Author: T. Moudiki

Feed: R-bloggers. Author: T. Moudiki. In a previous blog post, I presented time series cross-validation with crossvalidation::crossval_ts. The most recent version of package crossvalidation, v0.4.1, contains a new function called eval_ts. How does crossvalidation::eval_ts work? As described in the graph below: crossvalidation::crossval_ts and crossvalidation::eval_ts have the same arguments, except, an argument p for crossval_ts and an argument q for eval_ts. p is the percentage of original data used for cross-validation and hyperparameter tuning, and q is the percentage of unseen data used for model validation. In addition, and most importantly, we must have p + q = 1, to avoid ... Read More

Feed: R-bloggers. Author: T. Moudiki. Two weeks ago I presented ahead, an R package for univariate and multivariate time series forecasting. And last week, I’ve shown how ahead::dynrmf could be used for automatic univariate forecasting. This week, I compare the speeds of execution of ahead::ridge2f (quasi-randomized autoregressive network) and ahead::varf (Vector AutoRegressive model), with their default parameters (notably 1 lag, and 5-steps-ahead forecasting). For more examples of multivariate time series forecasting with ahead, you can type ?ahead::ridge2f, ?ahead::varf(x), or ?ahead::plot.mtsforecast in R console, once the package is installed and loaded. Here’s how to install ahead: 1st method: from R-universe In ... Read More

Feed: R-bloggers. Author: T. Moudiki. Last week I presented ahead, an R package for univariate and multivariate time series forecasting. In particular, the function dynrmf was introduced for univariate time series, with examples of Random Forest and Support Vector Machines fitting functions (fitting and predicting through fit_func and predict_func arguments of dynrmf). First things first, here’s how to install R package ahead: 1st method: from R-universe In R console: options(repos = c( techtonique = 'https://techtonique.r-universe.dev', CRAN = 'https://cloud.r-project.org')) install.packages("ahead") 2nd method: from Github In R console: devtools::install_github("Techtonique/ahead") Or remotes::install_github("Techtonique/ahead") In version 0.2.0 of ahead, Ridge regression is the default fitting ... Read More

Feed: R-bloggers. Author: T. Moudiki. A few weeks ago, I introduced a Forecasting API that I deployed on Heroku. Under the hood, this API is built on top of ahead (and through Python packages rpy2 and Flask); an R package for univariate and multivariate time series forecasting. As of October 13th, 2021, 5 forecasting methods are implemented in ahead: armagarchf: univariate time series forecasting method using simulation of an ARMA(1, 1) – GARCH(1, 1) dynrmf: univariate time series forecasting method adapted from forecast::nnetar to support any Statistical/Machine learning model (such as Ridge Regression, Random Forest, Support Vector Machines, etc.) eatf: ... Read More

Feed: R-bloggers. Author: T. Moudiki. options( repos = c(techtonique = 'https://techtonique.r-universe.dev', CRAN = 'https://cloud.r-project.org') ) install.packages("crossvalidation") library(crossvalidation) library(e1071) transforming model response into a factor y <- as.factor(as.numeric(iris$Species)) explanatory variables X <- as.matrix(iris[, c("Sepal.Length", "Sepal.Width", "Petal.Length", "Petal.Width")]) OF <- function(xx) { res <- crossvalidation::crossval_ml( x = X, y = y, k = 5, repeats = 3, p = 0.8, fit_func = e1071::svm, predict_func = predict, packages = "e1071", fit_params = list(gamma = xx[1], cost = xx[2]) ) # default metric is accuracy return(res$mean_training) } There are many, many ways to maximize this objective function. simulation of SVM’s hyperparameters’ matrix n_points <- ... Read More

Feed: R-bloggers. Author: T. Moudiki. options(repos = c( techtonique = 'https://techtonique.r-universe.dev', CRAN = 'https://cloud.r-project.org')) install.packages("crossvalidation") library(crossvalidation) library(randomForest) library(microbenchmark) set.seed(123) n <- 1000 ; p <- 10 X <- matrix(rnorm(n * p), n, p) y <- rnorm(n) tuning_grid <- base::expand.grid(mtry = c(2, 3, 4), ntree = c(100, 200, 300)) n_params <- nrow(tuning_grid) print(tuning_grid) n_cores <- 4 Sequential f1 <- function() base::lapply(1:n_params, function(i) crossvalidation::crossval_ml( x = X, y = y, k = 5, repeats = 3, fit_func = randomForest::randomForest, predict_func = predict, packages = "randomForest", fit_params = list(mtry = tuning_grid[i, "mtry"], ntree = tuning_grid[i, "ntree"]) )) Parallel 1 f2 <- function() parallel::mclapply(1:n_params, ... Read More

Feed: R-bloggers. Author: T. Moudiki. I had to rename my R package crossval – generic functions for cross-validation – to crossvalidation, because its name was clashing with an existing CRAN R package’s named crossval. Here is how to install crossvalidation: options(repos = c( techtonique = 'https://techtonique.r-universe.dev', CRAN = 'https://cloud.r-project.org')) install.packages("crossvalidation") What is the R-universe mentioned in the previous code snippet? It is, IMHO, a quite promising CRAN-like repository for storing, sharing and building R packages (for Linux, macOS and Windows). If you want to create your own repository on R-universe, read this. I’ve been looking for such an infrastructure for ... Read More

Feed: R-bloggers. Author: T. Moudiki. This post is about a predictive analytics tool (in beta version) I started building during the first lockdown in March 2020. I used R and Python for this purpose, and more specifically Flask and rpy2. It’s been a pretty cool and instructive experience for me, of having both R and Python interacting into a web app, and deploying the tool on Heroku: https://www.techtonique.net What’s inside? It’s a suite of Application Programming interfaces (APIs), a system which can receive requests from your computer keyboard, to carry out given tasks on given resources. This system is programming-language-agnostic ... Read More

Feed: R-bloggers. Author: T. Moudiki. I was selected and invited to present this family of univariate/multivariate time series models at R/Finance 2020 (in Chicago, IL). However, the COVID-19 pandemics decided differently. It’s still a work in progress; comments, remarks, pull requests are welcome as usual (in nnetsauce). But the general philosophy of model construction in this framework is already set and will be explained in this post. An R example, with less details about the implementation (so, R users read on) can also be found in this post: Bayesian forecasting for uni/multivariate time series. How does this family of time ... Read More

Feed: R-bloggers. Author: T. Moudiki. In previous posts, I introduced LSBoost; a gradient boosting machine that uses randomized and penalized least squares as a basis – instead of decision trees which are frequently used as base learners. mlsauce’s LSBoost takes into account a problem’s nonlinearity by including new, engineered explanatory variables (g(XW+b)) with: (g): an activation function (tanh, ReLU, sigmoid, …) (X): input data (covariates, explanatory variables) (W): a matrix containing numbers drawn from a multivariate uniform distribution on ([0, 1]) New activation functions were added to version 0.8.0 of mlsauce: ReLU6, tanh, sigmoid. These changes are available both in ... Read More