X-Git-Url: https://git.auder.net/%3C?a=blobdiff_plain;f=pkg%2FR%2Fz_util.R;fp=pkg%2FR%2Fz_util.R;h=996a5f82fe74431f9cb411abc020195981d949c6;hb=ae507b4e6ed3ae1f50a24a7c43d77be9501d791c;hp=0000000000000000000000000000000000000000;hpb=57dbbab63a48b42f85e0a711640e1dd10b58f18a;p=aggexp.git

diff --git a/pkg/R/z_util.R b/pkg/R/z_util.R
new file mode 100644
index 0000000..996a5f8
--- /dev/null
+++ b/pkg/R/z_util.R
@@ -0,0 +1,49 @@
+#Maximum size of stored data to predict next PM10
+MAX_HISTORY = 10000
+
+#Default lambda value (when too few data)
+LAMBDA = 2.
+
+#Maximum error to keep a line in (incremental) data
+MAX_ERROR = 20.
+
+#Turn a "vector" into 1D matrix if needed (because R auto cast 1D matrices)
+matricize = function(x)
+{
+	if (!is.null(dim(x)))
+		return (as.matrix(x))
+	return (t(as.matrix(x)))
+}
+
+#Moore-Penrose pseudo inverse
+mpPsInv = function(M)
+{
+	epsilon = 1e-10
+    s = svd(M)
+    sd = s$d ; sd[sd < epsilon] = Inf
+    sd = diag(1.0 / sd, min(nrow(M),ncol(M)))
+    return (s$v %*% sd %*% t(s$u))
+}
+
+#Heuristic for k in knn algorithms
+getKnn = function(n)
+{
+	return ( max(1, min(50, ceiling(n^(2./3.)))) )
+}
+
+#Minimize lambda*||u||^2 + ||Xu - Y||^2
+ridgeSolve = function(X, Y, lambda)
+{
+	s = svd(X)
+	deltaDiag = s$d / (s$d^2 + lambda)
+	deltaDiag[!is.finite(deltaDiag)] = 0.0
+	if (length(deltaDiag) > 1)
+		deltaDiag = diag(deltaDiag)
+	return (s$v %*% deltaDiag %*% t(s$u) %*% Y)
+}
+
+#Return the indices (of rows, by default) without any NA
+getNoNAindices = function(M, margin=1)
+{
+	return (apply(M, margin, function(z)(!any(is.na(z)))))
+}