v0.3.2

• (bug fix) Resolve performance issues in schema, and scitype(::AbstractString) in the MLJ convention (#70)

• (bug fix) Resolve problems with tables that are also abstract arrays (#71)

• Make conventions a type (#36)

v0.3.1

v0.3.1 (2019-12-19)

Diff since v0.3.0

Merged pull requests:

• adjusting test for CSV > 0.5.19 (#68) (tlienart)

v0.3.0

• Add an info method (migrated from MLJBase)
• Add table property to Schema instances and improve the show method for same

v0.2.7

• (Enhancement) Allow "global" coercions like coerce(X, Count => Continuous), coerce(X, Unknown => Multiclass) (#62, PR #63)

v0.2.6

• Enable in-place coercion for DataFrames, with coerce!(df, ...) (#50)
• Support coercion from categorical to count/continuous respecting order if there is one (#53)
• Add elscitype functionality which complements that of scitype_union (#59)

v0.2.5

• (fix) simplify the autotype logic, now when there are "few" values and the type is <:Real, the suggested type will be OrderedFactor (always) whereas before it could be Multiclass but that was inconsistent (PR #46)
• allow autotype on AbstractArray (in particular vector), issue #47

v0.2.4

• (Enhancement) Make CategoricalArrays, Tables, ColorTypes hard dependencies and dump use of AbstractTrees.

• (Bug fix) If a categorical array v has no missing elements, then its scitype will never be a Union{Missing, * } type (because the scitype is defined by scitype unions, not the array eltype) (#33)

• Define coerce(::CategoricalArray{Any}, T) where T<:Union{Missing,Finite} to re-wrap the categorical array as CategoricalArray{S} where S is the tightest type of the raw elements (PR #39)

v0.2.3

Some minor enhancements (#30, #31)

v0.2.2

• (Enhancement) Improve performance of scitype on arrays in the mlj (default) convention (#12, PR #23)

• Add compatibility bounds for Requires (PR #25)

v0.2.1

• (Enhancement) Add autotype method for guessing likely scitype of columns in tables (PR #4)