nashconv tests

WhiffleFish · Jul 10, 2023 · 872c172 · 872c172
1 parent 8d18247
commit 872c172
Show file tree

Hide file tree

Showing 2 changed files with 15 additions and 5 deletions.
diff --git a/test/callback.jl b/test/callback.jl
@@ -3,11 +3,15 @@
     game = Kuhn()
     sol = CFRSolver(game)
     cb = CFR.ExploitabilityCallback(sol, 100)
-    train!(sol, 100_000, cb = cb)
+    cb_nashconv = NashConvCallback(sol, 100)
+    train!(sol, 100_000, cb = CallbackChain(cb, cb_nashconv))
 
     @test RecipesBase.apply_recipe(Dict{Symbol,Any}(), cb) ≠ nothing
+    @test RecipesBase.apply_recipe(Dict{Symbol,Any}(), cb_nashconv) ≠ nothing
     @test length(cb.hist.y) == length(cb.hist.x) == 1_000
+    @test length(cb_nashconv.hist.y) == length(cb_nashconv.hist.x) == 1_000
     @test 0.0 < last(cb.hist.y) < 1e-2
+    @test 0.0 < last(cb_nashconv.hist.y) < 1e-2
 
     game = MatrixGame([(randn(), randn()) for i in 1:5, j in 1:5])
     sol = CFRSolver(game)

diff --git a/test/is-mcts.jl b/test/is-mcts.jl
@@ -4,18 +4,24 @@
     ## MaxUCB
     sol = CFRSolver(game)
     true_exploit_cb = ExploitabilityCallback(sol, 10)
-    max_ucb_cb      = ExploitabilityCallback(ISMCTS(sol; max_iter=100_000, criterion=CFR.MaxUCB()), 10)
-    poly_ucb_cb     = ExploitabilityCallback(ISMCTS(sol; max_iter=100_000, criterion=CFR.PolyUCB()), 10)
-    max_q_cb        = ExploitabilityCallback(ISMCTS(sol; max_iter=100_000, criterion=CFR.MaxQ()), 10)
+    true_nashconv_cb= NashConvCallback(sol, 10)
+    max_ucb_cb      = MCTSExploitabilityCallback(sol, 10; max_iter=100_000, criterion=CFR.MaxUCB())
+    poly_ucb_cb     = MCTSExploitabilityCallback(sol, 10; max_iter=100_000, criterion=CFR.PolyUCB())
+    max_q_cb        = MCTSExploitabilityCallback(sol, 10; max_iter=100_000, criterion=CFR.MaxQ())
+    nashconv_cb     = MCTSNashConvCallback(sol, 10; max_iter=100_000)
 
-    cb_chain = CFR.CallbackChain(true_exploit_cb, max_ucb_cb, poly_ucb_cb, max_q_cb)
+
+    cb_chain = CFR.CallbackChain(true_exploit_cb, true_nashconv_cb, max_ucb_cb, poly_ucb_cb, max_q_cb, nashconv_cb)
     train!(sol, 100; cb=cb_chain)
 
     err_ucb = abs.(max_ucb_cb.hist.y .- true_exploit_cb.hist.y)
     err_poly = abs.(poly_ucb_cb.hist.y .- true_exploit_cb.hist.y)
     err_q = abs.(max_q_cb.hist.y .- true_exploit_cb.hist.y)
 
+    err_nashconv = abs.(nashconv_cb.hist.y .- true_nashconv_cb.hist.y)
+
     @test all(err_ucb[2:end] .< 0.1)
     @test all(err_poly[2:end] .< 0.1)
     @test all(err_q[2:end] .< 0.2)
+    @test all(err_nashconv[2:end] .< 0.2)
 end