Working on debiasing experiments

justfile

@@ -0,0 +1,3 @@
+all:
+  @rm -f src/*.cov
+  @julia --project --color=yes --threads 6 make.jl

replication/debiasing/debiasing.jl

@@ -0,0 +1,45 @@
+using Distributions
+using PyPlot
+using MondrianForests
+
+rcParams = PyPlot.PyDict(PyPlot.matplotlib."rcParams")
+rcParams["text.usetex"] = true
+rcParams["font.family"] = "serif"
+plt.ioff()
+
+d = 1
+n = 100000
+x_evals = [ntuple(j -> 0.0, d)]
+y_evals = [0.0]
+n_evals = 1
+X_dist = Uniform(-1, 1)
+sigma = 0.01
+eps_dist = Normal(0, sigma)
+
+rand(X_dist)
+
+X = [ntuple(j -> rand(X_dist), d) for i in 1:n]
+Y = [X[i][1]^2 + rand(eps_dist) for i in 1:n]
+
+
+(fig, ax) = plt.subplots(figsize=(5, 5))
+plt.scatter(X, Y)
+savefig("replication/debiasing/plot.png", dpi=150)
+plt.close()
+
+#lambdas = collect(0.1:0.1:2)
+n_trees = 100
+#n_subsample = n
+#lambda = select_lifetime_gcv(lambdas, n_trees, X, Y, debias_order, n_subsample)
+#println(lambda)
+for debias_order in [0, 1]
+    #lambda = select_lifetime_polynomial(X, Y, debias_order)
+    gamma = d + 4*(debias_order + 1)
+    println(gamma)
+    lambda = n^(-1/gamma)
+    println(lambda)
+    forest = MondrianForest(lambda, n_trees, x_evals, X, Y)
+    mse = forest.mu_hat[]^2
+    println(mse)
+end
+