Schema-Driven Machine Learning and Visualization

Machine learning and visualization package with an 'S7' backend featuring comprehensive type checking and validation, observable runs, and auditable outputs. The functional API provides one-call access to supervised and unsupervised learning through train(), cluster(), and decomp(). All configuration steps are performed using setup functions and validated. A single call to train() can handle preprocessing, decomposition, hyperparameter tuning, model stacking, and testing with nested resampling. Supports 'data.frame', 'data.table', and 'tibble' inputs and parallel execution. Plot methods for trained and resampled supervised models draw interactive figures through 'rtemis.draw'. Includes native implementations of SuperLearner, RuleFit, and Linear Additive Trees, and supports conformal prediction. The package first appeared in E.D. Gennatas (2017) < https://repository.upenn.edu/entities/publication/d81892ea-3087-4b71-a6f5-739c58626d64>.

rtemis: Schema-Driven Machine Learning & Visualization.

• Schema-driven configuration: every step of a workflow, from data ingestion and preprocessing to resampling, tuning, learning, and execution, is a typed S7 config built by a setup_*() function and published as a JSON Schema at schema.rtemis.org. write_config() and read_config() move the same config between R, the rtemis CLI, and rtemislive.
• Validated inputs: type checks and validation run when a config is created and before any fitting starts, with corrective error messages.
• Functional API: train(), cluster(), and decomp() provide one-call supervised learning, clustering, and decomposition. A single train() call handles preprocessing, decomposition, hyperparameter tuning, stacking, and nested resampling.
• Observable runs: timestamped messages report each step, at the level set by verbosity.
• Auditable outputs: records state every resolved value and its origin; review() assesses a trained model, and writeup() drafts its Methods and Results with every number traceable to its source.
• Algorithms: a broad set of supervised learners, including native SuperLearner, RuleFit, and Linear Additive Trees, alongside clustering and decomposition methods, including torch autoencoders.
• Explanation and uncertainty: Shapley values (explain()), calibration, and conformal prediction.
• Parallel execution through future or mirai, and model visualization through rtemis.draw.
• Accepts data.frame, data.table, and tibble inputs.

Installation

From CRAN

pak::pak("rtemis")

or

install.packages("rtemis")

Latest version from r-universe

pak::repo_add(rtemis_universe = "https://rtemis-org.r-universe.dev")
pak::pak("rtemis")

or using install.packages:

install.packages(
  'rtemis',
  repos = c('https://rtemis-org.r-universe.dev', 'https://cloud.r-project.org')
)

Installation of dependencies

Every rtemis call that uses third-party packages includes a check for required dependencies and will print a message if any are missing.

Transparent messaging

It is essential to maintain transparency of operations at all times. rtemis functions often call multiple other functions, sometime recursively. The package uses a formatted messaging system to provide logging output which includes:

• Timestamp
• Message
• Origin (function name)

Most function include a verbosity argument to control the level of messaging output, with support for three levels:

• 0: silent
• 1: normal messaging
• 2: detailed messaging for debugging

setup_ functions

Machine learning workflows involve multiple steps, each with their own configuration options.

It is essential that a) the user has complete control over each step, while maintaining an intuitive, user-friendly interface, and b) user input is validated early, before a potentially long-running operation is started.

The following setup_ functions are available to configure each step of the workflow:

• Supervised Learning: setup_CART(), setup_GAM(), etc.
• Tuning: setup_GridSearch()
• Clustering: setup_CMeans(), setup_HardCL(), etc.
• Decomposition: setup_NMF(), setup_ICA(), etc.
• Resampling: setup_KFold(), setup_StratSub(), etc.
• Preprocessing: setup_Preprocessor()

Supervised Learning

The following will perform hyperparameter tuning and 10-fold cross-validation.
It will train (3*3*2*5 + 1) * 25 = 2275 models total.

mod_lightgbm <- train(
  dat,
  hyperparameters = setup_LightGBM(
        num_leaves = tune_over(2^(1:3)),
        learning_rate = tune_over(.001, .005, .01),
        subsample = tune_over(.6, .9)
  ),
  outer_resampling_config = setup_StratSub(
    n_resamples = 25L
  )
)

Clustering

clust_cmeans <- cluster(
  dat,
  config = setup_CMeans(k = 4L)
)

Decomposition

decomp_ica <- decomp(
  dat,
  config = setup_ICA(k = 12L)
)

Documentation

The documentation is available at docs.rtemis.org/r/ml, which includes walkthroughs of main features and full API reference.

© 2016–2026 E.D. Gennatas. Licensed under BSD 3-Clause.