tech.ml.train — techascent/tech.ml-base 2.1

average-prediction-error clj

(average-prediction-error train-fn
                          predict-fn
                          ds-entry->label-fn
                          loss-fn
                          dataset-seq)

Average prediction error across models generated with these datasets Page 242, https://web.stanford.edu/~hastie/ElemStatLearn/

dataset-seq->dataset-model-seq clj

(dataset-seq->dataset-model-seq train-fn dataset-seq)

Given a sequence of {:train-ds ...} datasets, produce a sequence of: {:model ...} train-ds is removed to keep memory usage as low as possible. See dataset/dataset->k-fold-datasets

find-best-options clj

(find-best-options train-fn
                   predict-fn
                   label-key
                   loss-fn
                   {:keys [parallelism top-n]
                    :or {parallelism (.availableProcessors (Runtime/getRuntime))
                         top-n 5}}
                   option-seq
                   dataset-seq)

Given a sequence of options and a sequence of datasets (for k-fold), run them and return the best options. train-fn: (train-fn options dataset) -> model predict-fn: (predict-fn options dataset) -> prediction-sequence label-key: key to get labels from dataset. loss-fn: (loss-fn label-sequence prediction-sequence)-> double Lowest number wins.

options-seq clj

(options-seq base-options parameter-sequence-map)

Given base options map and a map of parameter keyword -> value sequence produce a sequence of options maps that does a cartesian join across all of the parameter sequences