|
| 1 | +@testset "optimization" begin |
| 2 | + |
| 3 | +@testset "in_place_add!" begin |
| 4 | + # TODO |
| 5 | +end |
| 6 | + |
| 7 | +@testset "Accumulator" begin |
| 8 | + # TODO |
| 9 | +end |
| 10 | + |
| 11 | +@testset "Julia parameter store" begin |
| 12 | + |
| 13 | + store = JuliaParameterStore() |
| 14 | + |
| 15 | + @gen function foo() |
| 16 | + @param theta::Float64 |
| 17 | + @param phi::Vector{Float64} |
| 18 | + end |
| 19 | + register_parameters!(foo, [:theta, :phi]) |
| 20 | + |
| 21 | + # before the parameters are initialized in the store |
| 22 | + |
| 23 | + @test Gen.get_local_parameters(store, foo) == Dict{Symbol,Any}() |
| 24 | + |
| 25 | + @test_throws KeyError get_gradient((foo, :theta), store) |
| 26 | + @test_throws KeyError get_parameter_value((foo, :theta), store) |
| 27 | + @test_throws KeyError increment_gradient!((foo, :theta), 1.0, store) |
| 28 | + @test_throws KeyError reset_gradient!((foo, :theta), store) |
| 29 | + @test_throws KeyError Gen.set_parameter_value!((foo, :theta), 1.0, store) |
| 30 | + @test_throws KeyError Gen.get_gradient_accumulator((foo, :theta), store) |
| 31 | + |
| 32 | + @test_throws KeyError get_gradient((foo, :phi), store) |
| 33 | + @test_throws KeyError get_parameter_value((foo, :phi), store) |
| 34 | + @test_throws KeyError increment_gradient!((foo, :phi), [1.0, 1.0], store) |
| 35 | + @test_throws KeyError reset_gradient!((foo, :phi), store) |
| 36 | + @test_throws KeyError Gen.set_parameter_value!((foo, :phi), [1.0, 1.0], store) |
| 37 | + @test_throws KeyError Gen.get_gradient_accumulator((foo, :phi), store) |
| 38 | + |
| 39 | + # after the parameters are initialized in the store |
| 40 | + |
| 41 | + init_parameter!((foo, :theta), 1.0, store) |
| 42 | + init_parameter!((foo, :phi), [1.0, 2.0], store) |
| 43 | + |
| 44 | + dict = Gen.get_local_parameters(store, foo) |
| 45 | + @test length(dict) == 2 |
| 46 | + @test dict[:theta] == 1.0 |
| 47 | + @test dict[:phi] == [1.0, 2.0] |
| 48 | + |
| 49 | + @test get_gradient((foo, :theta), store) == 0.0 |
| 50 | + @test get_parameter_value((foo, :theta), store) == 1.0 |
| 51 | + increment_gradient!((foo, :theta), 1.1, store) |
| 52 | + @test get_gradient((foo, :theta), store) == 1.1 |
| 53 | + increment_gradient!((foo, :theta), 1.1, 2.0, store) |
| 54 | + @test get_gradient((foo, :theta), store) == (1.1 + 2.2) |
| 55 | + reset_gradient!((foo, :theta), store) |
| 56 | + @test get_gradient((foo, :theta), store) == 0.0 |
| 57 | + Gen.set_parameter_value!((foo, :theta), 2.0, store) |
| 58 | + @test get_parameter_value((foo, :theta), store) == 2.0 |
| 59 | + @test get_value(Gen.get_gradient_accumulator((foo, :theta), store)) == 0.0 |
| 60 | + |
| 61 | + @test get_gradient((foo, :phi), store) == [0.0, 0.0] |
| 62 | + @test get_parameter_value((foo, :phi), store) == [1.0, 2.0] |
| 63 | + increment_gradient!((foo, :phi), [1.1, 1.2], store) |
| 64 | + @test get_gradient((foo, :phi), store) == [1.1, 1.2] |
| 65 | + increment_gradient!((foo, :phi), [1.1, 1.2], 2.0, store) |
| 66 | + @test get_gradient((foo, :phi), store) == ([1.1, 1.2] .+ (2.0 * [1.1, 1.2])) |
| 67 | + reset_gradient!((foo, :phi), store) |
| 68 | + @test get_gradient((foo, :phi), store) == [0.0, 0.0] |
| 69 | + Gen.set_parameter_value!((foo, :phi), [2.0, 3.0], store) |
| 70 | + @test get_parameter_value((foo, :phi), store) == [2.0, 3.0] |
| 71 | + @test Gen.get_value(Gen.get_gradient_accumulator((foo, :phi), store)) == [0.0, 0.0] |
| 72 | + |
| 73 | + # FixedStepGradientDescent |
| 74 | + |
| 75 | + # DecayStepGradientDescent |
| 76 | + |
| 77 | + # init_optimizer and apply_update! for FixedStepGradientDescent and DecayStepGradientDescent |
| 78 | + # default_parameter_context and default_julia_parameter_store |
| 79 | +end |
| 80 | + |
| 81 | +@testset "composite optimizer" begin |
| 82 | + |
| 83 | +end |
| 84 | + |
| 85 | + |
| 86 | +end |
0 commit comments