Reconstruction of paleoclimate niches using phylogenetic comparative methods and projection reconstructed niches onto paleoclimate maps. The user can specify various models of trait evolution or estimate the best fit model, include fossils, use one or multiple phylogenies for inference, and make animations of shifting suitable habitat through time. This model was first used in Lawing and Polly (2011), and further implemented in Lawing et al (2016) and Rivera et al (2020). Lawing and Polly (2011) <doi:10.1371/journal.pone.0028554> "Pleistocene climate, phylogeny and climate envelope models: An integrative approach to better understand species response to climate change" Lawing et al (2016) <doi:10.1086/687202> "Including fossils in phylogenetic climate reconstructions: A deep time perspective on the climatic niche evolution and diversification of spiny lizards (Sceloporus)" Rivera et al (2020) <doi:10.1111/jbi.13915> "Reconstructing historical shifts in suitable habitat of Sceloporus lineages using phylogenetic niche modelling.".

Time series area-level models for small area estimation. The package supplements the functionality of the sae package. Specifically, it includes EBLUP fitting of the Rao-Yu model in the original form without a spatial component. The package also offers a modified ("dynamic") version of the Rao-Yu model, replacing the assumption of stationarity. Both univariate and multivariate applications are supported. Of particular note is the allowance for covariance of the area-level sample estimates over time, as encountered in rotating panel designs such as the U.S. National Crime Victimization Survey or present in a time-series of 5-year estimates from the American Community Survey. Key references to the methods include J.N.K. Rao and I. Molina (2015, ISBN:9781118735787), J.N.K. Rao and M. Yu (1994) <doi:10.2307/3315407>, and R.E. Fay and R.A. Herriot (1979) <doi:10.1080/01621459.1979.10482505>.

Machine coded genetic algorithm (MCGA) is a fast tool for real-valued optimization problems. It uses the byte representation of variables rather than real-values. It performs the classical crossover operations (uniform) on these byte representations. Mutation operator is also similar to classical mutation operator, which is to say, it changes a randomly selected byte value of a chromosome by +1 or -1 with probability 1/2. In MCGAs there is no need for encoding-decoding process and the classical operators are directly applicable on real-values. It is fast and can handle a wide range of a search space with high precision. Using a 256-unary alphabet is the main disadvantage of this algorithm but a moderate size population is convenient for many problems. Package also includes multi_mcga function for multi objective optimization problems. This function sorts the chromosomes using their ranks calculated from the non-dominated sorting algorithm.

Calculate the probability density functions (PDFs) for two threshold evidence accumulation models (EAMs). These are defined using the following Stochastic Differential Equation (SDE), dx(t) = v(x(t),t)*dt+D(x(t),t)*dW, where x(t) is the accumulated evidence at time t, v(x(t),t) is the drift rate, D(x(t),t) is the noise scale, and W is the standard Wiener process. The boundary conditions of this process are the upper and lower decision thresholds, represented by b_u(t) and b_l(t), respectively. Upper threshold b_u(t) > 0, while lower threshold b_l(t) < 0. The initial condition of this process x(0) = z where b_l(t) < z < b_u(t). We represent this as the relative start point w = z/(b_u(0)-b_l(0)), defined as a ratio of the initial threshold location. This package generates the PDF using the same approach as the python package it is based upon, PyBEAM by Murrow and Holmes (2023) <doi:10.3758/s13428-023-02162-w>. First, it converts the SDE model into the forwards Fokker-Planck equation dp(x,t)/dt = d(v(x,t)*p(x,t))/dt-0.5*d^2(D(x,t)^2*p(x,t))/dx^2, then solves this equation using the Crank-Nicolson method to determine p(x,t). Finally, it calculates the flux at the decision thresholds, f_i(t) = 0.5*d(D(x,t)^2*p(x,t))/dx evaluated at x = b_i(t), where i is the relevant decision threshold, either upper (i = u) or lower (i = l). The flux at each thresholds f_i(t) is the PDF for each threshold, specifically its PDF. We discuss further details of this approach in this package and PyBEAM publications. Additionally, one can calculate the cumulative distribution functions of and sampling from the EAMs.

Generate systems of ordinary differential equations (ODE) and integrate them, using a domain specific language (DSL). The DSL uses R's syntax, but compiles to C in order to efficiently solve the system. A solver is not provided, but instead interfaces to the packages deSolve and dde are generated. With these, while solving the differential equations, no allocations are done and the calculations remain entirely in compiled code. Alternatively, a model can be transpiled to R for use in contexts where a C compiler is not present. After compilation, models can be inspected to return information about parameters and outputs, or intermediate values after calculations. odin is not targeted at any particular domain and is suitable for any system that can be expressed primarily as mathematical expressions. Additional support is provided for working with delays (delay differential equations, DDE), using interpolated functions during interpolation, and for integrating quantities that represent arrays.

Estimate average treatment effects (ATEs) in stratified randomized experiments. `sreg` supports a wide range of stratification designs, including matched pairs, n-tuple designs, and larger strata with many units â possibly of unequal size across strata. sreg is designed to accommodate scenarios with multiple treatments and cluster-level treatment assignments, and accommodates optimal linear covariate adjustment based on baseline observable characteristics. sreg computes estimators and standard errors based on Bugni, Canay, Shaikh (2018) <doi:10.1080/01621459.2017.1375934>; Bugni, Canay, Shaikh, Tabord-Meehan (2024+) <doi:10.48550/arXiv.2204.08356>; Jiang, Linton, Tang, Zhang (2023+) <doi:10.48550/arXiv.2201.13004>; Bai, Jiang, Romano, Shaikh, and Zhang (2024) <doi:10.1016/j.jeconom.2024.105740>; Bai (2022) <doi:10.1257/aer.20201856>; Bai, Romano, and Shaikh (2022) <doi:10.1080/01621459.2021.1883437>; Liu (2024+) <doi:10.48550/arXiv.2301.09016>; and Cytrynbaum (2024) <doi:10.3982/QE2475>.

Facilitates nonresponse bias analysis (NRBA) for survey data. Such data may arise from a complex sampling design with features such as stratification, clustering, or unequal probabilities of selection. Multiple types of analyses may be conducted: comparisons of response rates across subgroups; comparisons of estimates before and after weighting adjustments; comparisons of sample-based estimates to external population totals; tests of systematic differences in covariate means between respondents and full samples; tests of independence between response status and covariates; and modeling of outcomes and response status as a function of covariates. Extensive documentation and references are provided for each type of analysis. Krenzke, Van de Kerckhove, and Mohadjer (2005) <http://www.asasrms.org/Proceedings/y2005/files/JSM2005-000572.pdf> and Lohr and Riddles (2016) <https://www150.statcan.gc.ca/n1/en/pub/12-001-x/2016002/article/14677-eng.pdf?st=q7PyNsGR> provide an overview of the methods implemented in this package.

In practice, we will encounter problems where the longitudinal performance of processes needs to be monitored over time. Dynamic screening systems (DySS) are methods that aim to identify and give signals to processes with poor performance as early as possible. This package is designed to implement dynamic screening systems and the related methods. References: Qiu, P. and Xiang, D. (2014) <doi:10.1080/00401706.2013.822423>; Qiu, P. and Xiang, D. (2015) <doi:10.1002/sim.6477>; Li, J. and Qiu, P. (2016) <doi:10.1080/0740817X.2016.1146423>; Li, J. and Qiu, P. (2017) <doi:10.1002/qre.2160>; You, L. and Qiu, P. (2019) <doi:10.1080/00949655.2018.1552273>; Qiu, P., Xia, Z., and You, L. (2020) <doi:10.1080/00401706.2019.1604434>; You, L., Qiu, A., Huang, B., and Qiu, P. (2020) <doi:10.1002/bimj.201900127>; You, L. and Qiu, P. (2021) <doi:10.1080/00224065.2020.1767006>.

Imbalanced domain learning has almost exclusively focused on solving classification tasks, where the objective is to predict cases labelled with a rare class accurately. Such a well-defined approach for regression tasks lacked due to two main factors. First, standard regression tasks assume that each value is equally important to the user. Second, standard evaluation metrics focus on assessing the performance of the model on the most common cases. This package contains methods to tackle imbalanced domain learning problems in regression tasks, where the objective is to predict extreme (rare) values. The methods contained in this package are: 1) an automatic and non-parametric method to obtain such relevance functions; 2) visualisation tools; 3) suite of evaluation measures for optimisation/validation processes; 4) the squared-error relevance area measure, an evaluation metric tailored for imbalanced regression tasks. More information can be found in Ribeiro and Moniz (2020) <doi:10.1007/s10994-020-05900-9>.

This package provides a set of model-assisted survey estimators and corresponding variance estimators for single stage, unequal probability, without replacement sampling designs. All of the estimators can be written as a generalized regression estimator with the Horvitz-Thompson, ratio, post-stratified, and regression estimators summarized by Sarndal et al. (1992, ISBN:978-0-387-40620-6). Two of the estimators employ a statistical learning model as the assisting model: the elastic net regression estimator, which is an extension of the lasso regression estimator given by McConville et al. (2017) <doi:10.1093/jssam/smw041>, and the regression tree estimator described in McConville and Toth (2017) <arXiv:1712.05708>. The variance estimators which approximate the joint inclusion probabilities can be found in Berger and Tille (2009) <doi:10.1016/S0169-7161(08)00002-3> and the bootstrap variance estimator is presented in Mashreghi et al. (2016) <doi:10.1214/16-SS113>.

The general principle relies on calculating the cumulative signal of nascent RNA sequencing over the gene body of any given gene or transcription unit. tepr can identify transcription attenuation sites by comparing profile to a null model which assumes uniform read density over the entirety of the transcription unit. It can also identify increased or diminished transcription attenuation by comparing two conditions. Besides rigorous statistical testing and high sensitivity, a major feature of tepr is its ability to provide the elongation pattern of each individual gene, including the position of the main attenuation point when such a phenomenon occurs. Using tepr', users can visualize and refine genome-wide aggregated analyses of elongation patterns to robustly identify effects specific to subsets of genes. These metrics are suitable for internal comparisons (between genes in each condition) and for studying elongation of the same gene in different conditions or comparing it to a perfect theoretical uniform elongation.

Spline regression, generalized additive models and component-wise gradient boosting utilizing geometrically designed (GeD) splines. GeDS regression is a non-parametric method inspired by geometric principles, for fitting spline regression models with variable knots in one or two independent variables. It efficiently estimates the number of knots and their positions, as well as the spline order, assuming the response variable follows a distribution from the exponential family. GeDS models integrate the broader category of generalized (non-)linear models, offering a flexible approach to model complex relationships. A description of the method can be found in Kaishev et al. (2016) <doi:10.1007/s00180-015-0621-7> and Dimitrova et al. (2023) <doi:10.1016/j.amc.2022.127493>. Further extending its capabilities, GeDS's implementation includes generalized additive models (GAM) and functional gradient boosting (FGB), enabling versatile multivariate predictor modeling, as discussed in the forthcoming work of Dimitrova et al. (2025).

This package provides a unified framework for symbolic regression (SR) and multi-view symbolic regression (MvSR) designed for complex, nonlinear systems, with particular applicability to ecological datasets. The package implements a four-stage workflow: data subset generation, functional form discovery, numerical parameter optimization, and multi-objective evaluation. It provides a high-level formula-style interface that abstracts and extends multiple discovery engines: genetic programming (via PySR), Reinforcement Learning with Monte Carlo Tree Search (via RSRM), and exhaustive generalized linear model search. leaf extends these methods by enabling multi-view discovery, where functional structures are shared across groups while parameters are fitted locally, and by supporting the enforcement of domain-specific constraints, such as sign consistency across groups. The framework automatically handles data normalization, link functions, and back-transformation, ensuring that discovered symbolic equations remain interpretable and valid on the original data scale. Implements methods following ongoing work by the authors (2026, in preparation).

Provide the core functionality to transform longitudinal data to complex-time (kime) data using analytic and numerical techniques, visualize the original time-series and reconstructed kime-surfaces, perform model based (e.g., tensor-linear regression) and model-free classification and clustering methods in the book Dinov, ID and Velev, MV. (2021) "Data Science: Time Complexity, Inferential Uncertainty, and Spacekime Analytics", De Gruyter STEM Series, ISBN 978-3-11-069780-3. <https://www.degruyter.com/view/title/576646>. The package includes 18 core functions which can be separated into three groups. 1) draw longitudinal data, such as Functional magnetic resonance imaging(fMRI) time-series, and forecast or transform the time-series data. 2) simulate real-valued time-series data, e.g., fMRI time-courses, detect the activated areas, report the corresponding p-values, and visualize the p-values in the 3D brain space. 3) Laplace transform and kimesurface reconstructions of the fMRI data.

Dynamic CUR (dCUR) boosts the CUR decomposition (Mahoney MW., Drineas P. (2009) <doi:10.1073/pnas.0803205106>) varying the k, the number of columns and rows used, and its final purposes to help find the stage, which minimizes the relative error to reduce matrix dimension. The goal of CUR Decomposition is to give a better interpretation of the matrix decomposition employing proper variable selection in the data matrix, in a way that yields a simplified structure. Its origins come from analysis in genetics. The goal of this package is to show an alternative to variable selection (columns) or individuals (rows). The idea proposed consists of adjusting the probability distributions to the leverage scores and selecting the best columns and rows that minimize the reconstruction error of the matrix approximation ||A-CUR||. It also includes a method that recalibrates the relative importance of the leverage scores according to an external variable of the user's interest.

GPU'/CPU Benchmarking on Debian-package based systems This package benchmarks performance of a few standard linear algebra operations (such as a matrix product and QR, SVD and LU decompositions) across a number of different BLAS libraries as well as a GPU implementation. To do so, it takes advantage of the ability to plug and play different BLAS implementations easily on a Debian and/or Ubuntu system. The current version supports - Reference BLAS ('refblas') which are un-accelerated as a baseline - Atlas which are tuned but typically configure single-threaded - Atlas39 which are tuned and configured for multi-threaded mode - Goto Blas which are accelerated and multi-threaded - Intel MKL which is a commercial accelerated and multithreaded version. As for GPU computing, we use the CRAN package - gputools For Goto Blas', the gotoblas2-helper script from the ISM in Tokyo can be used. For Intel MKL we use the Revolution R packages from Ubuntu 9.10.

Helper functions to implement univariate and bivariate latent change score models in R using the lavaan package. For details about Latent Change Score Modeling (LCSM) see McArdle (2009) <doi:10.1146/annurev.psych.60.110707.163612> and Grimm, An, McArdle, Zonderman and Resnick (2012) <doi:10.1080/10705511.2012.659627>. The package automatically generates lavaan syntax for different model specifications and varying timepoints. The lavaan syntax generated by this package can be returned and further specifications can be added manually. Longitudinal plots as well as simplified path diagrams can be created to visualise data and model specifications. Estimated model parameters and fit statistics can be extracted as data frames. Data for different univariate and bivariate LCSM can be simulated by specifying estimates for model parameters to explore their effects. This package combines the strengths of other R packages like lavaan', broom', and semPlot by generating lavaan syntax that helps these packages work together.

Genomic coordinates of CTCF binding sites, with strand orientation (directionality of binding). Position weight matrices (PWMs) from JASPAR, HOCOMOCO, CIS-BP, CTCFBSDB, SwissRegulon, Jolma 2013, were used to uniformly predict CTCF binding sites using FIMO (default settings) on human (hg18, hg19, hg38, T2T) and mouse (mm9, mm10, mm39) genome assemblies. Extra columns include motif/PWM name (e.g., MA0139.1), score, p-value, q-value, and the motif sequence. It is recommended to filter FIMO-predicted sites by 1e-6 p-value threshold instead of using the default 1e-4 threshold. Experimentally obtained CTCF-bound cis-regulatory elements from ENCODE SCREEN and predicted CTCF sites from CTCFBSDB are also included. Selected data are lifted over from a different genome assembly as we demonstrated liftOver is a viable option to obtain CTCF coordinates in different genome assemblies. CTCF sites obtained using JASPAR's MA0139.1 PWM and filtered at 1e-6 p-value threshold are recommended.

Universally unique identifiers ('UUIDs') can be sub-optimal for many uses-cases because they are not the most character efficient way of encoding 128 bits of randomness; v1/v2 versions are impractical in many environments, as they require access to a unique, stable MAC address; v3/v5 versions require a unique seed and produce randomly distributed IDs, which can cause fragmentation in many data structures; v4 provides no other information than randomness which can cause fragmentation in many data structures. Providing an alternative, ULIDs (<https://github.com/ulid/spec>) have 128-bit compatibility with UUID', 1.21e+24 unique ULIDs per millisecond, support standard (text) sorting, canonically encoded as a 26 character string, as opposed to the 36 character UUID', use base32 encoding for better efficiency and readability (5 bits per character), are case insensitive, have no special characters (i.e. are URL safe) and have a monotonic sort order (correctly detects and handles the same millisecond).

This package provides a sizable genomics study such as microarray often involves the use of multiple batches (groups) of experiment due to practical complication. To minimize batch effects, a careful experiment design should ensure the even distribution of biological groups and confounding factors across batches. OSAT (Optimal Sample Assignment Tool) is developed to facilitate the allocation of collected samples to different batches. With minimum steps, it produces setup that optimizes the even distribution of samples in groups of biological interest into different batches, reducing the confounding or correlation between batches and the biological variables of interest. It can also optimize the even distribution of confounding factors across batches. Our tool can handle challenging instances where incomplete and unbalanced sample collections are involved as well as ideal balanced RCBD. OSAT provides a number of predefined layout for some of the most commonly used genomics platform. Related paper can be find at http://www.biomedcentral.com/1471-2164/13/689 .

Classification using Richard A. Harshman's Parallel Factor Analysis-1 (Parafac) model or Parallel Factor Analysis-2 (Parafac2) model fit to a three-way or four-way data array. See Harshman and Lundy (1994): <doi:10.1016/0167-9473(94)90132-5>. Classification using principal component analysis (PCA) fit to a two-way data matrix is also supported. Uses component weights from one mode of a Parafac, Parafac2, or PCA model as features to tune parameters for one or more classification methods via a k-fold cross-validation procedure. Allows for constraints on different tensor modes. Allows for inclusion of additional features alongside features generated by the component model. Supports penalized logistic regression, support vector machine, random forest, feed-forward neural network, regularized discriminant analysis, and gradient boosting machine. Supports binary and multiclass classification. Predicts class labels or class probabilities, and calculates multiple classification performance measures. Implements parallel computing via the foreach', doParallel', and doRNG packages.

Kevin Dowd's book Measuring Market Risk is a widely read book in the area of risk measurement by students and practitioners alike. As he claims, MATLAB indeed might have been the most suitable language when he originally wrote the functions, but, with growing popularity of R it is not entirely valid. As Dowd's code was not intended to be error free and were mainly for reference, some functions in this package have inherited those errors. An attempt will be made in future releases to identify and correct them. Dowd's original code can be downloaded from www.kevindowd.org/measuring-market-risk/. It should be noted that Dowd offers both MMR2 and MMR1 toolboxes. Only MMR2 was ported to R. MMR2 is more recent version of MMR1 toolbox and they both have mostly similar function. The toolbox mainly contains different parametric and non parametric methods for measurement of market risk as well as backtesting risk measurement methods.

This package provides a comprehensive suite of genome-wide association study (GWAS) methods specifically designed for biobank-scale data, including but not limited to, robust approaches for time-to-event traits (Li et al., 2025 <doi:10.1038/s43588-025-00864-z>) and ordinal categorical traits (Bi et al., 2021 <doi:10.1016/j.ajhg.2021.03.019>). The package also offers general frameworks for GWAS of any trait type (Bi et al., 2020 <doi:10.1016/j.ajhg.2020.06.003>), while accounting for sample relatedness (Xu et al., 2025 <doi:10.1038/s41467-025-56669-1>) or population structure (Ma et al., 2025 <doi:10.1186/s13059-025-03827-9>). By accurately approximating score statistic distributions using saddlepoint approximation (SPA), these methods can effectively control type I error rates for rare variants and in the presence of unbalanced phenotype distributions. Additionally, the package includes functions for simulating genotype and phenotype data to support research and method development.

Obtaining relevant set of trait specific genes from gene expression data is important for clinical diagnosis of disease and discovery of disease mechanisms in plants and animals. This process involves identification of relevant genes and removal of redundant genes as much as possible from a whole gene set. This package returns the trait specific gene set from the high dimensional RNA-seq count data by applying combination of two conventional machine learning algorithms, support vector machine (SVM) and genetic algorithm (GA). GA is used to control and optimize the subset of genes sent to the SVM for classification and evaluation. Genetic algorithm uses repeated learning steps and cross validation over number of possible solution and selects the best. The algorithm selects the set of genes based on a fitness function that is obtained via support vector machines. Using SVM as the classifier performance and the genetic algorithm for feature selection, a set of trait specific gene set is obtained.