update MSE normalization

Author: rasbt

ERRATA/README.md

@@ -12,6 +12,10 @@ The doc strings of the LogisticRegressionGD classifier reference "Mean squared e
 
 Add bias unit to the net input.
 
+**Page 354**
+
+The MSE is normalized via `mse = mse/i` but should be normalized via `mse = mse/(i+1)` instead. (This does not affect the results in shown below though. The MSE is still 0.3.)
+
 **Page 361**
 
 $$\frac{\partial}{\partial w_{j, l}^{(l)}}=L(\boldsymbol{W}, \boldsymbol{b})$$ 
@@ -20,6 +24,8 @@ should be
 
 $$\frac{\partial L}{\partial w_{j, l}^{(l)}}$$ 
 
+
+
 ## Chapter 12
 
 **Page 380**

ch11/ch11.ipynb

@@ -375,7 +375,7 @@ def compute_mse_and_acc(nnet, X, y, num_labels=10, minibatch_size=100):
         num_examples += targets.shape[0]
         mse += loss
 
-    mse = mse/i
+    mse = mse/(i+1)
     acc = correct_pred/num_examples
     return mse, acc