Merge pull request #987 from JuliaAI/dev

For a 1.6 release

Project.toml

@@ -1,7 +1,7 @@
 name = "MLJBase"
 uuid = "a7f614a8-145f-11e9-1d2a-a57a1082229d"
 authors = ["Anthony D. Blaom <[email protected]>"]
-version = "1.5.0"
+version = "1.6"
 
 [deps]
 CategoricalArrays = "324d7699-5711-5eae-9e2f-1d82baa6b597"
@@ -47,7 +47,7 @@ DelimitedFiles = "1"
 Distributions = "0.25.3"
 InvertedIndices = "1"
 LearnAPI = "0.1"
-MLJModelInterface = "1.10"
+MLJModelInterface = "1.11"
 Missings = "0.4, 1"
 OrderedCollections = "1.1"
 Parameters = "0.12"
@@ -58,7 +58,7 @@ Reexport = "1.2"
 ScientificTypes = "3"
 StatisticalMeasures = "0.1.1"
 StatisticalMeasuresBase = "0.1.1"
-StatisticalTraits = "3.3"
+StatisticalTraits = "3.4"
 Statistics = "1"
 StatsBase = "0.32, 0.33, 0.34"
 Tables = "0.2, 1.0"

src/composition/learning_networks/nodes.jl

@@ -277,7 +277,7 @@ function _formula(stream, X::Node, depth, indent)
     if X.machine !== nothing
         print(stream, crind(indent + length(operation_name) - anti))
         printstyled(IOContext(stream, :color=>SHOW_COLOR[]),
-#                        handle(X.machine),
+                        #handle(X.machine),
                         X.machine,
                         bold=SHOW_COLOR[])
         n_args == 0 || print(stream, ", ")

src/composition/learning_networks/signatures.jl

@@ -307,7 +307,8 @@ See also [`MLJBase.Signature`](@ref).
 """
 fitted_params_supplement(signature::Signature) = call_and_copy(fitted_params_nodes(signature))
 
-""" report(signature; supplement=true)
+"""
+    report(signature; supplement=true)
 
 **Private method.**
 

src/composition/models/pipelines.jl

@@ -182,7 +182,7 @@ or what `transform` returns if it is `Unsupervised`.
 Names for the component fields are automatically generated unless
 explicitly specified, as in
 
-```
+```julia
 Pipeline(encoder=ContinuousEncoder(drop_last=false),
          stand=Standardizer())
 ```
@@ -225,6 +225,15 @@ implements it (some clustering models). Similarly, calling `transform`
 on a supervised pipeline calls `transform` on the supervised
 component.
 
+### Transformers that need a target in training
+
+Some transformers that have type `Unsupervised` (so that the output of `transform` is
+propagated in pipelines) may require a target variable for training. An example are
+so-called target encoders (which transform categorical input features, based on some
+target observations). Provided they appear before any `Supervised` component in the
+pipelines, such models are supported. Of course a target must be provided whenever
+training such a pipeline, whether or not it contains a `Supervised` component.
+
 ### Optional key-word arguments
 
 - `prediction_type`  -
@@ -444,9 +453,13 @@ function extend(front::Front{Pred}, ::Static, name, cache, args...)
     Front(transform(mach, active(front)), front.transform, Pred())
 end
 
-function extend(front::Front{Trans}, component::Unsupervised, name, cache, args...)
+function extend(front::Front{Trans}, component::Unsupervised, name, cache, ::Any, sources...)
     a = active(front)
-    mach = machine(name, a; cache=cache)
+    if target_in_fit(component)
+        mach = machine(name, a, first(sources); cache=cache)
+    else
+        mach = machine(name, a; cache=cache)
+    end
     Front(predict(mach, a), transform(mach, a), Trans())
 end
 
@@ -598,6 +611,7 @@ function MMI.iteration_parameter(pipe::SupervisedPipeline)
 end
 
 MMI.target_scitype(p::SupervisedPipeline) = target_scitype(supervised_component(p))
+MMI.target_in_fit(p::SomePipeline) = any(target_in_fit, components(p))
 
 MMI.package_name(::Type{<:SomePipeline}) = "MLJBase"
 MMI.load_path(::Type{<:SomePipeline}) = "MLJBase.Pipeline"

src/data/data.jl

@@ -401,12 +401,18 @@ _isnan(x::Number) = isnan(x)
 
 skipnan(x) = Iterators.filter(!_isnan, x)
 
+isinvalid(x) = ismissing(x) || _isnan(x)
+
 """
     skipinvalid(itr)
 
 Return an iterator over the elements in `itr` skipping `missing` and
 `NaN` values. Behaviour is similar to [`skipmissing`](@ref).
 
+"""
+skipinvalid(v) = v |> skipmissing |> skipnan
+
+"""
     skipinvalid(A, B)
 
 For vectors `A` and `B` of the same length, return a tuple of vectors
@@ -417,10 +423,6 @@ always returns a vector. Does not remove `Missing` from the element
 types if present in the original iterators.
 
 """
-skipinvalid(v) = v |> skipmissing |> skipnan
-
-isinvalid(x) = ismissing(x) || _isnan(x)
-
 function skipinvalid(yhat, y)
     mask = .!(isinvalid.(yhat) .| isinvalid.(y))
     return yhat[mask], y[mask]

src/data/datasets.jl

@@ -199,7 +199,7 @@ function load_smarket()
 end
 
 """Load a well-known sunspot time series (table with one column).
-[https://www.sws.bom.gov.au/Educational/2/3/6]](https://www.sws.bom.gov.au/Educational/2/3/6)
+<https://www.sws.bom.gov.au/Educational/2/3/6>
 """
 load_sunspots() = load_dataset("sunspots.csv", COERCE_SUNSPOTS)
 
@@ -250,9 +250,10 @@ macro load_crabs()
     end
 end
 
-""" Load S&P Stock Market dataset, as used in (An Introduction to
-Statistical Learning with applications in
-R)[https://rdrr.io/cran/ISLR/man/Smarket.html](https://rdrr.io/cran/ISLR/man/Smarket.html),
+"""
+Load S&P Stock Market dataset, as used in
+[An Introduction to Statistical Learning with applications in
+R](https://rdrr.io/cran/ISLR/man/Smarket.html),
 by Witten et al (2013), Springer-Verlag, New York."""
 macro load_smarket()
     quote

src/data/datasets_synthetic.jl

@@ -21,12 +21,12 @@ Internal function to  finalize the `make_*` functions.
 function finalize_Xy(X, y, shuffle, as_table, eltype, rng; clf::Bool=true)
     # Shuffle the rows if required
     if shuffle
-                X, y = shuffle_rows(X, y; rng=rng)
-        end
-        if eltype != Float64
-                X = convert.(eltype, X)
-        end
-        # return as matrix if as_table=false
+        X, y = shuffle_rows(X, y; rng=rng)
+    end
+    if eltype != Float64
+        X = convert.(eltype, X)
+    end
+    # return as matrix if as_table=false
     as_table || return X, y
     clf && return MLJBase.table(X), categorical(y)
     if length(size(y)) > 1
@@ -172,7 +172,6 @@ membership to the smaller or larger circle, respectively.
 * `noise=0`: standard deviation of the Gaussian noise added to the data,
 
 * `factor=0.8`: ratio of the smaller radius over the larger one,
-
 $(EXTRA_KW_MAKE*EXTRA_CLASSIFICATION)
 
 ### Example
@@ -318,7 +317,12 @@ Make portion `s` of vector `θ` exactly 0.
 """
 sparsify!(rng, θ, s) = (θ .*= (rand(rng, length(θ)) .< s))
 
-"""Add outliers to portion s of vector."""
+"""
+    outlify!(rng, y, s)
+
+Add outliers to portion `s` of vector `y`.
+
+"""
 outlify!(rng, y, s) =
         (n = length(y); y .+= 20 * randn(rng, n) .* (rand(rng, n) .< s))
 
@@ -329,7 +333,7 @@ const SIGMOID_32 = log(Float32(1)/eps(Float32) - Float32(1))
     sigmoid(x)
 
 Return the sigmoid computed in a numerically stable way:
-``σ(x) = 1/(1+exp(-x))``
+``σ(x) = 1/(1+\\exp(-x))``
 
 """
 function sigmoid(x::Float64)

src/hyperparam/one_dimensional_range_methods.jl

@@ -66,31 +66,31 @@ In the first case iteration is over all `values` stored in the range
 iteration is over approximately `n` ordered values, generated as
 follows:
 
-(i) First, exactly `n` values are generated between `U` and `L`, with a
-spacing determined by `r.scale` (uniform if `scale=:linear`) where `U`
-and `L` are given by the following table:
-
-| `r.lower`   | `r.upper`  | `L`                 | `U`                 |
-|-------------|------------|---------------------|---------------------|
-| finite      | finite     | `r.lower`           | `r.upper`           |
-| `-Inf`      | finite     | `r.upper - 2r.unit` | `r.upper`           |
-| finite      | `Inf`      | `r.lower`           | `r.lower + 2r.unit` |
-| `-Inf`      | `Inf`      | `r.origin - r.unit` | `r.origin + r.unit` |
-
-(ii) If a callable `f` is provided as `scale`, then a uniform spacing
-is always applied in (i) but `f` is broadcast over the results. (Unlike
-ordinary scales, this alters the effective range of values generated,
-instead of just altering the spacing.)
-
-(iii) If `r` is a discrete numeric range (`r isa NumericRange{<:Integer}`)
-then the values are additionally rounded, with any duplicate values
-removed. Otherwise all the values are used (and there are exacltly `n`
-of them).
-
-(iv) Finally, if a random number generator `rng` is specified, then the values are
-returned in random order (sampling without replacement), and otherwise
-they are returned in numeric order, or in the order provided to the
-range constructor, in the case of a `NominalRange`.
+1. First, exactly `n` values are generated between `U` and `L`, with a
+   spacing determined by `r.scale` (uniform if `scale=:linear`) where `U`
+   and `L` are given by the following table:
+
+   | `r.lower`   | `r.upper`  | `L`                 | `U`                 |
+   |-------------|------------|---------------------|---------------------|
+   | finite      | finite     | `r.lower`           | `r.upper`           |
+   | `-Inf`      | finite     | `r.upper - 2r.unit` | `r.upper`           |
+   | finite      | `Inf`      | `r.lower`           | `r.lower + 2r.unit` |
+   | `-Inf`      | `Inf`      | `r.origin - r.unit` | `r.origin + r.unit` |
+
+2. If a callable `f` is provided as `scale`, then a uniform spacing
+   is always applied in (1) but `f` is broadcast over the results. (Unlike
+   ordinary scales, this alters the effective range of values generated,
+   instead of just altering the spacing.)
+
+3. If `r` is a discrete numeric range (`r isa NumericRange{<:Integer}`)
+   then the values are additionally rounded, with any duplicate values
+   removed. Otherwise all the values are used (and there are exacltly `n`
+   of them).
+
+4. Finally, if a random number generator `rng` is specified, then the values are
+   returned in random order (sampling without replacement), and otherwise
+   they are returned in numeric order, or in the order provided to the
+   range constructor, in the case of a `NominalRange`.
 
 """
 iterator(rng::AbstractRNG, r::ParamRange, args...) =

src/machines.jl

@@ -529,7 +529,7 @@ err_missing_model(model) = ErrorException(
 )
 
 """
-   last_model(mach::Machine)
+    last_model(mach::Machine)
 
 Return the last model used to train the machine `mach`. This is a bona fide model, even if
 `mach.model` is a symbol.
@@ -572,31 +572,31 @@ the true model given by `getproperty(composite, model)`. See also [`machine`](@r
 For the action to be a no-operation, either `mach.frozen == true` or
 or none of the following apply:
 
-- (i) `mach` has never been trained (`mach.state == 0`).
+1. `mach` has never been trained (`mach.state == 0`).
 
-- (ii) `force == true`.
+2. `force == true`.
 
-- (iii) The `state` of some other machine on which `mach` depends has
-  changed since the last time `mach` was trained (ie, the last time
-  `mach.state` was last incremented).
+3. The `state` of some other machine on which `mach` depends has
+   changed since the last time `mach` was trained (ie, the last time
+   `mach.state` was last incremented).
 
-- (iv) The specified `rows` have changed since the last retraining and
-  `mach.model` does not have `Static` type.
+4. The specified `rows` have changed since the last retraining and
+   `mach.model` does not have `Static` type.
 
-- (v) `mach.model` is a model and different from the last model used for training, but has
-  the same type.
+5. `mach.model` is a model and different from the last model used for training, but has
+   the same type.
 
-- (vi) `mach.model` is a model but has a type different from the last model used for
-  training.
+6. `mach.model` is a model but has a type different from the last model used for
+   training.
 
-- (vii) `mach.model` is a symbol and `(composite, mach.model)` is different from the last
-  model used for training, but has the same type.
+7. `mach.model` is a symbol and `(composite, mach.model)` is different from the last
+   model used for training, but has the same type.
 
-- (viii) `mach.model` is a symbol and `(composite, mach.model)` has a different type from
-  the last model used for training.
+8. `mach.model` is a symbol and `(composite, mach.model)` has a different type from
+   the last model used for training.
 
-In any of the cases (i) - (iv), (vi), or (viii), `mach` is trained ab initio. If (v) or
-(vii) is true, then a training update is applied.
+In any of the cases (1) - (4), (6), or (8), `mach` is trained ab initio.
+If (5) or (7) is true, then a training update is applied.
 
 To freeze or unfreeze `mach`, use `freeze!(mach)` or `thaw!(mach)`.
 
@@ -658,7 +658,7 @@ function fit_only!(
     rows === nothing && (rows = (:))
     rows_is_new = !isdefined(mach, :old_rows) || rows != mach.old_rows
 
-    condition_iv = rows_is_new && !(mach.model isa Static)
+    condition_4 = rows_is_new && !(mach.model isa Static)
 
     upstream_has_changed = mach.old_upstream_state != upstream_state
 
@@ -672,16 +672,16 @@ function fit_only!(
 
     # build or update cached `resampled_data` if necessary (`mach.data` is already defined
     # above if needed here):
-    if cache_data && (!data_is_valid || condition_iv)
+    if cache_data && (!data_is_valid || condition_4)
         mach.resampled_data = selectrows(model, rows, mach.data...)
     end
 
     # `fit`, `update`, or return untouched:
-    if mach.state == 0 ||       # condition (i)
-        force == true ||        # condition (ii)
-        upstream_has_changed || # condition (iii)
-        condition_iv ||         # condition (iv)
-        modeltype_changed       # conditions (vi) or (vii)
+    if mach.state == 0 ||       # condition (1)
+        force == true ||        # condition (2)
+        upstream_has_changed || # condition (3)
+        condition_4 ||          # condition (4)
+        modeltype_changed       # conditions (6) or (7)
 
         isdefined(mach, :report) || (mach.report = LittleDict{Symbol,Any}())
 
@@ -709,7 +709,7 @@ function fit_only!(
                 rethrow()
             end
 
-    elseif model != mach.old_model # condition (v)
+    elseif model != mach.old_model # condition (5)
 
         # update the model:
         fitlog(mach, :update, verbosity)
@@ -1044,9 +1044,10 @@ To serialise using a different format, see [`serializable`](@ref).
 Machines are deserialized using the `machine` constructor as shown in
 the example below.
 
-> The implementation of `save` for machines changed in MLJ 0.18
->  (MLJBase 0.20). You can only restore a machine saved using older
->  versions of MLJ using an older version.
+!!! note
+    The implementation of `save` for machines changed in MLJ 0.18
+    (MLJBase 0.20). You can only restore a machine saved using older
+    versions of MLJ using an older version.
 
 ### Example
 
@@ -1073,8 +1074,7 @@ predict(predict_only_mach, X)
     general purpose serialization formats, can allow for arbitrary code
     execution during loading. This means it is possible for someone
     to use a JLS file that looks like a serialized MLJ machine as a
-    [Trojan
-    horse](https://en.wikipedia.org/wiki/Trojan_horse_(computing)).
+    [Trojan horse](https://en.wikipedia.org/wiki/Trojan_horse_(computing)).
 
 See also [`serializable`](@ref), [`machine`](@ref).