Merge pull request #138 from invenia/rofinn-patch-1

Update Project.toml and DataDep

.github/workflows/CI.yml

@@ -28,9 +28,8 @@ jobs:
           - os: windows-latest
             arch: x86
         include:
-          # Add a 1.5 job because that's what Invenia actually uses
           - os: ubuntu-latest
-            version: 1.5
+            version: 1.6
             arch: x64
     steps:
       - uses: actions/checkout@v2

Project.toml

@@ -1,7 +1,7 @@
 name = "Impute"
 uuid = "f7bf1975-0170-51b9-8c5f-a992d46b9575"
 authors = ["Invenia Technical Computing"]
-version = "0.6.10"
+version = "0.6.11"
 
 [deps]
 BSON = "fbb218c0-5317-5bc6-957e-2ee96dd4b1f0"
@@ -20,7 +20,7 @@ TableOperations = "ab02a1b2-a7df-11e8-156e-fb1833f50b87"
 Tables = "bd369af6-aec1-5ad0-b16a-f7cc5008161c"
 
 [compat]
-AxisKeys = "0.1"
+AxisKeys = "0.1, 0.2"
 BSON = "0.2, 0.3"
 CSV = "0.6, 0.7, 0.8, 0.9, 0.10"
 DataDeps = "0.7"
@@ -32,7 +32,7 @@ NearestNeighbors = "0.4"
 StatsBase = "0.32, 0.33, 0.34"
 TableOperations = "0.2, 1.0"
 Tables = "1"
-julia = "1"
+julia = "1.6"
 
 [extras]
 AxisArrays = "39de3d68-74b9-583c-8d2d-e117c070f3a9"

src/data.jl

@@ -3,9 +3,10 @@ function register_datadep()
         DataDep(
             "impute-v1.0.0",
             "Datasets for testing and demonstrating Impute.jl",
-            "https://invenia-public-datasets.s3.amazonaws.com/Impute/v1.0.0/datasets.tar.gz",
-            "938b3705752eb73141476a2abc7a36cfdaba9ec45f99f0796f44e0870e006e1c",
-            post_fetch_method=unpack,
+            "https://www.dropbox.com/scl/fi/bgtfqea9qqoug42gcnnsl/datasets.tar.gz?rlkey=11xsae0wi32m8gcfxhbqgo030&dl=0",
+            "cf1fff2e7f3ce28eb4264060bc7b9ee561bccfce2c5915c4cf758ec48477ddfe",
+            fetch_method=DataDeps.fetch_base,
+            post_fetch_method=DataDeps.unpack,
         )
     )
 end

src/imputors/knn.jl

@@ -73,13 +73,16 @@ function impute!(data::AbstractMatrix{Union{T, Missing}}, imp::KNN; dims=nothing
 
             # Weight valid neighbors based on inverse distance
             neighbor_dists = dists[.!neighbor_mask]
-            wv = weights(1.0 ./ neighbor_dists)
+
+            # Delay creating Weights as they don't support Inf or NaN anymore.
+            wv = 1.0 ./ neighbor_dists
+            Σ = sum(wv)
 
             # Only fill with the weighted mean of neighbors if the sum of the weights are
             # non-zero and finite.
-            if isfinite(sum(wv)) && !iszero(sum(wv))
+            if isfinite(Σ) && !iszero(Σ)
                 neighbor_vals = X[i, idxs[.!neighbor_mask]]
-                X[i, j] = mean(neighbor_vals, wv)
+                X[i, j] = mean(neighbor_vals, Weights(wv, Σ))
             end
         end
     end