checkpoint

src/explicit.jl

@@ -13,7 +13,7 @@ function PoissonEPCA()
     # assumes X = {integers}
     epsilon = eps()
     @. begin
-        Bregman(p, q) = p * log((p + epsilon) / (q + epsilon)) + q - p
+        Bregman(p, q) = p * (log(p + epsilon) - log(q + epsilon)) + q - p
         g(theta) = exp(theta)
     end
     mu = g(0)
@@ -25,7 +25,7 @@ function BernoulliEPCA()
     # assumes X = {0, 1}
     epsilon = eps()
     @. begin
-        Bregman(p, q) = p * log((p + epsilon)/ (q + epsilon)) + (1 - p) * log((1 - p + epsilon) / (1 - q + epsilon))
+        Bregman(p, q) = p * (log(p + epsilon) - log(q + epsilon)) + (1 - p) * (log(1 - p + epsilon) - log(1 - q + epsilon))
         g(theta) = exp(theta) / (1 + exp(theta))
     end
     mu = g(1)
@@ -57,7 +57,7 @@ function fit!(
 )
     B, g = epca.Bregman, epca.g
     L(theta) = begin
-        X_hat= g.(theta)
+        X_hat = g.(theta)
         sum(B(X, X_hat) + epsilon * B(mu, X_hat))
     end    
     n, d = size(X)
@@ -66,9 +66,9 @@ function fit!(
     for i in 1:maxiter
         V = Optim.minimizer(optimize(V_hat->L(A * V_hat), V))
         result = optimize(A_hat->L(A_hat * V), A)
-        loss = Optim.minimum(result)
         A = Optim.minimizer(result)
         if verbose && (i % print_steps == 0 || i == 1)
+            loss = Optim.minimum(result)
             println("Iteration: $i/$maxiter | Loss: $loss")
         end
     end
@@ -81,22 +81,24 @@ function compress(
     epca::ExplicitEPCA, 
     X;
     mu=epca.mu,
-    maxoutdim=1, 
     maxiter=100,
     verbose=false,
-    print_stesp=10,
+    print_steps=10,
     epsilon=eps()
 )
     B, g, V = epca.Bregman, epca.g, epca.V
     L(theta) = begin
-        X_hat= g.(theta)
+        X_hat = g.(theta)
         sum(@. B(X, X_hat) + epsilon * B(mu, X_hat))
     end    
     n, _ = size(X)
-    A = ones(n, maxoutdim)
+    outdim = size(V)[1]
+    A = ones(n, outdim)
     for _ in 1:maxiter
-        A = Optim.minimizer(optimize(A_hat->L(A_hat * V), A))
+        result = optimize(A_hat->L(A_hat * V), A)
+        A = Optim.minimizer(result)
         if verbose && (i % print_steps == 0 || i == 1)
+            loss = Optim.minimum(result)
             println("Iteration: $i/$maxiter | Loss: $loss")
         end
     end

src/implicit.jl

@@ -84,9 +84,9 @@ function fit!(
     for i in 1:maxiter
         V = Optim.minimizer(optimize(V_hat->L(A * V_hat), V))
         result = optimize(A_hat->L(A_hat * V), A)
-        loss = Optim.minimum(result)
         A = Optim.minimizer(result)
         if verbose && (i % print_steps == 0 || i == 1)
+            loss = Optim.minimum(result)
             println("Iteration: $i/$maxiter | Loss: $loss")
         end
     end
@@ -96,8 +96,8 @@ end
 
 function compress(
     epca::ImplicitEPCA, 
-    X,
-    mu;
+    X;
+    mu=1,  # NOTE: mu = 1 may not be valid for all link functions. 
     maxiter=100,
     verbose=false,
     print_steps=10,
@@ -106,11 +106,14 @@ function compress(
 )
     L = _make_loss(epca, X, mu, epsilon; tol=tol)
     n, _ = size(X)
-    A = ones(n, maxoutdim)
     V = epca.V
+    outdim = size(V)[1]
+    A = ones(n, outdim)
     for i in 1:maxiter
-        A = Optim.minimizer(optimize(A_hat->L(A_hat * V), A))
+        result = optimize(A_hat->L(A_hat * V), A)
+        A = Optim.minimizer(result)
         if verbose && (i % print_steps == 0 || i == 1)
+            loss = Optim.minimum(result)
             println("Iteration: $i/$maxiter | Loss: $loss")
         end
     end