Merge pull request #26 from JuliaAI/add-callable-and-better-errors

EssamWisam · web-flow · commit 08d973ff6668 · 2025-05-26T23:07:10.000-05:00
✨ Add callable features and better error testing
diff --git a/.github/workflows/CI.yml b/.github/workflows/CI.yml
@@ -24,7 +24,6 @@ jobs:
       matrix:
         version:
           - '1.10'
-          - '1.6'
           - 'nightly'
         os:
           - ubuntu-latest
diff --git a/Project.toml b/Project.toml
@@ -26,7 +26,7 @@ ScientificTypes = "3.0"
 StatsBase = "0.34"
 TableOperations = "1.2"
 Tables = "1.11"
-julia = "1.6.7"
+julia = "1.10"
 
 [extras]
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
diff --git a/src/encoders/contrast_encoder/contrast_encoder.jl b/src/encoders/contrast_encoder/contrast_encoder.jl
@@ -79,7 +79,7 @@ Fit a contrast encoing scheme on given data in `X`.
 """
 function contrast_encoder_fit(
     X,
-    features::AbstractVector{Symbol} = Symbol[];
+    features = Symbol[];
     mode::Union{Symbol, AbstractVector{Symbol}} = :dummy,
     buildmatrix = nothing,
     ignore::Bool = true,
diff --git a/src/encoders/contrast_encoder/interface_mlj.jl b/src/encoders/contrast_encoder/interface_mlj.jl
@@ -1,11 +1,11 @@
 ### ContrastEncoding with MLJ Interface
 
 # 1. Interface Struct
-mutable struct ContrastEncoder{AS <: AbstractVector{Symbol}} <: Unsupervised
-    features::AS
+mutable struct ContrastEncoder{ASS <: Union{Symbol, AbstractVector{Symbol}}, A1 <: Any, A2 <: Any} <: Unsupervised
+    features::A1
     ignore::Bool
-    mode::Union{Symbol, AS}
-    buildmatrix::Any
+    mode:: ASS
+    buildmatrix::A2
     ordered_factor::Bool
 end;
 
diff --git a/src/encoders/frequency_encoding/frequency_encoding.jl b/src/encoders/frequency_encoding/frequency_encoding.jl
@@ -20,7 +20,7 @@ categorical features with their (normalized or raw) frequencies of occurrence in
 """
 function frequency_encoder_fit(
     X,
-    features::AbstractVector{Symbol} = Symbol[];
+    features = Symbol[];
     ignore::Bool = true,
     ordered_factor::Bool = false,
     normalize::Bool = false,
diff --git a/src/encoders/frequency_encoding/interface_mlj.jl b/src/encoders/frequency_encoding/interface_mlj.jl
@@ -1,8 +1,8 @@
 ### FrequencyEncoding with MLJ Interface
 
 # 1. Interface Struct
-mutable struct FrequencyEncoder{AS <: AbstractVector{Symbol}} <: Unsupervised
-    features::AS
+mutable struct FrequencyEncoder{A <: Any} <: Unsupervised
+    features::A
     ignore::Bool
     ordered_factor::Bool
     normalize::Bool
diff --git a/src/encoders/missingness_encoding/interface_mlj.jl b/src/encoders/missingness_encoding/interface_mlj.jl
@@ -2,11 +2,10 @@
 
 # 1. Interface Struct
 mutable struct MissingnessEncoder{
-    AS <: AbstractVector{Symbol},
     T <: Type,
     A <: Any,
 } <: Unsupervised
-    features::AS
+    features::A
     ignore::Bool
     ordered_factor::Bool
     label_for_missing::Dict{T, A}
diff --git a/src/encoders/missingness_encoding/missingness_encoding.jl b/src/encoders/missingness_encoding/missingness_encoding.jl
@@ -27,7 +27,7 @@ types that are in `Char`, `AbstractString`, and `Number`.
 """
 function missingness_encoder_fit(
     X,
-    features::AbstractVector{Symbol} = Symbol[];
+    features = Symbol[];
     ignore::Bool = true,
     ordered_factor::Bool = false,
     label_for_missing::Dict{<:Type, <:Any} = Dict(
diff --git a/src/encoders/ordinal_encoding/interface_mlj.jl b/src/encoders/ordinal_encoding/interface_mlj.jl
@@ -1,8 +1,8 @@
 ### OrdinalEncoding with MLJ Interface
 
 # 1. Interface Struct
-mutable struct OrdinalEncoder{AS <: AbstractVector{Symbol}} <: Unsupervised
-    features::AS
+mutable struct OrdinalEncoder{A <: Any} <: Unsupervised
+    features::A
     ignore::Bool
     ordered_factor::Bool
     output_type::Type
diff --git a/src/encoders/ordinal_encoding/ordinal_encoding.jl b/src/encoders/ordinal_encoding/ordinal_encoding.jl
@@ -18,7 +18,7 @@ Fit an encoder to encode the levels of categorical variables in a given table as
 """
 function ordinal_encoder_fit(
     X,
-    features::AbstractVector{Symbol} = Symbol[];
+    features = Symbol[];
     ignore::Bool = true,
     ordered_factor::Bool = false,
     output_type::Type = Float32,
diff --git a/src/encoders/target_encoding/interface_mlj.jl b/src/encoders/target_encoding/interface_mlj.jl
@@ -1,9 +1,9 @@
 ### TargetEncoding with MLJ Interface
 
 # 1. Interface Struct
-mutable struct TargetEncoder{R1 <: Real, R2 <: Real, AS <: AbstractVector{Symbol}} <:
+mutable struct TargetEncoder{R1 <: Real, R2 <: Real, A <: Any} <:
                Unsupervised
-    features::AS
+    features::A
     ignore::Bool
     ordered_factor::Bool
     lambda::R1
@@ -45,7 +45,7 @@ end
 struct TargetEncoderResult{
     I <: Integer,
     S <: AbstractString,
-    A <: Any            # Useless but likely can't do much better
+    A <: Any,            # Useless but likely can't do much better
 } <: MMI.MLJType
     # target statistic for each level of each categorical feature
     y_stat_given_feat_level::Dict{A, A}
diff --git a/src/encoders/target_encoding/target_encoding.jl b/src/encoders/target_encoding/target_encoding.jl
@@ -132,7 +132,7 @@ Fit a target encoder on table X with target y by computing the necessary statist
 function target_encoder_fit(
     X,
     y::AbstractVector,
-    features::AbstractVector{Symbol} = Symbol[];
+    features = Symbol[];
     ignore::Bool = true,
     ordered_factor::Bool = false,
     lambda::Real = 1.0,
diff --git a/src/generic.jl b/src/generic.jl
@@ -13,21 +13,23 @@ logic?"
 
 # Arguments
 
-    - `X`: A table where the elements of the categorical features have [scitypes](https://juliaai.github.io/ScientificTypes.jl/dev/) 
+    - X: A table where the elements of the categorical features have [scitypes](https://juliaai.github.io/ScientificTypes.jl/dev/) 
     `Multiclass` or `OrderedFactor`
-    - `features=[]`: A list of names of categorical features given as symbols to exclude or include from encoding
-    - `ignore=true`: Whether to exclude or includes the features given in `features`
-    - `ordered_factor=false`: Whether to encode `OrderedFactor` or ignore them
-    - `feature_mapper`: Defined above. 
+    - features=[]: A list of names of categorical features given as symbols to exclude or include from encoding,
+      according to the value of `ignore`, or a single symbol (which is treated as a vector with one symbol),
+      or a callable that returns true for features to be included/excluded
+    - ignore=true: Whether to exclude or includes the features given in features
+    - ordered_factor=false: Whether to encode OrderedFactor or ignore them
+    - feature_mapper: Defined above. 
 
 # Returns
 
-    - `mapping_per_feat_level`: Maps each level for each feature in a subset of the categorical features of
+    - mapping_per_feat_level: Maps each level for each feature in a subset of the categorical features of
      X into a scalar or a vector. 
-    - `encoded_features`: The subset of the categorical features of X that were encoded
+    - encoded_features: The subset of the categorical features of X that were encoded
 """
 function generic_fit(X,
-    features::AbstractVector{Symbol} = Symbol[],
+    features = Symbol[],
     args...;
     ignore::Bool = true,
     ordered_factor::Bool = false,
@@ -38,7 +40,22 @@ function generic_fit(X,
     feat_names = Tables.schema(X).names
 
     #2.  Modify column_names based on features 
-    feat_names = (ignore) ? setdiff(feat_names, features) : intersect(feat_names, features)
+    if features isa Symbol
+        features = [features]
+    end
+    
+    if features isa AbstractVector{Symbol}
+        # Original behavior for vector of symbols
+        feat_names =
+            (ignore) ? setdiff(feat_names, features) : intersect(feat_names, features)
+    else
+        # If features is a callable, apply it to each feature name
+        if ignore
+            feat_names = filter(name -> !features(name), feat_names)
+        else
+            feat_names = filter(features, feat_names)
+        end
+    end
 
     # 3. Define mapping per column per level dictionary
     mapping_per_feat_level = Dict()
diff --git a/src/transformers/cardinality_reducer/cardinality_reducer.jl b/src/transformers/cardinality_reducer/cardinality_reducer.jl
@@ -32,7 +32,7 @@ types that are in `Char`, `AbstractString`, and `Number`.
 """
 function cardinality_reducer_fit(
     X,
-    features::AbstractVector{Symbol} = Symbol[];
+    features = Symbol[];
     ignore::Bool = true,
     ordered_factor::Bool = false,
     min_frequency::Real = 3,
diff --git a/src/transformers/cardinality_reducer/interface_mlj.jl b/src/transformers/cardinality_reducer/interface_mlj.jl
@@ -2,12 +2,11 @@
 
 # 1. Interface Struct
 mutable struct CardinalityReducer{
-    AS <: AbstractVector{Symbol},
     R <: Real,
     T <: Type,
     A <: Any,
 } <: Unsupervised
-    features::AS
+    features::A
     ignore::Bool
     ordered_factor::Bool
     min_frequency::R
diff --git a/test/encoders/contrast_encoder.jl b/test/encoders/contrast_encoder.jl
@@ -9,7 +9,6 @@ age    = [23, 23, 14, 23])
 
 
 @testset "Contrast Encoder Error Handling" begin
-
     # Example definitions to allow the test to run
     function dummy_buildmatrix(colname, k)
         # Simple dummy function to generate a matrix of correct size
@@ -23,21 +22,35 @@ age    = [23, 23, 14, 23])
     )
 
     # Test IGNORE_MUST_FALSE_VEC_MODE error
-    @test_throws ArgumentError contrast_encoder_fit(data, [:A], mode=[:contrast], ignore=true)
+    @test_throws MLJTransforms.IGNORE_MUST_FALSE_VEC_MODE begin
+        contrast_encoder_fit(data, [:A], mode=[:contrast], ignore=true)
+    end
 
     # Test LENGTH_MISMATCH_VEC_MODE error
-    @test_throws ArgumentError contrast_encoder_fit(data, [:A], mode=[:contrast, :dummy], buildmatrix=dummy_buildmatrix, ignore=false)
+    @test_throws MLJTransforms.LENGTH_MISMATCH_VEC_MODE(2, 1) begin
+        contrast_encoder_fit(data, [:A], mode=[:contrast, :dummy], buildmatrix=dummy_buildmatrix, ignore=false)
+    end
 
     # Test BUILDFUNC_MUST_BE_SPECIFIED error
-    @test_throws ArgumentError contrast_encoder_fit(data, [:A], mode=:contrast, ignore=false)
+    @test_throws MLJTransforms.BUILDFUNC_MUST_BE_SPECIFIED begin
+        contrast_encoder_fit(data, [:A], mode=:contrast, ignore=false)
+    end
 
     # Test MATRIX_SIZE_ERROR
     wrong_buildmatrix = (levels, k) -> randn(k, k)  # Incorrect dimensions
-    @test_throws ArgumentError contrast_encoder_fit(data, [:A], mode=:contrast, buildmatrix=wrong_buildmatrix, ignore=false)
+    k = 3  # Number of levels in data[:A]
+    wrong_size = (k, k)
+    @test_throws MLJTransforms.MATRIX_SIZE_ERROR(k, wrong_size, :A) begin
+        contrast_encoder_fit(data, [:A], mode=:contrast, buildmatrix=wrong_buildmatrix, ignore=false)
+    end
 
     # Test MATRIX_SIZE_ERROR_HYP
     wrong_buildmatrix_hyp = (levels, k) -> randn(k, k+1)  # Incorrect dimensions for hypothesis matrix
-    @test_throws ArgumentError contrast_encoder_fit(data, [:A], mode=:hypothesis, buildmatrix=wrong_buildmatrix_hyp, ignore=false)
+    wrong_size_hyp = (k, k+1)
+    @test_throws MLJTransforms.MATRIX_SIZE_ERROR_HYP(k, wrong_size_hyp, :A) begin
+        contrast_encoder_fit(data, [:A], mode=:hypothesis, buildmatrix=wrong_buildmatrix_hyp, ignore=false)
+    end
+    
 end
 
 @testset "Dummy Coding Tests" begin
diff --git a/test/encoders/missingness_encoding.jl b/test/encoders/missingness_encoding.jl
@@ -1,21 +1,26 @@
 using MLJTransforms: missingness_encoder_fit, missingness_encoder_transform
 
-@testset "Throws errors when needed" begin
-    @test_throws ArgumentError begin
+@testset "Missingness Encoder Error Handling" begin
+    # Test COLLISION_NEW_VAL_ME error - when label_for_missing value already exists in levels
+    @test_throws MLJTransforms.COLLISION_NEW_VAL_ME("missing") begin
         X = generate_X_with_missingness(;john_name="missing")
         cache = missingness_encoder_fit(
             X;
             label_for_missing = Dict(AbstractString => "missing", Char => 'm'),
         )
     end
-    @test_throws ArgumentError begin
+    
+    # Test VALID_TYPES_NEW_VAL_ME error - when label_for_missing key is not a supported type
+    @test_throws MLJTransforms.VALID_TYPES_NEW_VAL_ME(Bool) begin
         X = generate_X_with_missingness()
         cache = missingness_encoder_fit(
             X;
             label_for_missing = Dict(AbstractString => "Other", Bool => 'X'),
         )
     end
-    @test_throws ArgumentError begin
+    
+    # Test UNSPECIFIED_COL_TYPE_ME error - when column type isn't in label_for_missing
+    @test_throws MLJTransforms.UNSPECIFIED_COL_TYPE_ME(Char, Dict(AbstractString => "X")) begin
         X = generate_X_with_missingness()
         cache = missingness_encoder_fit(
             X;
diff --git a/test/encoders/target_encoding.jl b/test/encoders/target_encoding.jl
@@ -332,13 +332,15 @@ end
         @test fitresult.task == generic_cache[:task]
 
         # Test invalid `m`
-        @test_throws ArgumentError begin
-            t = TargetEncoder(ignore = true, ordered_factor = false, lambda = 0.5, m = -5)
+        invalid_m = -5
+        @test_throws MLJTransforms.NON_NEGATIVE_m(invalid_m) begin
+            t = TargetEncoder(ignore = true, ordered_factor = false, lambda = 0.5, m = invalid_m)
         end
-
-        # Test invalid `lambda`
-        @test_throws ArgumentError begin
-            t = TargetEncoder(ignore = true, ordered_factor = false, lambda = 1.1, m = 1)
+        
+        # Test invalid `lambda` (value > 1)
+        invalid_lambda = 1.1
+        @test_throws MLJTransforms.INVALID_lambda(invalid_lambda) begin
+            t = TargetEncoder(ignore = true, ordered_factor = false, lambda = invalid_lambda, m = 1)
         end
 
         # Test report
diff --git a/test/generic.jl b/test/generic.jl
@@ -62,7 +62,7 @@ end
 # Dummy encoder that maps each level to its hash (some arbitrary function)
 function dummy_encoder_fit(
     X,
-    features::AbstractVector{Symbol} = Symbol[];
+    features = Symbol[];
     ignore::Bool = true,
     ordered_factor::Bool = false,
 )
@@ -81,6 +81,7 @@ function dummy_encoder_fit(
     )
     cache = Dict(
         :hash_given_feat_val => hash_given_feat_val,
+        :encoded => encoded_features,
     )
     return cache
 end
@@ -161,4 +162,40 @@ end
         F = [enc(:F, X[:F][i]) for i in 1:10]
     )
     @test X_tr == target
+end
+
+@testset "Callable feature functionality tests" begin
+    X = dataset_forms[1]
+    feat_names = Tables.schema(X).names
+
+    # Define a predicate: include only columns with name in uppercase list [:A, :C, :E]
+    predicate = name -> name in [:A, :C, :E]
+
+    # Test 1: ignore=true should exclude predicate columns
+    cache1 = dummy_encoder_fit(X, predicate; ignore=true, ordered_factor=false)
+    @test !(:A in cache1[:encoded]) && !(:C in cache1[:encoded]) && !(:E in cache1[:encoded])
+
+    # Test 2: ignore=false should include only predicate columns
+    cache2 = dummy_encoder_fit(X, predicate; ignore=false, ordered_factor=false)
+    @test Set(cache2[:encoded]) == Set([:A, :C])
+
+    # Test 3: predicate with ordered_factor=true picks up ordered factors (e.g., :E)
+    cache3 = dummy_encoder_fit(X, predicate; ignore=false, ordered_factor=true)
+    @test Set(cache3[:encoded]) == Set([:A, :C, :E])
+end
+
+@testset "Single Symbol and list of one symbol equivalence" begin
+     X = dataset_forms[1]
+    feat_names = Tables.schema(X).names
+
+    # Test 1: Single Symbol
+    single_symbol = :A
+    cache1 = dummy_encoder_fit(X, single_symbol; ignore=true, ordered_factor=false)
+    @test !(:A in cache1[:encoded])
+    # Test 2: List of one symbol
+    single_symbol_list = [:A]
+    cache2 = dummy_encoder_fit(X, single_symbol_list; ignore=true, ordered_factor=false)
+    @test !(:A in cache2[:encoded])
+    # Test 3: Both should yield the same result
+    @test cache1[:encoded] == cache2[:encoded]
 end
diff --git a/test/transformers/cardinality_reducer.jl b/test/transformers/cardinality_reducer.jl
