeholder that was not replaced with the actual value, though Table II does report per-dog n_events values.
  2. §II.B: The Butterworth high-pass filter cutoff is stated as 0.5 Hz, but no justification is provided for this particular cutoff. Given that stride frequencies in galloping dogs are likely 2–4 Hz, a brief note on why 0.5 Hz was chosen would help readers assess whether stride-relevant signal components are preserved.
  3. §II.D: The UMAP parameters (n_neighbors=12, min_dist=0.22) and DBSCAN parameters (eps=0.6, min_samples=10) are stated, but the rationale for these specific values is not provided. While Appendix A demonstrates robustness, a brief note on how these values were selected would improve reproducibility.
  4. §III.B.2, Eq. (6): The population switching activity R(t) is defined as the sum of switch events across dogs per second, but the mean value R̄=7.54 is reported without units or context (is this switches per second across 10 dogs?). Clarifying this would help interpretation.
  5. §IV: The paper notes that 'joint embeddings were explored' but that 'individual variation dominated the embedding structure.' A brief description of what was attempted (e.g., concatenated embeddings, Procrustes alignment) would help readers understand the limitation and potential solutions.
  6. Figure 5D: The phase discrepancy Δϕ_Fk is defined in the caption but not in the main text where the logistic model is described (§III.C, Eq. 8). The model uses ϕ_Fk directly, while the figure shows Δϕ_Fk. This inconsistency could confuse readers.
  7. Table II: The column header 'n events' should be formatted consistently with other tables (e.g., 'n_events' or 'N_events').
  8. §II.C: The minimum crossing separation of 0.25 s (4 Hz maximum frequency) is stated, but it would be useful to note the typical stride frequency range observed in the dataset to confirm this threshold is appropriate.

Circularity Check

0 steps flagged

No significant circularity; self-citation provides validation data but is not load-bearing for central claims

full rationale

The paper's derivation chain is self-contained against external data. The central claims rest on three independent pipelines: (1) acceleration data from physical sensors → stride segmentation (zero-crossings in az) → MDTW pairwise distances → UMAP embedding → DBSCAN clustering → stride-state assignments → switching statistics; (2) tugline-force data from load cells → logistic regression → weak predictive power (AUC 0.52–0.62); (3) GPS-derived speed/heading for context. None of these steps reduce to their inputs by construction. The cluster purity validation (Eq. 4) checks unsupervised clusters against manually labeled video data from Ref [2] (Ding, Seleb, Bhamla, et al.), which is a self-citation, but it serves as an independent external check—manual video annotations are not defined by the present paper's acceleration pipeline. The self-citation is not load-bearing: the central claims (frequent switching, weak force prediction, limited team coordination) are derived from sensor measurements and standard unsupervised methods, not from Ref [2]'s conclusions. The logistic regression's weak performance is an honest negative result, not a fitted-then-predicted circularity. The skeptic's concern about segmentation artifacts creating spurious intermediate states is a legitimate correctness/methodology risk, but it is not circularity—the segmentation convention is a methodological choice that could be wrong, not a definition that forces the conclusion. Score 1 reflects the minor self-citation for validation data without independent load-bearing function.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 0 invented entities

The paper introduces no new physical entities, forces, or conserved quantities. The stride states are data-driven clusters, not postulated mechanisms. The free parameters are all standard algorithmic hyperparameters for UMAP and DBSCAN, with sensitivity analysis provided. The axioms are domain assumptions about the relationship between acceleration signals and biomechanics, most of which are supported by the validation analysis.

free parameters (7)
  • UMAP n_neighbors = 12
    Chosen by the authors; sensitivity assessed in Appendix A showing results are robust to {6, 12, 20}.
  • UMAP min_dist = 0.22
    Chosen by the authors; sensitivity assessed in Appendix A.
  • DBSCAN eps = 0.6
    Chosen by the authors; sits at transition between over-fragmentation and merging regimes per Appendix A.
  • DBSCAN min_samples = 10
    Chosen by the authors; sensitivity assessed in Appendix A.
  • Butterworth high-pass cutoff = 0.5 Hz
    Standard choice for removing gravity component from accelerometry; not swept.
  • Minimum crossing separation = 0.25 s
    Corresponds to max frequency of 4 Hz; prevents false detections from small oscillations.
  • Logistic regression coefficients (beta_0, beta_cos, beta_sin, beta_F) = per-dog, reported in Table II
    Fitted by maximum likelihood to predict switching from force features; weak predictive power (AUC 0.52-0.62).
axioms (5)
  • domain assumption Steep positive zero-crossings in dorsoventral acceleration correspond to a consistent biomechanical event usable as stride boundary
    Invoked in §II.C for stride segmentation. The paper acknowledges this marker is 'an arbitrary signal feature closest to the touchdown of the trailing forelimb' and that it differs from the hind-limb touchdown convention used in labeled data.
  • domain assumption MDTW distance between tri-axial acceleration subsequences captures meaningful gait dissimilarity
    Invoked in §II.D. Supported by the validation showing 96% cluster purity against labeled gait patterns in Individual 1.
  • domain assumption UMAP embedding preserves stride-state structure present in the MDTW distance matrix
    Invoked in §II.D. Standard assumption for UMAP; supported by parameter robustness analysis showing consistent cluster assignments across parameter sweeps.
  • domain assumption DBSCAN density-based clustering identifies recurrent stride states rather than artifacts of the embedding
    Invoked in §II.D. Supported by validation against labeled data and by the observation that identified states form distinct braid-like trajectories in acceleration space.
  • ad hoc to paper Single-stride peak force timing and magnitude are sufficient features to test the mechanical-forcing hypothesis
    Invoked in §III.C. The authors acknowledge this is insufficient and that multi-stride force history, speed fluctuations, and neckline tension were not assessed.

pith-pipeline@v1.1.0-glm · 15471 in / 3134 out tokens · 364658 ms · 2026-07-10T03:43:35.712879+00:00 · methodology

0 comments
read the original abstract

Continuous video is difficult to obtain during field studies of sprint sled dogs, limiting analysis of stride-to-stride variation during load-pulling gallop. We developed an acceleration-based pipeline to identify recurrent stride states from harness-mounted tri-axial accelerometers without manual gait labels. Using multivariate dynamic time warping, manifold embedding, and density-based clustering, we analyzed more than 20,000 strides from a 10-dog team and identified recurrent, dog-specific stride states. In one previously annotated individual, acceleration-derived states were broadly consistent with manually labeled gallop patterns. Across dogs, transitions between stride states were frequent, with substantial inter-individual variation and limited evidence of strong team-level coordination. A simple logistic model based on local tugline-force timing and magnitude had weak predictive power for transition events. These results suggest that sprint sled dog gallop occupies a variable set of nearby stride states and that local tugline-force fluctuations alone do not explain the observed switching.

Figures

Figures reproduced from arXiv: 2607.08644 by Benjamin Seleb, Saad Bhamla.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages · 1 internal anchor

  1. [1]

    Stride syllables and switching dynamics The cluster assignment sequencesk can be interpreted as a series of discrete behavioral syllables [20], providing a compact representation of real-time stride-state varia- tion. Because each dog’s embedding is stochastic and unique, stride clusters are not directly comparable across individuals (e.g., state 0 for on...

    work page 2000

  2. [2]

    Validation with labeled data To verify that the unsupervised clusters correspond to identifiable gait patterns, we compared cluster assign- mentss k with the manually labeled gait identitiesg k from Individual 1. For each gait labelg, the dominant clusterc ∗(g)was identified, and cluster purity was com- puted as Pg = 1 ng X k:gk=g 1[ck =c ∗(g)],(4) wheren...

    work page

  3. [3]

    4.Stride-space representation for a single dog.(A) Two-dimensional UMAP embedding of the per-stride MDTW distance matrix

    Individual dwell statistics For each dog, contiguous non-switch segments inb definedwell periodsof lengthL i, corresponding to the 5 FIG. 4.Stride-space representation for a single dog.(A) Two-dimensional UMAP embedding of the per-stride MDTW distance matrix. Each point represents a stride, colored by cluster assignments; black points denote outliers (top...

    work page

  4. [4]

    Collective switching dynamics We next examined whether stride-state transitions oc- curred in a coordinated manner across the 10-dog team (Fig. 5A,B). Each dog’s binary switch train was treated as a point process, and switching events were aggregated across dogs using 1s time bins. The instantaneous pop- ulation switching activity was defined as R(t) = 1 ...

    work page

  5. [5]

    Ar- eas under the ROC curve ranged from 0.52–0.62, indicat- ing discrimination only slightly above chance (AUC = 0.5)

    Model predictions Across all dogs, the logistic models showed only weak predictive power for stride-state switching (Table II). Ar- eas under the ROC curve ranged from 0.52–0.62, indicat- ing discrimination only slightly above chance (AUC = 0.5). Point-biserial correlations between switching and force peak phase were small (|rpb|<0.15), and the fitted cir...

    work page 2047

  6. [6]

    Description of breed ancestry and genetic health traits in arctic sled dog breeds.Canine medicine and genetics, 8:1–13, 2021

    Joseph A Thorsrud and Heather J Huson. Description of breed ancestry and genetic health traits in arctic sled dog breeds.Canine medicine and genetics, 8:1–13, 2021

    work page 2021

  7. [7]

    Gait Transitions in Load-Pulling Quadrupeds: Insights from Sled Dogs and a Minimal SLIP Model

    Jiayu Ding, Benjamin Seleb, Heather J Huson, Saad Bhamla, and Zhenyu Gan. Gait transitions in load- pulling quadrupeds: Insights from sled dogs and a mini- mal slip model.arXiv preprint arXiv:2507.14727, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  8. [8]

    Gait and the en- ergetics of locomotion in horses.Nature, 292(5820):239– 240, July 1981

    Donald F Hoyt and C Richard Taylor. Gait and the en- ergetics of locomotion in horses.Nature, 292(5820):239– 240, July 1981

    work page 1981

  9. [9]

    A mechanical trigger for the trot-gallop transition in horses.Science, 253(5017):306–308, July 1991

    Claire T Farley and C Richard Taylor. A mechanical trigger for the trot-gallop transition in horses.Science, 253(5017):306–308, July 1991

    work page 1991

  10. [10]

    A Vahedipour, O Haji Maghsoudi, S Wilshin, P Shamble, B Robertson, and A Spence. Uncovering the structure of the mouse gait controller: Mice respond to substrate perturbations with adaptations in gait on a continuum between trot and bound.Journal of biomechanics, 78:77– 86, 2018

    work page 2018

  11. [11]

    Dog galloping on rough terrain exhibits similar limb co- ordinationpatternsandgaitvariabilitytothatonflatter- rain.Bioinspiration & Biomimetics, 16(1):015001, 2020

    Simon Wilshin, Michelle A Reeve, and Andrew J Spence. Dog galloping on rough terrain exhibits similar limb co- ordinationpatternsandgaitvariabilitytothatonflatter- rain.Bioinspiration & Biomimetics, 16(1):015001, 2020

    work page 2020

  12. [12]

    Quantifying the movement, behaviour and environmen- tal context of group-living animals using drones and com- puter vision.Journal of Animal Ecology, 2023

    Benjamin Koger, Adwait Deshpande, Jeffrey T Kerby, Jacob M Graving, Blair R Costelloe, and Iain D Couzin. Quantifying the movement, behaviour and environmen- tal context of group-living animals using drones and com- puter vision.Journal of Animal Ecology, 2023

    work page 2023

  13. [13]

    Anal- ysis of agile canine gait characteristics using accelerome- try.Sensors, 2019

    Hasti Hayati, Fatemeh Mahdavi, and David Eager. Anal- ysis of agile canine gait characteristics using accelerome- try.Sensors, 2019

    work page 2019

  14. [14]

    An exploratory clustering approach for extracting stride parameters from tracking collars on free-ranging wild animals.Journal of Experimental Biol- ogy, 220(3):341–346, 2017

    Oliver P Dewhirst, Kyle Roskilly, Tatjana Y Hubel, Neil R Jordan, Krystyna A Golabek, J Weldon McNutt, and Alan M Wilson. An exploratory clustering approach for extracting stride parameters from tracking collars on free-ranging wild animals.Journal of Experimental Biol- ogy, 220(3):341–346, 2017

    work page 2017

  15. [15]

    Identification of behaviours from accelerometer data in a wild social primate.Animal Biotelemetry, 5:1–11, 2017

    Gaelle Fehlmann, M Justin O’Riain, Phil W Hopkins, Jack O’Sullivan, Mark D Holton, Emily LC Shepard, and Andrew J King. Identification of behaviours from accelerometer data in a wild social primate.Animal Biotelemetry, 5:1–11, 2017

    work page 2017

  16. [16]

    Observing the unwatchable through acceleration logging of animal behavior.Animal Biotelemetry, 1:1–16, 2013

    Danielle D Brown, Roland Kays, Martin Wikelski, Rory Wilson, and A Peter Klimley. Observing the unwatchable through acceleration logging of animal behavior.Animal Biotelemetry, 1:1–16, 2013

    work page 2013

  17. [17]

    Perspectives in machine learn- ing for wildlife conservation.Nature communications, 13(1):792, 2022

    Devis Tuia, Benjamin Kellenberger, Sara Beery, Blair R Costelloe, Silvia Zuffi, Benjamin Risse, Alexander Mathis, Mackenzie W Mathis, Frank van Langevelde, Tilo Burghardt, et al. Perspectives in machine learn- ing for wildlife conservation.Nature communications, 13(1):792, 2022

    work page 2022

  18. [18]

    Re-wilding collective behaviour: An ecological perspective.Trends in Ecology and Evolution, 2018

    Andrew J King, Gaelle Fehlmann, Dora Biro, Ashley J W Ward, and Ines Fürtbauer. Re-wilding collective behaviour: An ecological perspective.Trends in Ecology and Evolution, 2018

    work page 2018

  19. [19]

    Ethology as a physical science.bioRxiv, 2018

    André Ex Brown and Benjamin L de Bivort. Ethology as a physical science.bioRxiv, 2018

    work page 2018

  20. [20]

    Emerging technologies for behavioral research in changing environments.Trends in Ecology & Evolution, 38(4):346–354, 2023

    Iain D Couzin and Conor Heins. Emerging technologies for behavioral research in changing environments.Trends in Ecology & Evolution, 38(4):346–354, 2023

    work page 2023

  21. [21]

    Generalizing dtw to the multi-dimensional case requires an adaptive approach

    Mohammad Shokoohi-Yekta, Bing Hu, Hongxia Jin, Jun Wang, and Eamonn Keogh. Generalizing dtw to the multi-dimensional case requires an adaptive approach. Data mining and knowledge discovery, 31:1–31, 2017

    work page 2017

  22. [22]

    De- constructing hunting behavior reveals a tightly coupled stimulus-response loop.Current Biology, 30(1):54–69, 2020

    Duncan S Mearns, Joseph C Donovan, António M Fer- nandes, Julia L Semmelhack, and Herwig Baier. De- constructing hunting behavior reveals a tightly coupled stimulus-response loop.Current Biology, 30(1):54–69, 2020

    work page 2020

  23. [23]

    Uniform manifold approximation and projection for dimension reduction

    L McInnes, J Healy, and J Melville. Uniform manifold approximation and projection for dimension reduction. Preprint at arXiv,(UMAP, 2020)

    work page 2020

  24. [24]

    A density-based algorithm for discovering clus- ters in large spatial databases with noise

    Martin Ester, Hans-Peter Kriegel, Jörg Sander, Xiaowei Xu, et al. A density-based algorithm for discovering clus- ters in large spatial databases with noise. Inkdd, vol- ume 96, pages 226–231, 1996

    work page 1996

  25. [25]

    Mapping sub-second structure in mouse behavior.Neuron, 88(6):1121–1135, 2015

    Alexander B Wiltschko, Matthew J Johnson, Giuliano Iurilli, Ralph E Peterson, Jesse M Katon, Stan L Pashkovski, Victoria E Abraira, Ryan P Adams, and Sandeep Robert Datta. Mapping sub-second structure in mouse behavior.Neuron, 88(6):1121–1135, 2015

    work page 2015

  26. [26]

    Measuring multi- ple spike train synchrony.Journal of neuroscience meth- ods, 183(2):287–299, 2009

    Thomas Kreuz, Daniel Chicharro, Ralph G Andrzejak, Julie S Haas, and Henry DI Abarbanel. Measuring multi- ple spike train synchrony.Journal of neuroscience meth- ods, 183(2):287–299, 2009

    work page 2009

  27. [27]

    Fano factor: a potentially useful information.Frontiers in computational neuroscience, 14:569049, 2020

    Kamil Rajdl, Petr Lansky, and Lubomir Kostal. Fano factor: a potentially useful information.Frontiers in computational neuroscience, 14:569049, 2020

    work page 2020

  28. [28]

    Scikit-learn: Machine learning 12 in python.the Journal of machine Learning research, 12:2825–2830, 2011

    Fabian Pedregosa, Gaël Varoquaux, Alexandre Gram- fort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vin- cent Dubourg, et al. Scikit-learn: Machine learning 12 in python.the Journal of machine Learning research, 12:2825–2830, 2011

    work page 2011

  29. [29]

    An introduction to roc analysis.Pattern recognition letters, 27(8):861–874, 2006

    Tom Fawcett. An introduction to roc analysis.Pattern recognition letters, 27(8):861–874, 2006

    work page 2006

  30. [30]

    Point biserial correlation.Wiley Stat- sRef: Statistics Reference Online, 2014

    Diana Kornbrot. Point biserial correlation.Wiley Stat- sRef: Statistics Reference Online, 2014

    work page 2014

  31. [31]

    Motions of the running cheetah and horse.Journal of Mammalogy, 40(4):481–495, 1959

    Milton Hildebrand. Motions of the running cheetah and horse.Journal of Mammalogy, 40(4):481–495, 1959

    work page 1959

  32. [32]

    Ground forces applied by galloping dogs.Journal of Experimental Bi- ology, 210(2):208–216, 2007

    Rebecca M Walter and David R Carrier. Ground forces applied by galloping dogs.Journal of Experimental Bi- ology, 210(2):208–216, 2007

    work page 2007

  33. [33]

    A critical characteristic in the transverse galloping pattern.Applied Bionics and Biomechanics, 2015(1):631354, 2015

    Xiaohui Wei, Yongjun Long, Chunlei Wang, and Shi- gang Wang. A critical characteristic in the transverse galloping pattern.Applied Bionics and Biomechanics, 2015(1):631354, 2015

    work page 2015

  34. [34]

    Generalized procrustes analysis.Psy- chometrika, 40(1):33–51, 1975

    John C Gower. Generalized procrustes analysis.Psy- chometrika, 40(1):33–51, 1975

    work page 1975

  35. [35]

    Analysis of asymmetrical gaits.Jour- nal of Mammalogy, 58(2):131–156, May 1977

    Milton Hildebrand. Analysis of asymmetrical gaits.Jour- nal of Mammalogy, 58(2):131–156, May 1977

    work page 1977

  36. [36]

    Thequadrupedalgaitsofvertebrates

    MiltonHildebrand. Thequadrupedalgaitsofvertebrates. BioScience, 39(11):766–775, December 1989

    work page 1989

  37. [37]

    Limb preference in the gallop of dogs and the half-bound of pikas on flat ground.Laterality, 13(4):310–319, 2008

    Rémi Hackert, LD Maes, Marc Herbin, PA Libourel, and A Abourachid. Limb preference in the gallop of dogs and the half-bound of pikas on flat ground.Laterality, 13(4):310–319, 2008

    work page 2008

  38. [38]

    Optimal interval for pe- riodical lead limb changes during straight gallop in race horses.Congresso Nazionale della Società Italiana di Fi- siologia, 2012

    CM Biancardi and AE Minetti. Optimal interval for pe- riodical lead limb changes during straight gallop in race horses.Congresso Nazionale della Società Italiana di Fi- siologia, 2012

    work page 2012