Add serialization reference test & bump Legolas compat (#22)

* add serialization test, bump compat

* format example

* Update examples/digits.jl

* Update examples/digits.jl

Co-authored-by: Dave Kleinschmidt <[email protected]>

* Bump to Legolas v0.4

Co-authored-by: Dave Kleinschmidt <[email protected]>

.JuliaFormatter.toml

@@ -0,0 +1 @@
+style = "yas"

.github/workflows/CI.yml

@@ -23,18 +23,12 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        legolas-version:
-          - '0.2'
-          - '0.3'
         flux-version:
           - '0.12'
           - '0.13'
         version:
           - '1' # current release
-        include:
-          - version: '1.5' # earliest supported version
-            legolas-version: '0.2'
-            flux-version: '0.12'
+          - '1.6' # earliest supported version
     steps:
       - uses: actions/checkout@v2
         with:
@@ -43,12 +37,11 @@ jobs:
         with:
           version: ${{ matrix.version }}
           arch: x64
-      - name: "Install Legolas and Flux"
+      - name: "Install Flux"
         shell: julia --color=yes --project=. {0}
         run: |
           using Pkg
-          Pkg.add([Pkg.PackageSpec(; name="Legolas", version="${{ matrix.legolas-version }}"),
-                   Pkg.PackageSpec(; name="Flux", version="${{ matrix.flux-version }}")])
+          Pkg.add([Pkg.PackageSpec(; name="Flux", version="${{ matrix.flux-version }}")])
       - uses: actions/cache@v2
         with:
           path: ~/.julia/artifacts

Project.toml

@@ -1,7 +1,7 @@
 name = "LegolasFlux"
 uuid = "eb5f792d-d1b1-4535-bae3-d5649ec7daa4"
 authors = ["Beacon Biosignals, Inc."]
-version = "0.1.7"
+version = "0.1.8"
 
 [deps]
 Arrow = "69666777-d1a9-59fb-9406-91d4454c9d45"
@@ -13,9 +13,9 @@ Tables = "bd369af6-aec1-5ad0-b16a-f7cc5008161c"
 Arrow = "1, 2"
 Flux = "0.12, 0.13"
 Functors = "0.2.6, 0.3"
-Legolas = "0.1, 0.2, 0.3"
+Legolas = "0.4"
 Tables = "1"
-julia = "1.5"
+julia = "1.6"
 
 [extras]
 Flux = "587475ba-b771-5e3f-ad9e-33799f191a9c"

examples/digits.jl

@@ -8,13 +8,13 @@ using StableRNGs
 using Flux: onehotbatch, onecold, crossentropy, throttle
 using Base.Iterators: repeated, partition
 using Legolas, LegolasFlux
+using Tables
 
 # This should store all the information needed
 # to construct the model.
-Base.@kwdef struct DigitsConfig
-    seed::Int = 5
-    dropout_rate::Float32 = 0f1
-end
+const DigitsConfig = Legolas.@row("digits-config@1",
+                                  seed::Int = 5,
+                                  dropout_rate::Float32 = 0.0f1,)
 
 # Here's our model object itself, just a `DigitsConfig` and
 # a `chain`. We keep the config around so it's easy to save out
@@ -57,9 +57,9 @@ end
 # Here, we define a schema extension of the `legolas-flux.model` schema.
 # We add our `DigitsConfig` object, as well as the epoch and accuracy.
 const DigitsRow = Legolas.@row("digits.model@1" > "legolas-flux.model@1",
-                               config::DigitsConfig,
-                               epoch::Union{Missing, Int},
-                               accuracy::Union{Missing, Float32})
+                               config::DigitsConfig = DigitsConfig(config),
+                               epoch::Union{Missing,Int},
+                               accuracy::Union{Missing,Float32})
 
 # Construct a `DigitsRow` from a model by collecting the weights.
 # This can then be saved with e.g. `LegolasFlux.write_model_row`.
@@ -76,7 +76,6 @@ function DigitsModel(row)
     return m
 end
 
-
 # Increase to get more training/test data
 N_train = 1_000
 N_test = 50
@@ -109,11 +108,12 @@ function accuracy(m, x, y)
     return val
 end
 
-function train_model!(m; N = N_train)
+function train_model!(m; N=N_train)
     loss = (x, y) -> crossentropy(m(x), y)
     opt = ADAM()
     evalcb = throttle(() -> @show(accuracy(m, tX, tY)), 5)
-    Flux.@epochs 1 Flux.train!(loss, Flux.params(m), Iterators.take(train, N), opt; cb=evalcb)
+    Flux.@epochs 1 Flux.train!(loss, Flux.params(m), Iterators.take(train, N), opt;
+                               cb=evalcb)
     return accuracy(m, tX, tY)
 end
 
@@ -138,3 +138,22 @@ testmode!(m2)
 output2 = m2(input)
 
 @test output ≈ output2
+
+path = joinpath(pkgdir(LegolasFlux), "examples", "test.digits-model.arrow")
+# The saved weights in this repo were generated by running the command:
+# Legolas.write(path, [row], Legolas.Schema("digits.model@1"))
+# We don't run this every time, since we want to test that we can continue to deserialize previously saved out weights.
+table = Legolas.read(path)
+roundtripped = DigitsRow(only(Tables.rows(table)))
+@test roundtripped isa DigitsRow
+@test roundtripped.config isa DigitsConfig
+
+roundtripped_model = DigitsModel(roundtripped)
+output3 = roundtripped_model(input)
+@test output3 isa Matrix{Float32}
+
+# Here, we've hardcoded the results at the time of serialization.
+# This lets us check that the model we've saved gives the same answers now as it did then.
+# It is OK to update this test w/ a new reference if the answers are *supposed* to change for some reason. Just make sure that is the case.
+@test output3 ≈
+      Float32[0.09915658; 0.100575574; 0.101189725; 0.10078623; 0.09939819; 0.099650174; 0.1013182; 0.09952383; 0.0991391; 0.09926238;;]