unifiedml: a unified machine learning interface for R, now on CRAN + Discussion on AI replacing humans | R bloggers

unifiedml is now available on CRAN. Just because. I wanted to see what was new 🙂

We had an interesting discussion about the process of package publishing and AI replace human beings (the kind of discussion I’ll have with you when I’m gone). this mysterious prison).

Here’s what I had to say something about it:

>      > I see it has changed :) Just a question for my personal culture:
>     Is this going to be done by AI someday? Because it could/should.
>
>     You want to be replaced by AI?


On 11.11.2025 09:23, Thierry Moudiki wrote:
> The question of whether one would want to be replaced by AI or not is
> too narrow (in general and in particular in this context).
>
> To me, personally, it's like asking someone, back in the day, whether
> he/she wants to use the wheel or not.
>
> It's already much better than us in many mundane tasks. And it could be
> way, way more objective in this particular context.

On 11.12.2025 00:09, Thierry Moudiki wrote:
I said "AI" to follow up on your argument but this has nothing to do with text models. Not everything has to be AI.

In order to validate or push such a thing in prod, you typically need GitHub Actions and a set of objective rules (e.g what's a critical warning, or a critical note, if there's such a thing...). It's not even AI (see e.g https://github.com/Techtonique/techtonique-r-pkgs). 

Typically, since submitting this package to your highest authority, I've shipped ~5 packages to PyPI (within minutes, and they also check if the code is malicious, or it will be removed) without being bothered by gatekeeping ("this comma is missing", "oh english grammar requires a full stop there"). And this is not an insult, it's food for thought (if you're humble enough to accept it), when Python is eating R at breakfast everyday. 

R seems to be stuck somewhere in a very distant past with discussions like this. 

That is the comment of a stupid boy. If so foolishnessthat’s what I want to be stupid forever.

The package, unifiedmlprovides a consistent interface for machine learning in R, making it easier to work with different ML algorithms using a unified API.

R has an incredibly rich ecosystem of machine learning packages, but each has its own syntax and conventions. unifiedml is an attempt to bridge this gap by:

• Extremely light (see for yourself: https://github.com/Techtonique/unifiedml/blob/main/R/model.R) and consistent API for different ML algorithms
• Automatic task detection (regression vs. classification)
• Built-in cross-validation with appropriate metrics
• Tools for model interpretation, including feature importance and partial dependence plots
• Seamless integration with existing R packages (once installed) such as glmnet, randomForest and more

install.packages("unifiedml", repos = "https://cran.r-project.org/")

library(unifiedml)

This is how easy it is to build a regression model:

library(glmnet)
data(mtcars)

# Prepare data
X <- as.matrix(mtcars[, -1])
y <- mtcars$mpg  # numeric → automatic regression

# Fit model
mod <- Model$new(glmnet::glmnet)
mod$fit(X, y, alpha = 0, lambda = 0.1)

# Make predictions
predictions <- mod$predict(X)

# Get model summary with feature importance
mod$summary()

# Visualize partial dependence
mod$plot(feature = 1)

# Cross-validation (automatically uses RMSE for regression)
cv_scores <- cross_val_score(mod, X, y, cv = 5)
cat("Mean RMSE:", mean(cv_scores), "\n")

The same intuitive API works for classification:

library(randomForest)

data(iris)

# Prepare data
X <- as.matrix(iris[, 1:4])
y <- iris$Species  # factor → automatic classification

# Fit model
mod <- Model$new(randomForest::randomForest)
mod$fit(X, y, ntree = 100)

# Make predictions
predictions <- mod$predict(X)

# Cross-validation (automatically uses accuracy for classification)
cv_scores <- cross_val_score(mod, X, y, cv = 5)
cat("Mean Accuracy:", mean(cv_scores), "\n")

Main features

1. Consistent interface Whether you use glmnet, randomForest, xgboost or any other compatible algorithm, the interface remains the same. This makes it easy to:

• Switch between algorithms
• Compare model performance
• Build reproducible workflows

1. Automatic task detection You don’t have to specify whether you are running regression or classification: unifiedml automatically detects this based on the type of target variable.
2. Model interpretation (in progress) Built-in tools to understand your models: feature importance rankings Partial dependency graphs Extensive model summaries

unifiedml is still very young, so there power are some rough edges. The package will mature over time as I refine the lightweight API (see for yourself: https://github.com/Techtonique/unifiedml/blob/main/R/model.R), extend functionality and incorporate user feedback.

Feedback Welcome.

Related