Lightning Fast Serialization of Data Frames

Multithreaded serialization of compressed data frames using the 'fst' format. The 'fst' format allows for full random access of stored data and a wide range of compression settings using the LZ4 and ZSTD compressors.

Dynamic Documents for R

Convert R Markdown documents into a variety of formats.

Simple Engine for Generating Reports using R

Runs R-code present in a pandoc markdown file and includes the resulting output in the resulting markdown file. This file can then be converted into any of the output formats supported by pandoc. The package can also be used as an engine for writing package vignettes.

The Uniform Manifold Approximation and Projection (UMAP) Method for Dimensionality Reduction

An implementation of the Uniform Manifold Approximation and Projection dimensionality reduction by McInnes et al. (2018) . It also provides means to transform new data and to carry out supervised dimensionality reduction. An implementation of the related LargeVis method of Tang et al. (2016) is also provided. This is a complete re-implementation in R (and C++, via the 'Rcpp' package): no Python installation is required. See the uwot website (< https://github.com/jlmelville/uwot>) for more documentation and examples.

'Memoisation' of Functions

Cache the results of a function so that when you call it again with the same arguments it returns the previously computed value.

Various R Programming Tools

Functions to assist in R programming, including: - assist in developing, updating, and maintaining R and R packages ('ask', 'checkRVersion', 'getDependencies', 'keywords', 'scat'), - calculate the logit and inverse logit transformations ('logit', 'inv.logit'), - test if a value is missing, empty or contains only NA and NULL values ('invalid'), - manipulate R's .Last function ('addLast'), - define macros ('defmacro'), - detect odd and even integers ('odd', 'even'), - convert strings containing non-ASCII characters (like single quotes) to plain ASCII ('ASCIIfy'), - perform a binary search ('binsearch'), - sort strings containing both numeric and character components ('mixedsort'), - create a factor variable from the quantiles of a continuous variable ('quantcut'), - enumerate permutations and combinations ('combinations', 'permutation'), - calculate and convert between fold-change and log-ratio ('foldchange', 'logratio2foldchange', 'foldchange2logratio'), - calculate probabilities and generate random numbers from Dirichlet distributions ('rdirichlet', 'ddirichlet'), - apply a function over adjacent subsets of a vector ('running'), - modify the TCP_NODELAY ('de-Nagle') flag for socket objects, - efficient 'rbind' of data frames, even if the column names don't match ('smartbind'), - generate significance stars from p-values ('stars.pval'), - convert characters to/from ASCII codes ('asc', 'chr'), - convert character vector to ASCII representation ('ASCIIfy'), - apply title capitalization rules to a character vector ('capwords').

Read, Write and Edit xlsx Files

Simplifies the creation of Excel .xlsx files by providing a high level interface to writing, styling and editing worksheets. Through the use of 'Rcpp', read/write times are comparable to the 'xlsx' and 'XLConnect' packages with the added benefit of removing the dependency on Java.

R Interface to 'Keras'

Interface to 'Keras' < https://keras.io>, a high-level neural networks 'API'. 'Keras' was developed with a focus on enabling fast experimentation, supports both convolution based networks and recurrent networks (as well as combinations of the two), and runs seamlessly on both 'CPU' and 'GPU' devices.

The Scalable Highly Adaptive Lasso

A scalable implementation of the highly adaptive lasso algorithm, including routines for constructing sparse matrices of basis functions of the observed data, as well as a custom implementation of Lasso regression tailored to enhance efficiency when the matrix of predictors is composed exclusively of indicator functions. For ease of use and increased flexibility, the Lasso fitting routines invoke code from the 'glmnet' package by default. The highly adaptive lasso was first formulated and described by MJ van der Laan (2017) , with practical demonstrations of its performance given by Benkeser and van der Laan (2016) . This implementation of the highly adaptive lasso algorithm was described by Hejazi, Coyle, and van der Laan (2020) .

Targeted Maximum Likelihood Estimation

Targeted maximum likelihood estimation of point treatment effects (Targeted Maximum Likelihood Learning, The International Journal of Biostatistics, 2(1), 2006. This version automatically estimates the additive treatment effect among the treated (ATT) and among the controls (ATC). The tmle() function calculates the adjusted marginal difference in mean outcome associated with a binary point treatment, for continuous or binary outcomes. Relative risk and odds ratio estimates are also reported for binary outcomes. Missingness in the outcome is allowed, but not in treatment assignment or baseline covariate values. The population mean is calculated when there is missingness, and no variation in the treatment assignment. The tmleMSM() function estimates the parameters of a marginal structural model for a binary point treatment effect. Effect estimation stratified by a binary mediating variable is also available. An ID argument can be used to identify repeated measures. Default settings call 'SuperLearner' to estimate the Q and g portions of the likelihood, unless values or a user-supplied regression function are passed in as arguments.