Online Sampling Approach

This PR is meant to apply a more efficient sampling process which will hopefully increase the speed of inference significantly by reducing allocations.

Experiments/Scripts/comparison_exps.jl

@@ -1,14 +1,12 @@
-
-using Profile
 include("../Experiments.jl")
 
 #datasets = [aids, yeast, hprd, dblp, youtube, wordnet]
-datasets = [lubm80, aids, human]
+datasets = [lubm80, aids, human, dblp]
 
 experiment_params = Vector{ExperimentParams}()
 for dataset in datasets
     push!(experiment_params, ExperimentParams(dataset=dataset))
-    push!(experiment_params, ExperimentParams(dataset=dataset, num_colors=256))
+#    push!(experiment_params, ExperimentParams(dataset=dataset, num_colors=256))
 end
 
 #build_experiments(experiment_params)

Source/QuasiStableCardinalityEstimator.jl

@@ -4,7 +4,7 @@
 # Equivalently, they perform a groupby on all other nodes of the query graph. The goal of this is to prevent
 # an exponential growth in the number of paths through the lifted color graph. However, we can only remove query nodes whose
 # edges have already been processed.
-function sum_over_node!(partial_paths::Vector{Tuple{Vector{Color}, Vector{Float64}}}, current_query_nodes, node_to_remove)
+@inline function sum_over_node(partial_paths::Matrix{Color}, partial_weights::Matrix{Float64}, current_query_nodes::Vector{Int}, node_to_remove::Int)
     nodeIdx = 1
     for node in current_query_nodes
         if node == node_to_remove
@@ -150,7 +150,6 @@ function get_all_simple_path_bools(start::Int, finish::Int, max_length::Int,
     return path_bools
 end
 
-
 # gets the directed path from the start to finish node
 function get_matching_graph(start::Int, finish::Int, query::QueryGraph)
     # convert the graph to be undirected