LoginSign Up
← Back to all sparks
M

modeltime

ANALYTICS
Velocity0.0

Time series forecasting framework for R

github.com/business-science/modeltime ↗

modeltime built conformal intervals in, then went quiet on features.

forecastingconformal-predictiontidymodelsparallelismr-language
◆Current state
modeltime is at 1.3.3, a single change making the package robust to xgboost version shifts. The feature weight sits in 1.3.2, which added a future-based parallel backend, the maape() accuracy metric and dials helpers for ADAM engine tuning, and further back in the 1.2.8 and 1.3.0 pair that introduced conformal prediction intervals and then carried them through the nested forecasting workflow.
◆Where it's heading
The arc runs from uncertainty quantification to execution. Conformal intervals arrived first and were then threaded through nested fitting, refitting and the printed forecast tables so users can see which confidence method produced an interval. The later work moves down a layer to how forecasts are computed — a portable future backend replacing foreach tuning — rather than what they express.
◆Prediction
With only an xgboost compatibility fix since the 1.3.2 feature release, the entries do not support a confident prediction about what comes next beyond continued dependency maintenance.

◆Recent moves

  1. 7mo ago

    Robustness to xgboost version changes

    One change making the package resilient to xgboost version shifts. Dependency maintenance, not forecasting capability.

    View source ↗
  2. 11mo ago

    future parallel backend, maape() metric and ADAM tuning helpers

    parallel_start() gains a future method bridged through doFuture, cutting tuning warnings and making parallel setup portable. Adds the maape() metric to the extended accuracy set and dials helpers exposing ADAM model and loss selection.

    View source ↗
  3. 2y ago

    Conformal intervals reach the nested forecasting workflow

    Carries conformal prediction intervals into the nested forecasting workflow via modeltime_nested_fit() and refit, and reworks the printed forecast tables to show which confidence method and interval are in play. The second half of a two-release arc that begins in 1.2.8.

    View source ↗
  4. 2y ago

    Conformal prediction intervals introduced

    Introduces conformal predictions with a dedicated vignette, and drops default parallel processing from all available cores to one across the control functions. The starting point for the interval work 1.3.0 completes.

    View source ↗
  5. 3y ago

    Fixes the Smooth es() model

    A one-line release fixing the Smooth es() model. Narrow bug fix.

    View source ↗
  6. 3y ago

    Fixes failing developer-tools tests

    Fixes failing tests in a developer-tools test file. Internal only.

    View source ↗