Version 0.4.0

Merge pull request #80 from JuliaStats/an/revert

Revert "Handle prediction that returns a confidence interval (#77)"

Version 0.3.1

• Fix droprandomeffects on Julia 1.0.

Drop support for 0.6

This release is 0.7 only, and fixes the remaining deprecations on 0.7 (#69).

ModelFrame.msng has been renamed nonmissing which more clearly matches its meaning (#70)

v0.2.6: bug fixes and compatibility with 0.7

• Use something instead of coalesce for unspecified values represented by nothing (#67)
• Fix copy(f::Formula) when one side is a Symbol (#68)

v0.2.5: bug fixes and compatibility with Base

• StatsBase.stderr is now StatsBase.stderror (#59, JuliaStats/StatsBase.jl#368) to avoid conflicts with Base.stderr (renamed from STDERR)
• Bugfix in error message for Formula (@matthieugomez #56)
• Fix missing StatsBase coefnames method for DataFrameStatisticalModel and DataFrameRegressionModel

Internal updates to formula parsing

Major internal changes to how formula expressions are parsed. There are no differences if you're interacting with formulae through the ModelFrame or Terms layers and the @formula macro, but two important changes if you interact with Formulas directly:

1. the formula expression is processed according to the DSL at parse time (e.g., in the @formula macro) instead of at run-time.
2. the Formula struct holds the lhs and rhs sub-expressions, but also holds the original (un-parsed) expression in the ex_orig field, and the parsed expression (i.e., a :call to ~) in the ex field. The lhs and rhs fields now correspond to the lhs and rhs of the parsed expression, rather than the un-processed original expression.

These should only affect users that actually rely on the internals of the formula struct or construct them manually. There's a deprecated two-argument Formula(lhs, rhs) constructor that mimics the old Formula constructor, creating a Formula with empty ex_orig and ex = :($lhs ~ $rhs), but does no further processing of the expression.

Internally, the parsing code has been completely re-written to use rewrites rules with a uniform API.

Version 0.2.3

unique from categoricalarrays 0.3.8 (#55)

* sort unique values returned by unique(CategoricalArrays)

* don't sort, use intersect to grab order from levels

v0.2.2: Merge pull request #51 from JuliaStats/anj/deunion

Fix small issue in deunionization fix

Version 0.2.1

• Add drop_intercept function for model types which cannot estimate an intercept (like the Cox model).

Version 0.2.0

• coefnames is now defined in StatsBase (since 0.19.2), but still reexported by the package.
• This release also incorporates changes for compatibility with julia 0.7 although those cannot be checked on Travis because of problems with dependencies.