add: df and AIC/BIC to Reg. Matrix Regression

tensorPredictors/R/RMReg.R

@@ -149,8 +149,14 @@ RMReg <- function(X, Z, y, lambda = 0,
         a1 <- alpha(a1, iter)
     }
 
-    # ### Degrees of Freedom estimate
-    # df <- 
+    ### Degrees of Freedom estimate (TODO: this is like in `matrix_sparsereg.m`)
+    sigma <- c(La.svd(A, 0, 0)$d, rep(0, max(shape) - min(shape)))
+    df <- if (!is.null(ZZiZ)) { ncol(Z) } else { 0 }
+    for (i in seq_len(sum(b1 > 0))) {
+        df <- df + 1 + sigma[i] * (sigma[i] - delta * lambda) * (
+            sum(ifelse((1:shape[1]) != i, 1 / (sigma[i]^2 - sigma[1:shape[1]]^2), 0)) +
+            sum(ifelse((1:shape[2]) != i, 1 / (sigma[i]^2 - sigma[1:shape[2]]^2), 0)))
+    }
 
     # return estimates and some additional stats
     list(
@@ -158,13 +164,10 @@ RMReg <- function(X, Z, y, lambda = 0,
         beta = if(is.null(ZZiZ)) { NULL } else { beta },
         singular.values = b1,
         iter = iter,
-        df = "?",
-        AIC = "?",
-        BIC = "?",
+        df = df,
+        loss = loss1,
+        AIC = loss1 / var(y) + 2 * df,
+        BIC = loss1 / var(y) + log(nrow(X)) * df,
         call = match.call() # invocing function call, collects params like lambda
     )
 }
-
-rmr.fit <- RMReg(X, NULL, y, lambda = 0)
-rmr.fit$iter
-rmr.fit$singular.values[rmr.fit$singular.values > 0]