Add Flux 0.13 compatibility (#202)

* Add Flux 0.13 compatibility * `params` -> `Flux.params` * Forward-compatible (Flux 0.13 and 0.12) fix for tabular model * Subtype `AbstractOptimiser`
FluxML · Apr 20, 2022 · cbdf36b · cbdf36b
1 parent f51ecb6
commit cbdf36b
Show file tree

Hide file tree

Showing 6 changed files with 8 additions and 7 deletions.
diff --git a/Project.toml b/Project.toml
@@ -28,6 +28,7 @@ Markdown = "d6f4376e-aef5-505a-96c1-9c027394607a"
 MosaicViews = "e94cdb99-869f-56ef-bcf0-1ae2bcbe0389"
 Parameters = "d96e819e-fc66-5662-9728-84c9c7592b0a"
 PrettyTables = "08abe8d2-0d0c-5749-adfa-8a2ac140af0d"
+ReTest = "e0db7c4e-2690-44b9-bad6-7687da720f89"
 Reexport = "189a3867-3050-52da-a836-e630ba90ab69"
 Requires = "ae029012-a4dd-5104-9daa-d747884805df"
 Setfield = "efcf1570-3423-57d1-acb7-fd33fddbac46"
@@ -51,7 +52,7 @@ DataLoaders = "0.1"
 FileIO = "1.7"
 FilePathsBase = "0.9"
 FixedPointNumbers = "0.8"
-Flux = "0.12"
+Flux = "0.12, 0.13"
 FluxTraining = "0.2"
 Glob = "1"
 ImageInTerminal = "0.4"

diff --git a/src/Tabular/models.jl b/src/Tabular/models.jl
@@ -47,7 +47,7 @@ function TabularModel(
 
     tabularbackbone = Parallel(vcat, catbackbone, contbackbone)
 
-    classifierin = mapreduce(layer -> size(layer.weight)[1], +, catbackbone[2].layers;
+    classifierin = mapreduce(layer -> size(layer.weight)[1], +, Tuple(catbackbone[2].layers);
                              init = contbackbone.chs)
     dropout_rates = Iterators.cycle(dropout_rates)
     classifiers = []
@@ -139,7 +139,7 @@ function tabular_embedding_backbone(embedding_sizes, dropout_rate=0.)
     emb_drop = iszero(dropout_rate) ? identity : Dropout(dropout_rate)
     Chain(
         x -> tuple(eachrow(x)...),
-        Parallel(vcat, embedslist),
+        Parallel(vcat, embedslist...),
         emb_drop
     )
 end

diff --git a/src/Vision/models/blocks.jl b/src/Vision/models/blocks.jl
@@ -38,7 +38,7 @@ function visionhead(
     acts = vcat([relu for _ ∈ 1:n-2], [identity])
     pool = concat_pool ? AdaptiveConcatPool((1, 1)) : AdaptiveMeanPool((1, 1))
 
-    layers = [pool, flatten]
+    layers = [pool, Flux.flatten]
 
     for (h_in, h_out, act) in zip(hs, hs[2:end], acts)
         push!(layers, linbndrop(h_in, h_out, act=act, p=p))

diff --git a/src/training/discriminativelrs.jl b/src/training/discriminativelrs.jl
@@ -23,7 +23,7 @@ dlro = DiscriminativeLRs(paramgroups, Dict(1 => 0., 2 => 1.))
 o = Optimiser(dlro, Descent(0.1))
 ```
 """
-struct DiscriminativeLRs
+struct DiscriminativeLRs <: Flux.Optimise.AbstractOptimiser
     pg::ParamGroups
     factorfn
 end

diff --git a/src/training/lrfind.jl b/src/training/lrfind.jl
@@ -46,7 +46,7 @@ function lrfind(
     withfields(
         learner,
         model = modelcheckpoint,
-        params = params(modelcheckpoint),
+        params = Flux.params(modelcheckpoint),
         optimizer = deepcopy(learner.optimizer)
         ) do
 

diff --git a/src/training/paramgroups.jl b/src/training/paramgroups.jl
@@ -32,7 +32,7 @@ getgroup(pg::ParamGroups, x::AbstractArray) = get(pg.map, x, nothing)
 
 function assigngroups!(pg::ParamGroups, grouper, m)
     for (group, m_) in group(grouper, m)
-        for p in params(m_)
+        for p in Flux.params(m_)
             pg.map[p] = group
         end
     end