Drop Loss from loss functions

Context is already provided by the module.
dahong67 · Jun 14, 2024 · fa2a3ff · fa2a3ff
1 parent 1d9a07b
commit fa2a3ff
Show file tree

Hide file tree

Showing 10 changed files with 153 additions and 160 deletions.
diff --git a/benchmark/suites/gcp.jl b/benchmark/suites/gcp.jl
@@ -11,7 +11,7 @@ for sz in [(15, 20, 25), (30, 40, 50)], r in 1:2
     M = CPD(ones(r), rand.(sz, r))
     X = [M[I] for I in CartesianIndices(size(M))]
     SUITE["least-squares-size(X)=$sz, rank(X)=$r"] =
-        @benchmarkable gcp($X, $r; loss = GCPLosses.LeastSquaresLoss())
+        @benchmarkable gcp($X, $r; loss = GCPLosses.LeastSquares())
 end
 
 # Benchmark Poisson loss
@@ -20,7 +20,7 @@ for sz in [(15, 20, 25), (30, 40, 50)], r in 1:2
     M = CPD(fill(10.0, r), rand.(sz, r))
     X = [rand(Poisson(M[I])) for I in CartesianIndices(size(M))]
     SUITE["poisson-size(X)=$sz, rank(X)=$r"] =
-        @benchmarkable gcp($X, $r; loss = GCPLosses.PoissonLoss())
+        @benchmarkable gcp($X, $r; loss = GCPLosses.Poisson())
 end
 
 # Benchmark Gamma loss
@@ -30,7 +30,7 @@ for sz in [(15, 20, 25), (30, 40, 50)], r in 1:2
     k = 1.5
     X = [rand(Gamma(k, M[I] / k)) for I in CartesianIndices(size(M))]
     SUITE["gamma-size(X)=$sz, rank(X)=$r"] =
-        @benchmarkable gcp($X, $r; loss = GCPLosses.GammaLoss())
+        @benchmarkable gcp($X, $r; loss = GCPLosses.Gamma())
 end
 
 # Benchmark BernoulliOdds Loss
@@ -39,7 +39,7 @@ for sz in [(15, 20, 25), (30, 40, 50)], r in 1:2
     M = CPD(ones(r), rand.(sz, r))
     X = [rand(Bernoulli(M[I] / (M[I] + 1))) for I in CartesianIndices(size(M))]
     SUITE["bernoulliOdds-size(X)=$sz, rank(X)=$r"] =
-        @benchmarkable gcp($X, $r; loss = GCPLosses.BernoulliOddsLoss())
+        @benchmarkable gcp($X, $r; loss = GCPLosses.BernoulliOdds())
 end
 
 end
diff --git a/benchmark/suites/leastsquares.jl b/benchmark/suites/leastsquares.jl
@@ -13,7 +13,7 @@ for sz in [(15, 20, 25), (30, 40, 50), (60, 70, 80)], r in [1, 10, 50]
     M = CPD(ones(r), rand.(sz, r))
     X = [M[I] for I in CartesianIndices(size(M))]
     SUITE["size(X)=$sz, rank(X)=$r"] =
-        @benchmarkable gcp($X, $r; loss = GCPLosses.LeastSquaresLoss())
+        @benchmarkable gcp($X, $r; loss = GCPLosses.LeastSquares())
 end
 
 # Order-4 tensors
@@ -22,7 +22,7 @@ for sz in [(15, 20, 25, 30), (30, 40, 50, 60)], r in [1, 10, 50]
     M = CPD(ones(r), rand.(sz, r))
     X = [M[I] for I in CartesianIndices(size(M))]
     SUITE["least-squares-size(X)=$sz, rank(X)=$r"] =
-        @benchmarkable gcp($X, $r; loss = GCPLosses.LeastSquaresLoss())
+        @benchmarkable gcp($X, $r; loss = GCPLosses.LeastSquares())
 end
 
 # Order-5 tensors
@@ -31,7 +31,7 @@ for sz in [(15, 20, 25, 30, 35), (30, 30, 30, 30, 30)], r in [1, 10, 50]
     M = CPD(ones(r), rand.(sz, r))
     X = [M[I] for I in CartesianIndices(size(M))]
     SUITE["least-squares-size(X)=$sz, rank(X)=$r"] =
-        @benchmarkable gcp($X, $r; loss = GCPLosses.LeastSquaresLoss())
+        @benchmarkable gcp($X, $r; loss = GCPLosses.LeastSquares())
 end
 
 end
diff --git a/docs/src/quickstart.md b/docs/src/quickstart.md
@@ -143,7 +143,7 @@ For example,
 to try non-negative least-squares simply run
 
 ```@repl quickstart
-M_nonneg = gcp(X, 1; loss = GCPLosses.NonnegativeLeastSquaresLoss())
+M_nonneg = gcp(X, 1; loss = GCPLosses.NonnegativeLeastSquares())
 ```
 
 !!! tip "Congratulations!"

diff --git a/src/GCPDecompositions.jl b/src/GCPDecompositions.jl
@@ -30,7 +30,7 @@ end
 
 """
     gcp(X::Array, r;
-        loss = GCPLosses.LeastSquaresLoss(),
+        loss = GCPLosses.LeastSquares(),
         constraints = default_constraints(loss),
         algorithm = default_algorithm(X, r, loss, constraints),
         init = default_init(X, r, loss, constraints, algorithm))
@@ -42,7 +42,7 @@ Keyword arguments:
 + `constraints` : a `Tuple` of constraints on the factor matrices `U = (U[1],...,U[N])`.
 + `algorithm`   : algorithm to use
 
-Conventional CP corresponds to the default `GCPLosses.LeastSquaresLoss()` loss
+Conventional CP corresponds to the default `GCPLosses.LeastSquares()` loss
 with the default of no constraints (i.e., `constraints = ()`).
 
 If the LossFunctions.jl package is also loaded,
@@ -55,7 +55,7 @@ See also: `CPD`, `GCPLosses`, `GCPConstraints`, `GCPAlgorithms`.
 gcp(
     X::Array,
     r;
-    loss = GCPLosses.LeastSquaresLoss(),
+    loss = GCPLosses.LeastSquares(),
     constraints = default_constraints(loss),
     algorithm = default_algorithm(X, r, loss, constraints),
     init = default_init(X, r, loss, constraints, algorithm),
@@ -91,12 +91,8 @@ loss function `loss`, and tuple of constraints `constraints`.
 
 See also: `gcp`.
 """
-default_algorithm(
-    X::Array{<:Real},
-    r,
-    loss::GCPLosses.LeastSquaresLoss,
-    constraints::Tuple{},
-) = GCPAlgorithms.FastALS()
+default_algorithm(X::Array{<:Real}, r, loss::GCPLosses.LeastSquares, constraints::Tuple{}) =
+    GCPAlgorithms.FastALS()
 default_algorithm(X, r, loss, constraints) = GCPAlgorithms.LBFGSB()
 
 """

diff --git a/src/gcp-algorithms/als.jl b/src/gcp-algorithms/als.jl
@@ -4,7 +4,7 @@
     ALS
 
 **A**lternating **L**east **S**quares.
-Workhorse algorithm for `LeastSquaresLoss` with no constraints.
+Workhorse algorithm for `LeastSquares` loss with no constraints.
 
 Algorithm parameters:
 
@@ -17,7 +17,7 @@ end
 function _gcp(
     X::Array{TX,N},
     r,
-    loss::GCPLosses.LeastSquaresLoss,
+    loss::GCPLosses.LeastSquares,
     constraints::Tuple{},
     algorithm::GCPAlgorithms.ALS,
     init,

diff --git a/src/gcp-algorithms/fastals.jl b/src/gcp-algorithms/fastals.jl
@@ -24,7 +24,7 @@ end
 function _gcp(
     X::Array{TX,N},
     r,
-    loss::GCPLosses.LeastSquaresLoss,
+    loss::GCPLosses.LeastSquares,
     constraints::Tuple{},
     algorithm::GCPAlgorithms.FastALS,
     init,