update

Author: mhjensen

doc/pub/week7/ipynb/rnncode.ipynb

@@ -893,7 +893,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "We also want to be able to choose which method we want to use for gradient descent when training the RNN. This is done by using a scheduler defined below. These schedulers are identical to the ones used in FYS-STK4155."
+    "We also want to be able to choose which method we want to use for gradient descent when training the RNN. This is done by using a scheduler defined below. These schedulers are identical to the ones used in FYS-STK4155 and during our discussions of the CNN code, see codes from week 5."
    ]
   },
   {
@@ -1136,10 +1136,10 @@
     "\n",
     "<u>**Note:**</u> The input and output from a node are two-dimensional arrays, with different inputs along the first axis, and the features along the second axis. The equations we have derived for computing forward- and backpropagation assume that we are working with column vectors, but when performing matrix multiplication with these two-dimensional arrays we are in practice working with row vectors. This has to be taken into account in our code.\n",
     "\n",
-    "Consider, for instance, multiplying some weight matrix $W$ with a vector $\\mathbf{h}$. The way we have derived the equations this will look like\n",
+    "Consider, for instance, multiplying some weight matrix $\\mathbf{W}$ with a vector $\\mathbf{h}$. The way we have derived the equations this will look like\n",
     "\n",
     "$$\n",
-    "W\\mathbf{h} = \\begin{pmatrix}\n",
+    "\\mathbf{Wh} = \\begin{pmatrix}\n",
     "W_{11} & \\cdots & W_{1n}\\\\\n",
     "\\vdots & \\ddots & \\vdots\\\\\n",
     "W_{m1} & \\cdots & W_{mn}\n",