Fix more broadcast deprecations

rdeits · rdeits · commit 425d402009bc · 2017-05-03T18:32:49.000-04:00
diff --git a/src/layers/binary-cross-entropy-loss.jl b/src/layers/binary-cross-entropy-loss.jl
@@ -29,7 +29,7 @@ end
 function forward(backend::CPUBackend, state::BinaryCrossEntropyLossLayerState, inputs::Vector{Blob})
   pred = vec(inputs[1].data)
   label = vec(inputs[2].data)
-  loss = BLAS.dot(log.(pred), label) + BLAS.dot(log1p(-pred), (1-label))
+  loss = BLAS.dot(log.(pred), label) + BLAS.dot(log1p.(-pred), (1-label))
 
   num = get_num(inputs[1])
   state.loss = state.layer.weight * -loss/num
diff --git a/test/layers/accuracy.jl b/test/layers/accuracy.jl
@@ -12,7 +12,7 @@ function test_accuracy_layer(backend::Backend, tensor_dim, T)
   input = rand(T, dims)
   input_blob = make_blob(backend, input)
 
-  label = abs(rand(Int, dims_label)) % dims[op_dim]
+  label = abs.(rand(Int, dims_label)) .% dims[op_dim]
   label = convert(Array{T}, label)
   label_blob = make_blob(backend, label)
 
diff --git a/test/layers/binary-accuracy.jl b/test/layers/binary-accuracy.jl
@@ -8,9 +8,9 @@ function test_binary_accuracy_layer(backend::Backend, tensor_dim, T, threshold,
   println("    > $dims")
   if 0 == threshold
     preds = rand(T, dims)*4-2
-    labels = round(rand(T, dims))*2-1
+    labels = round.(rand(T, dims))*2-1
   elseif 0.5 == threshold
-    preds = round(rand(T, dims))
+    preds = round.(rand(T, dims))
     labels = rand(T, dims)
   else
     error("Threshold must be 0 or 0.5; was $threshold")
diff --git a/test/layers/hinge-loss.jl b/test/layers/hinge-loss.jl
@@ -8,7 +8,7 @@ function test_hinge_loss_layer(backend::Backend, T, eps)
   dims = tuple(rand(6:11, tensor_dim)...)
   println("    > $dims")
   preds = rand(T, dims)*4-2
-  labels = round(rand(T, dims))*2-1
+  labels = round.(rand(T, dims))*2-1
   errs_mask = preds.*labels .< one(T)
 
   ############################################################
@@ -29,7 +29,7 @@ function test_hinge_loss_layer(backend::Backend, T, eps)
 
   forward(backend, state, inputs)
 
-  loss = sum(max(one(T) .- preds.*labels, zero(T))) / dims[end]
+  loss = sum(max.(one(T) .- preds.*labels, zero(T))) / dims[end]
   @test -eps < loss-state.loss < eps
 
   backward(backend, state, inputs, diffs)
diff --git a/test/layers/multinomial-logistic-loss.jl b/test/layers/multinomial-logistic-loss.jl
@@ -13,9 +13,9 @@ function test_multinomial_logistic_loss_layer(backend::Backend, tensor_dim, clas
   dims = tuple(dims...)
   channels = dims[op_dim]
 
-  prob = abs(rand(T, dims)) + 0.01
+  prob = abs.(rand(T, dims)) .+ 0.01
 
-  label = abs(rand(Int, dims_label)) % channels
+  label = abs.(rand(Int, dims_label)) .% channels
   label = convert(Array{T}, label)
 
   prob_blob = make_blob(backend, prob)
@@ -27,7 +27,7 @@ function test_multinomial_logistic_loss_layer(backend::Backend, tensor_dim, clas
   elseif class_weights[1] == :local
     weights = rand(T, channels)
   elseif class_weights[1] == :global
-    weights = round(1000*rand(T, dims[1:end-1]))/1000
+    weights = round.(1000*rand(T, dims[1:end-1]))/1000
   else
     @assert class_weights[1] == :no
     weights = []
