Add UnivariateTimeTypeToContinous - take II

src/MLJModels.jl

@@ -9,7 +9,7 @@ import MLJBase: @load
 import MLJBase: Table, Continuous, Count, Finite, OrderedFactor, Multiclass
 
 using Requires, Pkg, Pkg.TOML, OrderedCollections, Parameters
-using Tables, CategoricalArrays, StatsBase, Statistics
+using Tables, CategoricalArrays, StatsBase, Statistics, Dates
 import Distributions
 
 # for administrators to update Metadata.toml:
@@ -28,7 +28,8 @@ export ConstantRegressor, ConstantClassifier,
 # from model/Transformers
 export FeatureSelector, StaticTransformer, UnivariateDiscretizer,
     UnivariateStandardizer, Standardizer, UnivariateBoxCoxTransformer,
-    OneHotEncoder, ContinuousEncoder, FillImputer, UnivariateFillImputer
+    OneHotEncoder, ContinuousEncoder, FillImputer, UnivariateFillImputer,
+    UnivariateTimeTypeToContinuous
 
 const srcdir = dirname(@__FILE__) # the directory containing this file
 

src/builtins/Transformers.jl

@@ -1,4 +1,4 @@
-## CONSTANTS
+## CONSTANTS 
 
 const N_VALUES_THRESH = 16 # for BoxCoxTransformation
 
@@ -23,6 +23,10 @@ const CONTINUOUS_ENCODER_DESCR = "Convert all `Finite` (categorical) and "*
 "`Count` features (columns) of a table to `Continuous` and drop all "*
 " remaining non-`Continuous` features. "
 "features. "
+const UNIVARIATE_TIME_TYPE_TO_CONTINUOUS = "Transform univariate "*
+"data with element scitype `ScientificDateTime` so that it has "*
+"`Continuous` element scitype, according to a learned scale. "
+
 
 ##
 ## IMPUTER
@@ -411,10 +415,152 @@ MLJBase.inverse_transform(transformer::UnivariateStandardizer, fitresult, w) =
     [inverse_transform(transformer, fitresult, y) for y in w]
 
 
+## CONTINUOUS TRANSFORM OF TIME TYPE FEATURES
+
+"""
+    UnivariateTimeTypeToContinuous(zero_time=nothing, step=Hour(24))
+
+Convert a `Date`, `DateTime`, and `Time` vector to `Float64` by
+assuming `0.0` corresponds to the `zero_time` parameter and the time
+increment to reach `1.0` is given by the `step` parameter. The type of
+`zero_time` should match the type of the column if provided. If not
+provided, then `zero_time` is inferred as the minimum time found in
+the data when `fit` is called.
+
+"""
+mutable struct UnivariateTimeTypeToContinuous <: Unsupervised
+    zero_time::Union{Nothing, TimeType}
+    step::Period
+end
+
+function UnivariateTimeTypeToContinuous(;
+    zero_time=nothing, step=Dates.Hour(24))
+    model = UnivariateTimeTypeToContinuous(zero_time, step)
+    message = MLJBase.clean!(model)
+    isempty(message) || @warn message
+    return model
+end
+
+function MLJBase.clean!(model::UnivariateTimeTypeToContinuous)
+    # Step must be able to be added to zero_time if provided.
+    msg = ""
+    if model.zero_time !== nothing
+        try
+            tmp = model.zero_time + model.step
+        catch err
+            if err isa MethodError
+                model.zero_time, model.step, status, msg = _fix_zero_time_step(
+                    model.zero_time, model.step)
+                if status === :error
+                    # Unable to resolve, rethrow original error.
+                    throw(err)
+                end
+            else
+                throw(err)
+            end
+        end
+    end
+    return msg
+end
+
+function _fix_zero_time_step(zero_time, step)
+    # Cannot add time parts to dates nor date parts to times.
+    # If a mismatch is encountered. Conversion from date parts to time parts
+    # is possible, but not from time parts to date parts because we cannot
+    # represent fractional date parts.
+    msg = ""
+    if zero_time isa Dates.Date && step isa Dates.TimePeriod
+        # Convert zero_time to a DateTime to resolve conflict.
+        if step % Hour(24) === Hour(0)
+            # We can convert step to Day safely
+            msg = "Cannot add `TimePeriod` `step` to `Date` `zero_time`. Converting `step` to `Day`."
+            step = convert(Day, step)
+        else
+            # We need datetime to be compatible with the step.
+            msg = "Cannot add `TimePeriod` `step` to `Date` `zero_time`. Converting `zero_time` to `DateTime`."
+            zero_time = convert(DateTime, zero_time)
+        end
+        return zero_time, step, :success, msg
+    elseif zero_time isa Dates.Time && step isa Dates.DatePeriod
+        # Convert step to Hour if possible. This will fail for
+        # isa(step, Month)
+        msg = "Cannot add `DatePeriod` `step` to `Time` `zero_time`. Converting `step` to `Hour`."
+        step = convert(Hour, step)
+        return zero_time, step, :success, msg
+    else
+        return zero_time, step, :error, msg
+    end
+end
+
+function MLJBase.fit(model::UnivariateTimeTypeToContinuous, verbosity::Int, X)
+    if model.zero_time !== nothing
+        min_dt = model.zero_time
+        step = model.step
+        # Check zero_time is compatible with X
+        example = first(X)
+        try
+            X - min_dt
+        catch err
+            if err isa MethodError
+                @warn "`$(typeof(min_dt))` `zero_time` is not compatible with `$(eltype(X))` vector. Attempting to convert `zero_time`."
+                min_dt = convert(eltype(X), min_dt)
+            else
+                throw(err)
+            end
+        end
+    else
+        min_dt = minimum(X)
+        step = model.step
+        message = ""
+        try
+            min_dt + step
+        catch err
+            if err isa MethodError
+                min_dt, step, status, message = _fix_zero_time_step(min_dt, step)
+                if status === :error
+                    # Unable to resolve, rethrow original error.
+                    throw(err)
+                end
+            else
+                throw(err)
+            end
+        end
+        isempty(message) || @warn message
+    end
+    cache = nothing
+    report = nothing
+    fitresult = (min_dt, step)
+    return fitresult, cache, report
+end
+
+function MLJBase.transform(model::UnivariateTimeTypeToContinuous, fitresult, X)
+    min_dt, step = fitresult
+    if typeof(min_dt) ≠ eltype(X)
+        # Cannot run if eltype in transform differs from zero_time from fit.
+        throw(ArgumentError("Different `TimeType` encountered during `transform` than expected from `fit`. Found `$(eltype(X))`, expected `$(typeof(min_dt))`"))
+    end
+    # Set the size of a single step.
+    next_time = min_dt + step
+    if next_time == min_dt
+        # Time type loops if step is a multiple of Hour(24), so calculate the
+        # number of multiples, then re-scale to Hour(12) and adjust delta to match original.
+        m = step / Dates.Hour(12)
+        delta = m * (
+            Float64(Dates.value(min_dt + Dates.Hour(12)) - Dates.value(min_dt)))
+    else
+        delta = Float64(Dates.value(min_dt + step) - Dates.value(min_dt))
+    end
+    return @. Float64(Dates.value(X - min_dt)) / delta
+end
+
+
 ## STANDARDIZATION OF ORDINAL FEATURES OF TABULAR DATA
 
 """
-    Standardizer(; features=Symbol[], ignore=false, ordered_factor=false, count=false)
+    Standardizer(; features=Symbol[],
+                   ignore=false,
+                   ordered_factor=false,
+                   count=false)
 
 Unsupervised model for standardizing (whitening) the columns of
 tabular data.  If `features` is unspecified then all columns
@@ -1096,7 +1242,8 @@ metadata_pkg.(
     (FeatureSelector, UnivariateStandardizer,
      UnivariateDiscretizer, Standardizer,
      UnivariateBoxCoxTransformer, UnivariateFillImputer,
-     OneHotEncoder, FillImputer, ContinuousEncoder),
+     OneHotEncoder, FillImputer, ContinuousEncoder,
+     UnivariateTimeTypeToContinuous),
     name       = "MLJModels",
     uuid       = "d491faf4-2d78-11e9-2867-c94bc002c0b7",
     url        = "https://github.com/alan-turing-institute/MLJModels.jl",
@@ -1170,4 +1317,9 @@ metadata_model(ContinuousEncoder,
     descr   = CONTINUOUS_ENCODER_DESCR,
     path    = "MLJModels.ContinuousEncoder")
 
-
+metadata_model(UnivariateTimeTypeToContinuous,
+    input   = AbstractVector{<:ScientificTypes.ScientificTimeType},
+    output  = AbstractVector{Continuous},
+    weights = false,
+    descr   = UNIVARIATE_TIME_TYPE_TO_CONTINUOUS,
+    path    = "MLJModels.UnivariateTimeTypeToContinuous")

src/registry/Metadata.toml

@@ -2,7 +2,7 @@
 [NearestNeighbors.KNNClassifier]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s105,1} where _s105<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s99,1} where _s99<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "NearestNeighbors"
 ":package_license" = "MIT"
@@ -1850,7 +1850,7 @@
 [NaiveBayes.GaussianNBClassifier]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s104,1} where _s104<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s173,1} where _s173<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "NaiveBayes"
 ":package_license" = "unknown"
@@ -1872,7 +1872,7 @@
 [NaiveBayes.MultinomialNBClassifier]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Count)`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s104,1} where _s104<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s173,1} where _s173<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "NaiveBayes"
 ":package_license" = "unknown"
@@ -2064,13 +2064,13 @@
 ":prediction_type" = ":unknown"
 ":implemented_methods" = [":clean!", ":fit", ":fitted_params", ":transform"]
 ":hyperparameters" = "`(:k, :alg, :fun, :do_whiten, :maxiter, :tol, :winit, :mean)`"
-":hyperparameter_types" = "`(\"Int64\", \"Symbol\", \"Symbol\", \"Bool\", \"Int64\", \"Real\", \"Union{Nothing, Array{_s197,2} where _s197<:Real}\", \"Union{Nothing, Array{Float64,1}, Real}\")`"
+":hyperparameter_types" = "`(\"Int64\", \"Symbol\", \"Symbol\", \"Bool\", \"Int64\", \"Real\", \"Union{Nothing, Array{_s148,2} where _s148<:Real}\", \"Union{Nothing, Array{Float64,1}, Real}\")`"
 ":hyperparameter_ranges" = "`(nothing, nothing, nothing, nothing, nothing, nothing, nothing, nothing)`"
 
 [MultivariateStats.BayesianLDA]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
-":target_scitype" = "`AbstractArray{_s226,1} where _s226<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s174,1} where _s174<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "MultivariateStats"
 ":package_license" = "MIT"
@@ -2092,7 +2092,7 @@
 [MultivariateStats.BayesianSubspaceLDA]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
-":target_scitype" = "`AbstractArray{_s226,1} where _s226<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s174,1} where _s174<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "MultivariateStats"
 ":package_license" = "MIT"
@@ -2158,7 +2158,7 @@
 [MultivariateStats.LDA]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
-":target_scitype" = "`AbstractArray{_s226,1} where _s226<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s174,1} where _s174<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "MultivariateStats"
 ":package_license" = "MIT"
@@ -2202,7 +2202,7 @@
 [MultivariateStats.SubspaceLDA]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
-":target_scitype" = "`AbstractArray{_s226,1} where _s226<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s174,1} where _s174<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "MultivariateStats"
 ":package_license" = "MIT"
@@ -2224,7 +2224,7 @@
 [DecisionTree.AdaBoostStumpClassifier]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:Union{AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous, AbstractArray{_s25,1} where _s25<:ScientificTypes.Count, AbstractArray{_s25,1} where _s25<:ScientificTypes.OrderedFactor}`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s226,1} where _s226<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s174,1} where _s174<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "DecisionTree"
 ":package_license" = "MIT"
@@ -2268,7 +2268,7 @@
 [DecisionTree.DecisionTreeClassifier]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:Union{AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous, AbstractArray{_s25,1} where _s25<:ScientificTypes.Count, AbstractArray{_s25,1} where _s25<:ScientificTypes.OrderedFactor}`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s226,1} where _s226<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s174,1} where _s174<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "DecisionTree"
 ":package_license" = "MIT"
@@ -2312,7 +2312,7 @@
 [DecisionTree.RandomForestClassifier]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:Union{AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous, AbstractArray{_s25,1} where _s25<:ScientificTypes.Count, AbstractArray{_s25,1} where _s25<:ScientificTypes.OrderedFactor}`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s226,1} where _s226<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s174,1} where _s174<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "DecisionTree"
 ":package_license" = "MIT"
@@ -2422,7 +2422,7 @@
 [XGBoost.XGBoostClassifier]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s492,1} where _s492<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s472,1} where _s472<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`false`"
 ":package_name" = "XGBoost"
 ":package_license" = "unknown"
@@ -2444,7 +2444,7 @@
 [LightGBM.LGBMClassifier]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s576,1} where _s576<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s565,1} where _s565<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`false`"
 ":package_name" = "LightGBM"
 ":package_license" = "MIT Expat"
@@ -2639,6 +2639,28 @@
 ":hyperparameter_types" = "``"
 ":hyperparameter_ranges" = "``"
 
+[MLJModels.UnivariateTimeTypeToContinuous]
+":input_scitype" = "`AbstractArray{_s55,1} where _s55<:ScientificTypes.ScientificTimeType`"
+":output_scitype" = "`AbstractArray{ScientificTypes.Continuous,1}`"
+":target_scitype" = "`ScientificTypes.Unknown`"
+":is_pure_julia" = "`true`"
+":package_name" = "MLJModels"
+":package_license" = "MIT"
+":load_path" = "MLJModels.UnivariateTimeTypeToContinuous"
+":package_uuid" = "d491faf4-2d78-11e9-2867-c94bc002c0b7"
+":package_url" = "https://github.com/alan-turing-institute/MLJModels.jl"
+":is_wrapper" = "`false`"
+":supports_weights" = "`false`"
+":supports_online" = "`false`"
+":docstring" = "Transform univariate data with element scitype `ScientificDateTime` so that it has `Continuous` element scitype, according to a learned scale. \n→ based on [MLJModels](https://github.com/alan-turing-institute/MLJModels.jl).\n→ do `@load UnivariateTimeTypeToContinuous pkg=\"MLJModels\"` to use the model.\n→ do `?UnivariateTimeTypeToContinuous` for documentation."
+":name" = "UnivariateTimeTypeToContinuous"
+":is_supervised" = "`false`"
+":prediction_type" = ":unknown"
+":implemented_methods" = [":clean!", ":fit", ":transform"]
+":hyperparameters" = "`(:zero_time, :step)`"
+":hyperparameter_types" = "`(\"Union{Nothing, Dates.TimeType}\", \"Dates.Period\")`"
+":hyperparameter_ranges" = "`(nothing, nothing)`"
+
 [MLJModels.OneHotEncoder]
 ":input_scitype" = "`ScientificTypes.Table`"
 ":output_scitype" = "`ScientificTypes.Table`"
@@ -2884,7 +2906,7 @@
 [LIBSVM.LinearSVC]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s575,1} where _s575<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s564,1} where _s564<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`false`"
 ":package_name" = "LIBSVM"
 ":package_license" = "unknown"
@@ -2928,7 +2950,7 @@
 [LIBSVM.NuSVC]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s575,1} where _s575<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s564,1} where _s564<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`false`"
 ":package_name" = "LIBSVM"
 ":package_license" = "unknown"
@@ -2950,7 +2972,7 @@
 [LIBSVM.SVC]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s575,1} where _s575<:ScientificTypes.Finite`"
+":target_scitype" = "`AbstractArray{_s564,1} where _s564<:ScientificTypes.Finite`"
 ":is_pure_julia" = "`false`"
 ":package_name" = "LIBSVM"
 ":package_license" = "unknown"
@@ -2971,7 +2993,7 @@
 
 [LIBSVM.OneClassSVM]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
-":output_scitype" = "`AbstractArray{_s575,1} where _s575<:ScientificTypes.Finite{2}`"
+":output_scitype" = "`AbstractArray{_s564,1} where _s564<:ScientificTypes.Finite{2}`"
 ":target_scitype" = "`ScientificTypes.Unknown`"
 ":is_pure_julia" = "`false`"
 ":package_name" = "LIBSVM"
@@ -2994,7 +3016,7 @@
 [GLM.LinearBinaryClassifier]
 ":input_scitype" = "`ScientificTypes.Table{_s23} where _s23<:(AbstractArray{_s25,1} where _s25<:ScientificTypes.Continuous)`"
 ":output_scitype" = "`ScientificTypes.Unknown`"
-":target_scitype" = "`AbstractArray{_s576,1} where _s576<:ScientificTypes.Finite{2}`"
+":target_scitype" = "`AbstractArray{_s565,1} where _s565<:ScientificTypes.Finite{2}`"
 ":is_pure_julia" = "`true`"
 ":package_name" = "GLM"
 ":package_license" = "MIT"