added generated script to report, removed non-UTF-8 chars

nncw.ipynb

@@ -716,13 +716,13 @@
     "The below function runs an iteration of layer/epoch investigations.\n",
     "Returns the amount of layers/epochs used as well as the results and the model.\n",
     "\n",
-    "Using cancer dataset (as in E2) and ‘trainscg’ or an optimiser of your choice, vary nodes and epochs (that is using early stopping for epochs) over suitable range, to find optimal choice in terms of classification test error rate of node/epochs for 50/50% random train/test split (no validation set). It is suggested that you initially try epochs = [ 1 2 4 8 16 32 64], nodes = [2 8 32], so there would be 21 node/epoch combinations. \n",
+    "Using cancer dataset (as in E2) and 'trainscg' or an optimiser of your choice, vary nodes and epochs (that is using early stopping for epochs) over suitable range, to find optimal choice in terms of classification test error rate of node/epochs for 50/50% random train/test split (no validation set). It is suggested that you initially try epochs = [ 1 2 4 8 16 32 64], nodes = [2 8 32], so there would be 21 node/epoch combinations. \n",
     "\n",
-    "(Hint1: from the ‘advanced script’in E2, nodes can be changed to xx, with hiddenLayerSize = xx; and epochs changed to xx by addingnet. trainParam.epochs = xx; placed afternet = patternnet(hiddenLayerSize, trainFcn); --see ‘trainscg’ help documentation for changing epochs). \n",
+    "(Hint1: from the 'advanced script' in E2, nodes can be changed to xx, with hiddenLayerSize = xx; and epochs changed to xx by addingnet. trainParam.epochs = xx; placed afternet = patternnet(hiddenLayerSize, trainFcn); --see 'trainscg' help documentation for changing epochs). \n",
     "\n",
     "Repeat each of the 21 node/epoch combinations at least thirty times, with different 50/50 split and take average and report classification error rate and standard deviation (std). Graph classification train and test error rate and std as node-epoch changes, that is plot error rate vs epochs for different number of nodes. Report the optimal value for test error rate and associated node/epoch values. \n",
     "\n",
-    "(Hint2: as epochs increases you can expect the test error rate to reach a minimum and then start increasing, you may need to set the stopping criteria to achieve the desired number of epochs – Hint 3: to find classification error rates for train and test set, you need to check the code from E2, to determine how you may obtain the train and test set patterns)\n"
+    "(Hint2: as epochs increases you can expect the test error rate to reach a minimum and then start increasing, you may need to set the stopping criteria to achieve the desired number of epochs - Hint 3: to find classification error rates for train and test set, you need to check the code from E2, to determine how you may obtain the train and test set patterns)\n"
    ]
   },
   {
@@ -1619,7 +1619,7 @@
     "\n",
     "For cancer dataset, choose an appropriate value of node and epochs, based on Exp 1) and use ensemble of individual (base) classifiers with random starting weights and Majority Vote to see if performance improves - repeat the majority vote ensemble at least thirty times with different 50/50 split and average and graph (Each classifier in the ensemble sees the same training patterns). Repeat for a different odd number (prevents tied vote) of individual classifiers between 3 and 25, and comment on the result of individualclassifier accuracy vs ensemble accuracy as number of base classifiers varies. Consider changing the number of nodes/epochs (both less complex and more complex) to see if you obtain better performance, and comment on the result with respect to why the optimal node/epoch combination may be different for an ensemble compared with the base classifier, as in Exp 1). \n",
     "\n",
-    "(Hint4: to implement majority vote you need to determine the predicted class labels –probably easier to implement yourself rather than use the ensemble matlab functions)\n"
+    "(Hint4: to implement majority vote you need to determine the predicted class labels -probably easier to implement yourself rather than use the ensemble matlab functions)\n"
    ]
   },
   {
@@ -2133,7 +2133,7 @@
    "source": [
     "# Experiment 3\n",
     "\n",
-    "Repeat Exp 2) for cancer dataset with two different optimisers of your choice e.g. ‘trainlm’ and ‘trainrp’. Comment and discuss the result and decide which is more appropriate training algorithm for the problem. In your discussion, include in your description a detailed account of how the training algorithms (optimisations) work."
+    "Repeat Exp 2) for cancer dataset with two different optimisers of your choice e.g. 'trainlm' and 'trainrp'. Comment and discuss the result and decide which is more appropriate training algorithm for the problem. In your discussion, include in your description a detailed account of how the training algorithms (optimisations) work."
    ]
   },
   {

report/report.lyx

@@ -120,7 +120,7 @@ todonotes
 \begin_layout Title
 
 \size giant
-Training Neural Networks With Backpropagation
+Training Neural Networks with Backpropagation
 \end_layout
 
 \begin_layout Author
@@ -252,7 +252,7 @@ LatexCommand tableofcontents
 \end_layout
 
 \begin_layout Standard
-\begin_inset Newpage newpage
+\begin_inset Newpage pagebreak
 \end_inset
 
 
@@ -344,5 +344,16 @@ options "bibtotoc"
 Source Code
 \end_layout
 
+\begin_layout Standard
+\begin_inset CommandInset include
+LatexCommand lstinputlisting
+filename "../nncw.py"
+lstparams "caption={Formatted Jupyter notebook containing experiment code},label={notebook-code}"
+
+\end_inset
+
+
+\end_layout
+
 \end_body
 \end_document