LoginSign Up
← Back to all sparks
T

timetk

ANALYTICS
Velocity0.0

Toolkit for visualizing, wrangling and feature engineering time series data in R

github.com/business-science/timetk ↗

timetk swallowed anomalize whole, then went quiet for two years

time-seriesanomaly-detectionvisualizationfeature-engineeringr-package
◆Current state
timetk handles time series wrangling, visualization, and feature engineering in a tidyverse idiom. Its defining recent move was absorbing the anomalize package outright in 2.9.0, bringing anomaly detection, cleaning, and the associated plots inside timetk rather than leaving them in a sibling package. Development then paused for nearly two years before 2.9.1, a robustness and documentation release.
◆Where it's heading
The trajectory before the pause was consolidation: fold in adjacent functionality, then make the visualization layer handle many series at once via trelliscopejs, then broaden feature generation with tk_tsfeatures(). The recent release works on the least glamorous layer — internal generics so date parsing and sequence generation behave consistently across Date, POSIXct, hms, yearmon, and yearqtr — which is the kind of foundation work a package does when it has accumulated too many special cases. The long gap and the CI-refresh content suggest maintenance attention rather than a new direction.
◆Prediction
The stated gap is the unimplemented twitter method for anomalize(), which is the one concrete outstanding item the release notes name; beyond that the recent work points to consolidation rather than expansion.

◆Recent moves

  1. 11mo ago

    timetk 2.9.1 makes date parsing consistent across time classes

    New exported internal generics make date parsing and future sequence generation consistent across Date, POSIXct, hms, yearmon, and yearqtr, alongside a vignette overhaul for anomaly workflows and a CI refresh. Foundation work after a long quiet period rather than new capability.

    View source ↗
  2. 2y ago

    timetk 2.9.0 absorbs the anomalize package

    ⚡ SPARK

    The anomalize package is folded into timetk, bringing anomalize(), plot_anomalies(), plot_anomalies_decomp(), and plot_anomalies_cleaned() into a single package, with the STL method supported and the twitter method planned. It ends the split between wrangling a time series and finding what is wrong with it.

    View source ↗
  3. 4y ago

    timetk 2.8.1 exposes trelliscope plotting parameters

    Exposes the facet_trelliscope() plotting parameters to plot_time_series(), plot_time_series_boxplot(), and plot_anomaly_diagnostics(). A follow-up giving users control over the trelliscope integration added in 2.8.0.

    View source ↗
  4. 4y ago

    timetk 2.8.0 adds trelliscopejs support for many-series plots

    Plotting functions gain trelliscopejs support for visualizing many time series at once, with new trelliscope, facet_strip_remove, and facet_nrow arguments, and facet_collapse now defaulting to FALSE. Aimed squarely at the many-series case that ordinary faceting cannot handle.

    View source ↗
  5. 4y ago

    timetk 2.7.0 adds tk_tsfeatures() for grouped feature matrices

    tk_tsfeatures() generates a time series feature matrix from grouped tibbles via tsfeatures, and plot_time_series_boxplot() adds period-aggregated boxplots, both with new vignettes including one on time series clustering. Feature generation is what makes clustering across many series practical.

    View source ↗
  6. 4y ago

    timetk 2.6.2 adds .week_start and facet direction controls

    summarize_by_time() gains a .week_start argument for Monday-start weeks, several plotting functions gain .facet_dir, and plot_acf_diagnostics() shows white noise bars by default. Small ergonomic corrections across the plotting and aggregation surface.

    View source ↗