Skip to content

Commit

Permalink
[cortex/dataset] map-sequence->dataset docstring improvement. (#70)
Browse files Browse the repository at this point in the history
  • Loading branch information
harold authored and Chris Nuernberger committed Jan 18, 2017
1 parent bb5ccc7 commit c851b20
Showing 1 changed file with 8 additions and 6 deletions.
14 changes: 8 additions & 6 deletions cortex/src/cortex/dataset.clj
Expand Up @@ -343,13 +343,15 @@ data."
[item-key (map #(get % item-key) seq-of-maps)]))
(into {})))


(defn map-sequence->dataset
"Given sequences of maps (one of them infinite for training
and the number of elements in an epoch construct an infinite dataset.
Uses m/ecount to divine the shape from the first item in the infinite sequence.
If a cross-validation sequence is not provided one will be created. IF a holdout
sequence is not create one with be created from the original infinite sequence."
"Turns an infinite training sequence of observation maps (and optionally
sequences of cv and holdout observation maps) into a dataset. The caller also
must provide the number of observations processed per epoch.
Uses m/ecount to divine a shape from the first item in the infinite sequence.
If a cross-validation sequence is not provided one will be created. If a
holdout sequence is not provided it will be created similarly."
[infinite-map-seq num-epoch-elements & {:keys [cv-map-seq holdout-map-seq
use-cv-for-holdout? shutdown-fn]
:or {use-cv-for-holdout? true}}]
Expand Down

0 comments on commit c851b20

Please sign in to comment.