Merge pull request #2 from MariaHei/deep_learning

Add support for DDL models

.github/workflows/CI.yml

@@ -15,6 +15,8 @@ jobs:
         version:
           - '1.7'
           - '1.8'
+          - '1.9'
+          - '1.10'
         os:
           - ubuntu-latest
           - macOS-latest

Project.toml

@@ -7,7 +7,6 @@ version = "0.1.0"
 Conda = "8f4d0f93-b110-5947-807f-2305c1781a2d"
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 Distances = "b4f34e82-e78d-54a5-968a-f98e89d6e8f7"
-Formatting = "59287772-0a20-5a39-b81b-1366585eb4c0"
 JudiLing = "b43a184b-0e9d-488b-813a-80fd5dbc9fd8"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"

README.md

@@ -7,8 +7,6 @@ This is code for JudiLingMeasures. Most measures are based on R implementations
 
 You can find the documentation [here](https://mariahei.github.io/JudiLingMeasures.jl/dev/index.html).
 
-PLEASE NOTE THAT THIS PACKAGE IS WORK IN PROGRESS. MAJOR CHANGES TO THE CODE ARE POSSIBLE AT ANY POINT AND NEW MEASURES ARE STILL BEING ADDED.
-
 ## Installation
 
 ```

src/helpers.jl

@@ -81,7 +81,7 @@ julia> sem_density_mean(cor_s, 2)
 function sem_density_mean(s_cor::Union{JudiLing.SparseMatrixCSC, Matrix},
                           n::Int)
     if n > size(s_cor,2)
-        throw(MethodError("n larger than the dimension of the semantic vectors"))
+        throw(ArgumentError("n larger than the dimension of the semantic vectors"))
     end
     sems = Vector{Union{Missing, Float32}}(missing, size(s_cor,1))
     for i in 1:size(s_cor)[1]
@@ -477,8 +477,8 @@ function compute_all_measures_train(data_train::DataFrame,
                                     Chat_train::Union{JudiLing.SparseMatrixCSC, Matrix},
                                     S_train::Union{JudiLing.SparseMatrixCSC, Matrix},
                                     Shat_train::Union{JudiLing.SparseMatrixCSC, Matrix},
-                                    F_train::Union{JudiLing.SparseMatrixCSC, Matrix},
-                                    G_train::Union{JudiLing.SparseMatrixCSC, Matrix};
+                                    F_train::Union{JudiLing.SparseMatrixCSC, Matrix, Missing},
+                                    G_train::Union{JudiLing.SparseMatrixCSC, Matrix, Missing};
                                     res_learn_train::Union{Array{Array{JudiLing.Result_Path_Info_Struct,1},1}, Missing}=missing,
                                     gpi_learn_train::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
                                     rpi_learn_train::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
@@ -514,7 +514,7 @@ function compute_all_measures_train(data_train::DataFrame,
     results[!,"EDNN"] = EDNN(Shat_train, S_train)
     results[!,"NNC"] = NNC(cor_s)
 
-    if !low_cost_measures_only
+    if !low_cost_measures_only && !ismissing(F_train)
         results[!,"DistanceTravelledF"] = total_distance(cue_obj_train, F_train, :F)
     end
 
@@ -531,7 +531,7 @@ function compute_all_measures_train(data_train::DataFrame,
     if calculate_production_uncertainty && !low_cost_measures_only
         results[!,"ProductionUncertainty"] = vec(uncertainty(cue_obj_train.C, Chat_train, method="cosine"))
     end
-    if !low_cost_measures_only
+    if !low_cost_measures_only && !ismissing(G_train)
         results[!,"DistanceTravelledG"] = total_distance(cue_obj_train, G_train, :G)
     end
 
@@ -567,6 +567,61 @@ function compute_all_measures_train(data_train::DataFrame,
     results
 end
 
+"""
+    function compute_all_measures_train(data_train::DataFrame,
+                                        cue_obj_train::JudiLing.Cue_Matrix_Struct,
+                                        Chat_train::Union{JudiLing.SparseMatrixCSC, Matrix},
+                                        S_train::Union{JudiLing.SparseMatrixCSC, Matrix},
+                                        Shat_train::Union{JudiLing.SparseMatrixCSC, Matrix};
+                                        res_learn_train::Union{Array{Array{JudiLing.Result_Path_Info_Struct,1},1}, Missing}=missing,
+                                        gpi_learn_train::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
+                                        rpi_learn_train::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
+                                        sem_density_n::Int64=8,
+                                        calculate_production_uncertainty::Bool=false,
+                                        low_cost_measures_only::Bool=false)
+Compute all measures currently available in JudiLingMeasures for the training data if F and G are not available (usually for DDL models).
+# Arguments
+- `data_train::DataFrame`: The data for which measures should be calculated (the training data).
+- `cue_obj_train::JudiLing.Cue_Matrix_Struct`: The cue object of the training data.
+- `Chat_train::Union{JudiLing.SparseMatrixCSC, Matrix}`: The Chat matrix of the training data.
+- `S_train::Union{JudiLing.SparseMatrixCSC, Matrix}`: The S matrix of the training data.
+- `Shat_train::Union{JudiLing.SparseMatrixCSC, Matrix}`: The Shat matrix of the training data.
+- `res_learn_train::Union{Array{Array{JudiLing.Result_Path_Info_Struct,1},1}, Missing}=missing`: The first output of JudiLing.learn_paths_rpi (with `check_gold_path=true`)
+- `gpi_learn_train::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing`: The second output of JudiLing.learn_paths_rpi (with `check_gold_path=true`)
+- `rpi_learn_train::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing`: The third output of JudiLing.learn_paths_rpi (with `check_gold_path=true`)
+- `sem_density_n::Int64=8`: Number of neighbours to take into account in Semantic Density measure.
+- `calculate_production_uncertainty`: "Production Uncertainty" is computationally very heavy for large C matrices, therefore its computation is turned off by default.
+- `low_cost_measures_only::Bool=false`: Only compute measures which are not computationally heavy. Recommended for very large datasets.
+# Returns
+- `results::DataFrame`: A dataframe with all information in `data_train` plus all the computed measures.
+"""
+function compute_all_measures_train(data_train::DataFrame,
+                                    cue_obj_train::JudiLing.Cue_Matrix_Struct,
+                                    Chat_train::Union{JudiLing.SparseMatrixCSC, Matrix},
+                                    S_train::Union{JudiLing.SparseMatrixCSC, Matrix},
+                                    Shat_train::Union{JudiLing.SparseMatrixCSC, Matrix};
+                                    res_learn_train::Union{Array{Array{JudiLing.Result_Path_Info_Struct,1},1}, Missing}=missing,
+                                    gpi_learn_train::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
+                                    rpi_learn_train::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
+                                    sem_density_n::Int64=8,
+                                    calculate_production_uncertainty::Bool=false,
+                                    low_cost_measures_only::Bool=false)
+
+    compute_all_measures_train(data_train,
+                                cue_obj_train,
+                                Chat_train,
+                                S_train,
+                                Shat_train,
+                                missing,
+                                missing;
+                                res_learn_train=res_learn_train,
+                                gpi_learn_train=gpi_learn_train,
+                                rpi_learn_train=rpi_learn_train,
+                                sem_density_n=sem_density_n,
+                                calculate_production_uncertainty=calculate_production_uncertainty,
+                                low_cost_measures_only=low_cost_measures_only)
+end
+
 """
     function compute_all_measures_val(data_val::DataFrame,
                                       cue_obj_train::JudiLing.Cue_Matrix_Struct,
@@ -608,8 +663,8 @@ function compute_all_measures_val(data_val::DataFrame,
                                   S_train::Union{JudiLing.SparseMatrixCSC, Matrix},
                                   S_val::Union{JudiLing.SparseMatrixCSC, Matrix},
                                   Shat_val::Union{JudiLing.SparseMatrixCSC, Matrix},
-                                  F_train::Union{JudiLing.SparseMatrixCSC, Matrix},
-                                  G_train::Union{JudiLing.SparseMatrixCSC, Matrix};
+                                  F_train::Union{JudiLing.SparseMatrixCSC, Matrix, Missing},
+                                  G_train::Union{JudiLing.SparseMatrixCSC, Matrix, Missing};
                                   res_learn_val::Union{Array{Array{JudiLing.Result_Path_Info_Struct,1},1}, Missing}=missing,
                                   gpi_learn_val::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
                                   rpi_learn_val::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
@@ -644,7 +699,7 @@ function compute_all_measures_val(data_val::DataFrame,
     results[!,"ALC"] = JudiLingMeasures.ALC(cor_s)
     results[!,"EDNN"] = EDNN(Shat_val, S_val, S_train)
     results[!,"NNC"] = JudiLingMeasures.NNC(cor_s)
-    if !low_cost_measures_only
+    if !low_cost_measures_only && !ismissing(F_train)
         results[!,"DistanceTravelledF"] = total_distance(cue_obj_val, F_train, :F)
     end
 
@@ -660,7 +715,7 @@ function compute_all_measures_val(data_val::DataFrame,
     if calculate_production_uncertainty && !low_cost_measures_only
         results[!,"ProductionUncertainty"] = vec(JudiLingMeasures.uncertainty(cue_obj_val.C, Chat_val, cue_obj_train.C, method="cosine"))
     end
-    if !low_cost_measures_only
+    if !low_cost_measures_only && !ismissing(G_train)
         results[!,"DistanceTravelledG"] = JudiLingMeasures.total_distance(cue_obj_val, G_train, :G)
     end
 
@@ -696,6 +751,68 @@ function compute_all_measures_val(data_val::DataFrame,
     results
 end
 
+"""
+    function compute_all_measures_val(data_val::DataFrame,
+                                      cue_obj_train::JudiLing.Cue_Matrix_Struct,
+                                      cue_obj_val::JudiLing.Cue_Matrix_Struct,
+                                      Chat_val::Union{JudiLing.SparseMatrixCSC, Matrix},
+                                      S_train::Union{JudiLing.SparseMatrixCSC, Matrix},
+                                      S_val::Union{JudiLing.SparseMatrixCSC, Matrix},
+                                      Shat_val::Union{JudiLing.SparseMatrixCSC, Matrix};
+                                      res_learn_val::Union{Array{Array{JudiLing.Result_Path_Info_Struct,1},1}, Missing}=missing,
+                                      gpi_learn_val::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
+                                      rpi_learn_val::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
+                                      sem_density_n::Int64=8,
+                                      calculate_production_uncertainty::Bool=false,
+                                      low_cost_measures_only::Bool=false)
+Compute all measures currently available in JudiLingMeasures for the validation data if F and G are not available (usually for DDL models).
+# Arguments
+- `data_val::DataFrame`: The data for which measures should be calculated (the validation data).
+- `cue_obj_train::JudiLing.Cue_Matrix_Struct`: The cue object of the training data.
+- `cue_obj_val::JudiLing.Cue_Matrix_Struct`: The cue object of the validation data.
+- `Chat_val::Union{JudiLing.SparseMatrixCSC, Matrix}`: The Chat matrix of the validation data.
+- `S_train::Union{JudiLing.SparseMatrixCSC, Matrix}`: The S matrix of the training data.
+- `S_val::Union{JudiLing.SparseMatrixCSC, Matrix}`: The S matrix of the validation data.
+- `Shat_val::Union{JudiLing.SparseMatrixCSC, Matrix}`: The Shat matrix of the data of interest.
+- `res_learn_val::Union{Array{Array{JudiLing.Result_Path_Info_Struct,1},1}, Missing}=missing`: The first output of JudiLing.learn_paths_rpi (with `check_gold_path=true`)
+- `gpi_learn_val::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing`: The second output of JudiLing.learn_paths_rpi (with `check_gold_path=true`)
+- `rpi_learn_val::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing`: The third output of JudiLing.learn_paths_rpi (with `check_gold_path=true`)
+- `low_cost_measures_only::Bool=false`: Only compute measures which are not computationally heavy. Recommended for very large datasets.
+# Returns
+- `results::DataFrame`: A dataframe with all information in `data_val` plus all the computed measures.
+"""
+function compute_all_measures_val(data_val::DataFrame,
+                                  cue_obj_train::JudiLing.Cue_Matrix_Struct,
+                                  cue_obj_val::JudiLing.Cue_Matrix_Struct,
+                                  Chat_val::Union{JudiLing.SparseMatrixCSC, Matrix},
+                                  S_train::Union{JudiLing.SparseMatrixCSC, Matrix},
+                                  S_val::Union{JudiLing.SparseMatrixCSC, Matrix},
+                                  Shat_val::Union{JudiLing.SparseMatrixCSC, Matrix};
+                                  res_learn_val::Union{Array{Array{JudiLing.Result_Path_Info_Struct,1},1}, Missing}=missing,
+                                  gpi_learn_val::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
+                                  rpi_learn_val::Union{Array{JudiLing.Gold_Path_Info_Struct,1}, Missing}=missing,
+                                  sem_density_n::Int64=8,
+                                  calculate_production_uncertainty::Bool=false,
+                                  low_cost_measures_only::Bool=false)
+
+  compute_all_measures_val(data_val,
+                                    cue_obj_train,
+                                    cue_obj_val,
+                                    Chat_val,
+                                    S_train,
+                                    S_val,
+                                    Shat_val,
+                                    missing,
+                                    missing;
+                                    res_learn_val=res_learn_val,
+                                    gpi_learn_val=gpi_learn_val,
+                                    rpi_learn_val=rpi_learn_val,
+                                    sem_density_n=sem_density_n,
+                                    calculate_production_uncertainty=calculate_production_uncertainty,
+                                    low_cost_measures_only=low_cost_measures_only)
+end
+
+
 function safe_divide(x, y)
     if (y != 0) & (!ismissing(y)) & (!ismissing(x))
         x/y

test/test_helpers.jl

@@ -45,7 +45,7 @@ end
     cs = JudiLingMeasures.correlation_rowwise([1. 2. 3.], [5. 1. 19.])
     @test isapprox(vec(JudiLingMeasures.sem_density_mean(cs,1)),
                    vec([0.7406128966515281]))
-    @test_throws MethodError JudiLingMeasures.sem_density_mean(cs,5)
+    @test_throws ArgumentError JudiLingMeasures.sem_density_mean(cs,5)
     cs = JudiLingMeasures.correlation_rowwise(ma2, ma3)
     @test isapprox(vec(JudiLingMeasures.sem_density_mean(cs, 3)),
                    vec([0.550947 0.28884766666666667 0.19702316666666667 0.15713063333333335]), rtol=1e-4)

test/test_measures.jl

@@ -134,7 +134,7 @@ cor_s2 = JudiLingMeasures.correlation_rowwise(ma2, ma5)
         @test JudiLingMeasures.density(zeros((1,1)), n=1) == [0]
         @test JudiLingMeasures.density(ones((1,1)), n=1) == [1]
         @test isequal(JudiLingMeasures.density([[1 2 missing]; [-1 -2 -3]; [1 2 3]], n=2), [missing; -1.5; 2.5])
-        @test_throws MethodError JudiLingMeasures.density(zeros((1,1))) == [0]
+        @test_throws ArgumentError JudiLingMeasures.density(zeros((1,1))) == [0]
     end
     @testset "Validation data" begin
         @test isapprox(JudiLingMeasures.density(cor_s2, n=2), vec([0.7393784999999999 0.6420815 0.44968675 0.2811505]), rtol=1e-4)
@@ -644,6 +644,16 @@ end
            @test "DistanceTravelledF" in names(all_measures)
            @test "DistanceTravelledG" in names(all_measures)
            @test !("WithinPathEntropies" in names(all_measures))
+
+           all_measures =  JudiLingMeasures.compute_all_measures_train(dat, # the data of interest
+                                                                cue_obj, # the cue_obj of the training data
+                                                                Chat, # the Chat of the data of interest
+                                                                S, # the S matrix of the data of interest
+                                                                Shat, # the Shat matrix of the data of interest
+                                                                sem_density_n=2)
+            @test all_measures != 1
+            @test !("DistanceTravelledF" in names(all_measures))
+            @test !("DistanceTravelledG" in names(all_measures))
     end
     @testset "Validation data" begin
         # just make sure that this function runs without error
@@ -710,5 +720,17 @@ end
            @test "DistanceTravelledF" in names(all_measures)
            @test "DistanceTravelledG" in names(all_measures)
            @test !("WithinPathEntropies" in names(all_measures))
+
+           all_measures =  JudiLingMeasures.compute_all_measures_val(val_dat, # the data of interest
+                                                                cue_obj, # the cue_obj of the training data
+                                                                cue_obj_val,
+                                                                Chat_val, # the Chat of the data of interest
+                                                                S, # the S matrix of the data of interest
+                                                                S_val,
+                                                                Shat_val,
+                                                                sem_density_n=2)
+            @test all_measures != 1
+            @test !("DistanceTravelledF" in names(all_measures))
+            @test !("DistanceTravelledG" in names(all_measures))
     end
 end