Change Log

[6.00-beta-16]

Changed

• [#42] [breaking] add-column default strategy is :strict now.

[6.00-beta-10]

Fixed

• [#41] dataset name not set on tensor path

[6.00-beta-7]

TMD upgrade, no changes in TC

[5.17]

TMD upgrade

Fixed

• [#36] reorder-columns on empty dataset returns nil

[5.11]

Fixed

• aggregate-columns didn't keep column order (#35)

[5.05.1]

Added

• pipeline functions have doc copied from original ones

[5.05]

Added

• split can turn off shuffling now (:shuffle? option)
• split :holdouts - sequence of consecutive holdouts

[5.04]

tech.ml.dataset version bump, this introduces the change of the order of the groups after group-by operation

[5.02]

Added

• split :holdout supports any number of splits (minimum 2) [#28]
• split supports split-names to provide custom names for subdatasets
• concat and concat-copying are working with grouped datasets

Fixed

• kfold split failed on small number of rows (due to partition-all behaviour

[5.01]

Added

• split->seq to return train/test splits as a sequence or datasets or as map of sequences for grouped datasets

Changed

• [breaking] tablecloth.pipeline returns a map with dataset under :metamorph/data key (see metamorph)
• [breaking] split returns now a dataset or grouped dataset with two new columns indicating train/test and split id. See split->seq for previous behaviour.

[5.00-beta-29.1]

Added

• without-grouping-> threading macro which allows operations on grouping dataset treated as a regular one.

Changed

• group-by accepts any java.util.Map for a collection of indexes (use LinkedHashMap to persist an order)
• some tablecloth.api.group-by functions moved to tablecloth.api.utils, no changes to API

[5.00-beta-29]

Changed

• add-or-replace-column(s) replaced by add-column(s) (add-or-replace-column(s) is marked as deprecated) (#16)

Fixed

• mark-as-group wasn't visible in API (#18)
• map-columns didn't propagate new-type for grouped case (#20)
• broken links (#14) in readme

[5.00-beta-28]

Added

• let-dataset - to simulate tibble from R

Fixed

• Adding a column to an empty dataset returned empty dataset

[5.00-beta-27]

Changed

• re-implementation of numerical arrays path dataset creation

[5.00-beta-25]

Added

• rows and columns new result: :as-double-arrays - convert rows to 2d double array
• dataset can be created from numerical arrays discusson

Fixed

• column from single value should create valid datatype (#10)

[5.00-beta-21a]

Added

• tablecloth.pipeline for pipeline operations

[5.00-beta-21]

Added

• concat-copying exposed.
• split function for splitting into train-test pairs with :kfold, :bootstrap, :loo and holdout strategies + stratified versions
• replace-missing with new strategy :midpoint

[5.00-beta-5a]

Fixed

• column names should keep order for provided names (#9)

[5.00-beta-5]

t.m.d update

[5.00-beta-3]

t.m.d update

Changed

• contribution guide in readme

[5.00-beta-2]

t.m.d update

Changed

• write-nippy! and read-nippy are deprecated, replaced by write! and dataset

[5.0-SNAPSHOT]

tech.ml.dataset version 5.0-alpha*

Added

• map-columns accepts optional target datatype
• ds/column->dataset functionality introduced in separate-column
• more datatypes included for conversion (:text among others)

Changed

• write-csv! replaced by write! (write-csv! is marked as deprecated)
• info field :size is replaced by :n-elems
• [breaking] separate-column 3-arity version accepts separator instead target-columns now

Fixed

• do not skip 1-row DS when folding
• do not attempt to fold empty dataset

[4.04]

tech.ml.dataset version 4.04

Added

• tests: dataset

Changed

• version number to match t.m.dataset version
• documentation:
  • gfm renderer for markdown

Fixed

• code block language alignment fix in css

[1.0.0-pre-alpha9]

tech.ml.dataset version 4.03

Added

• some operations on grouped dataset can be parallel (parallel? option set to true). These are: aggregate, unique-by, order-by, join-columns, separate-columns, ungroup

Fixed

• #2 - docs typo
• #3 - recover datatypes after ungrouping

Changed

• aggregation uses now in-place ungrouping which is much faster

[1.0.0-pre-alpha8]

tech.ml.dataset version 3.06

Added

• fill-range-replace to inject data to make continuous seqence in column
• write-nippy! and read-nippy

[1.0.0-pre-alpha7]

tech.ml.dataset version 2.13

Added

• replace-missing new strategies: :mid and :lerp, working also for dates.

Changed

• [breaking] replace-missing has different conctract and default strategy :mid. value argument is the last argument now.
• [breaking] replace-missing :up and :down strategies, when value is nil fills border missing values with nearest value.

[1.0.0-pre-alpha6]

tech.ml.dataset version 2.06

Added

• asof-join added

[1.0.0-pre-alpha4]

Added

• reshape tests
• pivot->wider accepts :drop-missing? option (default: true)

Changed

• pivot->wider drops missing rows by default
• pivto->wider order of concatenated column names is reversed (first: colnames, last: value), was opposite.
• pivot->longer :splitter accepts string used for splitting column name