From d62819d43d8c7d3f8ce9cd40c838a69c14bcde26 Mon Sep 17 00:00:00 2001
From: jw2249a <github@jackryanwilliams.com>
Date: Mon, 13 May 2024 11:15:56 +0000
Subject: [PATCH] adding tf version of the new gammakpar!

---
 scratch.jl              | 11 +++---
 src/gammas/gammaKpar.jl | 77 ++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 81 insertions(+), 7 deletions(-)

diff --git a/scratch.jl b/scratch.jl
index fe8b999..d00febb 100755
--- a/scratch.jl
+++ b/scratch.jl
@@ -1,14 +1,13 @@
-using FastLink
 using DataFrames
 using CSV
 using PooledArrays
-import Pkg.Artifacts: @artifact_str
-#Pkg.add(url="https://github.com/jw2249a/FastLink.jl")
-using StatsBase
+import Pkg
+Pkg.add(url="https://github.com/jw2249a/FastLink.jl")
+using FastLink
 using JSON
 
-a_fil = @artifact_str "dfA"
-b_fil = @artifact_str "dfB"
+a_fil = Pkg.Artifacts.@artifact_str "dfA"
+b_fil = Pkg.Artifacts.@artifact_str "dfB"
 
 
 dfA=CSV.read("$(a_fil)/dfA.csv", DataFrame,
diff --git a/src/gammas/gammaKpar.jl b/src/gammas/gammaKpar.jl
index c1b1ffd..e82ad59 100755
--- a/src/gammas/gammaKpar.jl
+++ b/src/gammas/gammaKpar.jl
@@ -129,6 +129,82 @@ function gammaKpar!(vecA::PooledVector,vecB::PooledVector,results::DiBitMatrix,
     return nothing
 end
 
+
+
+# vector version of gammakpar for highly entropic data (aka high ordinality compared to obs)
+function gammaKpar!(vecA::Vector,vecB::Vector,results::DiBitMatrix,
+                     tf_table_x::SubArray{Float16},
+                    tf_table_y::SubArray{Float16};
+                    tf_minimum_u_value=0.001)
+    vecA = PooledArray(vecA)
+    vecB = PooledArray(vecB)
+    if @isdefined(_dims) == false
+        _dims = (length(vecA), length(vecB))
+    end
+    # Segment unique keys from missing key
+    missingvals_x = findfirst(ismissing.(vecA.pool))
+    iter_x=filter(x -> x != missingvals_x, 0x00000001:UInt32(length(vecA.pool)))
+    
+    missingvals_y = findfirst(ismissing.(vecB.pool))
+    iter_y=filter(x -> x != missingvals_y, 0x00000001:UInt32(length(vecB.pool)))
+    
+    # Form match matrices based on differing levels of matches
+    Threads.@threads for x in iter_x
+        indices_x = findall(vecA.refs .=== x)
+         # term frequency adjustment for x
+        tf_val_x = length(indices_x)/_dims[1]
+        for tf_i in indices_x
+            tf_table_x[tf_i] = max(tf_val_x, tf_minimum_u_value)
+        end
+        for y in  iter_y
+            indices_y = findall(vecB.refs .=== y)
+             # term frequency adjustment for y
+            tf_val_y = length(indices_y)/_dims[2]
+            for tf_i in indices_y
+                tf_table_y[tf_i] = max(tf_val_y, tf_minimum_u_value)
+            end
+            # if matches at a threshold, go through result vector and assign new value
+            if vecA.pool[x] == vecB.pool[y]
+                for ix in indices_x,iy in indices_y
+                    results[ix,iy] = match2
+                end
+            end
+        end
+    end
+
+    # set all to missing where x is missing
+    if !isnothing(missingvals_x)
+        missingindices = findall(vecA.refs .== missingvals_x)
+        # term frequency adjustment for x
+        tf_val_x = length(missingindices)/_dims[1]
+        for tf_i in missingindices
+            tf_table_x[tf_i] = max(tf_val_x, tf_minimum_u_value)
+        end
+        Threads.@threads for iy in 1:_dims[2]
+            for ix in missingindices
+                results[ix,iy] = missingval
+            end
+        end
+    end
+    # set all to missing where y is missing
+    if !isnothing(missingvals_y)
+        missingindices = findall(vecB.refs .== missingvals_y)
+         # term frequency adjustment for y
+        tf_val_y = length(missingindices)/_dims[2]
+        for tf_i in missingindices
+            tf_table_y[tf_i] = max(tf_val_y, tf_minimum_u_value)
+        end
+        
+        Threads.@threads for ix in 1:_dims[1]
+            for iy in missingindices
+                results[ix,iy] = missingval
+            end
+        end
+    end
+    # Return nothing
+    return nothing
+end
+
 # vector version of gammakpar for highly entropic data (aka high ordinality compared to obs)
 function gammaKpar!(vecA::Vector,vecB::Vector,results::DiBitMatrix)
     if @isdefined(_dims) == false
@@ -162,4 +238,3 @@ function gammaKpar!(vecA::Vector,vecB::Vector,results::DiBitMatrix)
     # Return nothing
     return nothing
 end
-