add l15

ejuarezg · Oct 16, 2018 · 6ed3dd0 · 6ed3dd0
1 parent c64b777
commit 6ed3dd0
Show file tree

Hide file tree

Showing 3 changed files with 402 additions and 18 deletions.
diff --git a/Lecture12_CurseDim/Curse of Dimensionality and Principal Components Analysis.ipynb b/Lecture12_CurseDim/Curse of Dimensionality and Principal Components Analysis.ipynb
@@ -327,7 +327,9 @@
     "\n",
     "Thus, $\\mathbf{v}_j^T\\mathbf{v}_i = 0$. \n",
     "\n",
-    "* The eigenvecotrs can be interpreted as an orthogonal axis defined by the data. It makes sense to use these to look for data-centric coordinates/subspaces if the data projects differently to each axis. Since all data is noisy, we can concentrate on the axes corresponding to the largest eigenvectors if you are interested in preserving the variance of the data.  However, when using PCA within a classification problem, it is much more difficult because we are interested in discriminability (not necessarily variance).\n"
+    "* The eigenvectors can be interpreted as an orthogonal axis defined by the data. \n",
+    "* Since all data is noisy, we can concentrate on the axes corresponding to the largest eigenvectors if you are interested in preserving the variance of the data.  \n",
+    "* However, when using PCA within a classification problem, it is much more difficult because we are interested in discriminability (not necessarily variance).\n"
    ]
   },
   {