Package index • timesift

The one call

Two tables to a scored comparison of representations, and the prediction that follows from it.

timesift(): Fit and compare representations of time-varying data
c(<timesift_learner>) c(<timesift_models>) c(<timesift_representation>) c(<timesift_sift>): Combine learners, or representations, into a set
summary(<timesift>) print(<timesift>) print(<timesift_summary>): What a run found
predict(<timesift>): Predict from a fitted timesift
plot(<timesift>): Draw a run

How a series becomes the array a learner reads, and the set a run is compared across.

native() grain() multigrain() lookback(): How a series becomes an array a learner reads
grains() lookbacks() as_sift(): Several representations to run the same learners across
build_representation(): Build one representation for a set of targets

The arms that ship, how they are trained, and the interface a learner of your own goes through.

One fold map read by everything that scores, and the cells a score is defined on, computed with no model involved.

Weights fitted on the out-of-fold predictions alone.

tss(): The true skill statistic
roc_auc(): The area under the ROC curve
kappa_score() decision_threshold() model_agreement(): Cohen's kappa, and where two models disagree
score_predictions(): Score held-out predictions on the cells the mask allows
paired_contrast(): Compare two arms cell by cell
grain_contrasts(): Compare every grain against a learner's best one
tss_inflation(): How much a self-selected threshold inflates the reported level
implied_skill(): What population skill a reported level is consistent with
occlusion(): What part of the record a fitted model reads

Readings in long form to a [unit, bin, channel] array, reachable without the fitting layer.

grain_matrix(): Reduce sensor series to a temporal grain
lookback_matrix(): Reduce sensor series to a lookback anchored on each target
coverage(): Which units reach which bins
timesift_set(): Several built representations of the same targets
calendar_channels(): Where in the year each bin sits
bind_channels(): Put channels side by side
feature_matrix(): Bring an already-reduced feature table into a ladder

Fitting across a set of grains on its own split, and reading the grain a ladder saturates at.

grain_ladder() summary(<timesift_ladder>): Fit at every grain and see where skill saturates
fit_learner() predict(<timesift_fit>): Fit one learner at one grain
select_grain() summary(<timesift_selection>): Choose the grain inside the training data, and score the whole procedure
plot(<timesift_ladder>): Draw a ladder
plot(<timesift_selection>): Draw how stable the choice of grain was

The response head and the metric are registrations, never a fork of the fitting code.

The three artifacts a split is carried in, and the digest that says two arrays are the same array.

write_folds() read_folds() write_response() read_response() write_cells() read_cells(): Read and write the artifacts that cross the language boundary
digest_array(): The cross-language digest of a representation

simulate_records(): Simulate sensor records whose response acts at a known temporal grain

timesift-package: timesift: Temporal Climate Resolution for Ecological Prediction