tech.ml-base — techascent/tech.ml-base 2.6

auto-gridsearch-options clj

(auto-gridsearch-options system-name options)

gridsearch clj

(gridsearch system-name->options-seq
            feature-keys
            label-keys
            loss-fn
            dataset
            &
            {:keys [parallelism top-n gridsearch-depth k-fold scalar-labels?]
             :or {parallelism (.availableProcessors (Runtime/getRuntime))
                  top-n 5
                  gridsearch-depth 50
                  k-fold 5}
             :as options})

Gridsearch these system/option pairs by this dataset, averaging the errors across k-folds and taking the lowest top-n options. We are breaking out of 'simple' and into 'easy' here, this is pretty opinionated. The point is to make 80% of the cases work great on the first try.

predict clj

(predict model dataset)

train clj

(train system-name feature-keys label-keys options dataset)