Releases: JuliaAI/MLJBase.jl
v0.11.5
MLJBase v0.11.5
-
rebinding of measure names to non-abbreviated versions of their aliases
-
(enhancement) Allow specification of
shuffle=...
andrng=...
inunpack
(#167, PR #190) -
(enhancement) Enable loading of datasets from OpenML with
OpenML.load
Closed issues:
- Remove two letter measures (#187)
Merged pull requests:
v0.11.4
MLJBase v0.11.4
Add forgotten exports: tp, fp, tn, fn
Merged pull requests:
v0.11.3
v0.11.2
MLJBase v0.11.2
Closed issues:
- Towards next release (#145)
Merged pull requests:
- Install TagBot as a GitHub Action (#174) (@JuliaTagBot)
- Crayon (#179) (@tlienart)
v0.11.1
v0.11.0
v0.10.1
v0.10.0
-
Give the
partition
function a new keyword argumentstratify=nothing
for specifying aFinite
vector on which to base stratified partitioning. Query?partition
for details (#113) -
Optimize
selectrows
forDataFrames
, row tables, and column tables (JuliaAI/MLJ.jl#122) -
Add new methods for generating synthetic data sets:
make_blobs
,make_moons
,make_circles
,make_regression
(#155) -
Update to ScientifcTypes 0.5.1
-
Import resampling interface and its implementations from MLJ
-
Improve
show
method for the results of performance evaluations (callingevaluate!
,evaluate
) -
Add keyword argument
repeats=1
toevaluate!
/evaluate
for repeated resampling. For example, specifyingresampling=CV(nfolds=3, shuffle=true), repeats=2
is to generate 6per_fold
performance estimates for aggregation. Query?evaluate!
for details (JuliaAI/MLJ.jl#406) -
Import one-dimensional ranges (
ParamRange
objects) MLJ and add enhancements: unbounded nominal ranges allowed, with neworigin
andscale
fields; grid generation for unbounded ranges. Allow specification of a type rather than a model in the constructor. Query?range
and?iterator
for details. -
Add a model trait called
hyperparameter_ranges
to allow model implementers to specify default one-dimensional ranges for hyper parameters -
(breaking) Default for
is_julia(::Type{<:Model)
is returned tofalse
instead ofmissing
to prevent case-distinction headaches downstream -
improve
show
method forMLJType
objects that "show as constructed" (JuliaAI/MLJ.jl#351)
v0.9.2
- add functionalities to generate foo data (PR #136 and #155, issue #118)
- export
coerce!
(PR #155, issue #149) - stratified sampling for
partition
(PR #138, issue #113) - minor improvements for cross entropy and briers score (PR #148)
- improve performance of
selectrows
for DataFrames and julia native tables (PR #154)
v0.9.1
Fix bug in @from_network
(also effecting @pipeline
) preventing use from MLJ without first doing import MLJBase
.