Unit test pseudobulk projections

test/basic.jl

@@ -420,10 +420,16 @@
 	# @testset "pseudobulk $name" for (name,data,data_proj) in (("counts",counts,counts_proj), ("transformed",transformed,transformed_proj), ("reduced",reduced,reduced_proj))
 		d = copy(data)
 		d.obs.group2 = replace(d.obs.group, "C"=>missing)
-		d.obs.group3 = rand(StableRNG(276), ("a","b"), size(data,2))
+		d.obs.group3 = rand(StableRNG(276), ("a","b"), size(d,2))
 		d.obs.twogroup = replace(d.obs.group, "C"=>"A")
 		X = materialize(d.matrix)
 
+		d_proj = copy(data_proj)
+		d_proj.obs.group2 = replace(d_proj.obs.group, "C"=>missing)
+		d_proj.obs.group3 = rand(StableRNG(276), ("a","b"), size(d_proj,2))
+		d_proj.obs.twogroup = replace(d_proj.obs.group, "C"=>"A")
+		X_proj = materialize(d_proj.matrix)
+
 		@testset "$annot" for annot in ("group","group2","group3","twogroup")
 			unique_groups = collect(skipmissing(unique!(sort(d.obs[!,annot]))))
 
@@ -439,9 +445,26 @@
 			for g in unique_groups
 				x = vec(mean(X[:, isequal.(d.obs[!,annot], g)]; dims=2))
 				gi = findfirst(isequal(g), pb.obs.id)
-
 				@test x ≈ pb_X[:,gi]
 			end
+
+
+			unique_groups_proj = collect(skipmissing(unique!(sort(d_proj.obs[!,annot]))))
+
+			pb_proj = project(d_proj, pb)
+			@test names(pb_proj.obs) == ["id", annot]
+			@test unique!(sort(pb_proj.obs.id)) == unique_groups_proj
+			@test unique!(sort(pb_proj.obs[!,annot])) == unique_groups_proj
+
+			pb_proj_X = materialize(pb_proj.matrix)
+			@test size(pb_proj_X,1) == size(X_proj,1)
+			@test size(pb_proj_X,2) == length(unique_groups_proj)
+
+			for g in unique_groups_proj
+				x_proj = vec(mean(X_proj[:, isequal.(d_proj.obs[!,annot], g)]; dims=2))
+				gi = findfirst(isequal(g), pb_proj.obs.id)
+				@test x_proj ≈ pb_proj_X[:,gi]
+			end
 		end
 
 		@testset "$annot1, $annot2" for (annot1,annot2) in (("group","group3"),("group","group2"),("group","twogroup"))
@@ -455,7 +478,6 @@
 			@test unique!(sort(pb.obs[!,annot1])) == unique!(sort!(d.obs[mask,annot1]))
 			@test unique!(sort(pb.obs[!,annot2])) == unique!(sort!(d.obs[mask,annot2]))
 
-
 			pb_X = materialize(pb.matrix)
 			@test size(pb_X,1) == size(X,1)
 			@test size(pb_X,2) == length(unique_groups)
@@ -470,7 +492,32 @@
 					@test gi === nothing
 				end
 			end
-		end
 
+
+			groups_proj = string.(d_proj.obs[!,annot1],'_',d_proj.obs[!,annot2])
+			mask_proj = .!ismissing.(d_proj.obs[!,annot1]) .& .!ismissing.(d_proj.obs[!,annot2])
+			unique_groups_proj = unique!(sort!(groups_proj[mask_proj]))
+
+			pb_proj = project(d_proj, pb)
+			@test names(pb_proj.obs) == ["id", annot1, annot2]
+			@test unique!(sort(pb_proj.obs.id)) == unique_groups_proj
+			@test unique!(sort(pb_proj.obs[!,annot1])) == unique!(sort!(d_proj.obs[mask_proj,annot1]))
+			@test unique!(sort(pb_proj.obs[!,annot2])) == unique!(sort!(d_proj.obs[mask_proj,annot2]))
+
+			pb_proj_X = materialize(pb_proj.matrix)
+			@test size(pb_proj_X,1) == size(X_proj,1)
+			@test size(pb_proj_X,2) == length(unique_groups_proj)
+
+			for g1 in unique(d_proj.obs[mask_proj,annot1]), g2 in unique(d_proj.obs[mask_proj,annot2])
+				g_mask = isequal.(d_proj.obs[!,annot1], g1) .& isequal.(d_proj.obs[!,annot2], g2)
+				x_proj = vec(mean(X_proj[:, g_mask]; dims=2))
+				gi = findfirst(isequal(string(g1,'_',g2)), pb_proj.obs.id)
+				if any(g_mask) # are there any observations in this group?
+					@test x_proj ≈ pb_proj_X[:,gi]
+				else
+					@test gi === nothing
+				end
+			end
+		end
 	end
 end