"Body": "In the last demo you learned how to train a neural network using NeuroSolutions for Excel. In this demo you will use NeuroSolutions for Excel to test the classification performance of a trained neural network. This demo will use the data set and trained neural network from the previous demo. First, we will determine the classification performance of the \"Training\" data set. Then we will test the classification performance of data that the network has never seen. The last 50 rows of the data set have been tagged as \"Testing\" and will be used for this purpose. This will tell us whether the neural network memorized the training data or truly learned to model the underlying relationship.",
"Button": "None",
"Next": "None"
},
{
"Heading": "Training Data Classification",
"Body": "A testing process can be run on the \"Training\", \"Cross Validation\", or \"Testing\" data sets. In this step, we will measure the classification performance using the \"Training\" data set. Click the \"Test on Training Data\" button now to test the network. A dialog box will be displayed with the correct options pre-set such that the testing process will load the saved best weights (found in the last demo), run the training data set through the network, and produce a report summarizing the network classification performance. Examine the settings then click OK.",
"Button": "Test on Training Data",
"Next": "None"
},
{
"Heading": "Classification Results",
"Body": "A classification report has been generated within the active worksheet. The first thing you will notice on this report is a confusion matrix. The confusion matrix summarizes the classification results in an easy to interpret format. For example, the number of samples classified correctly as male are shown in the top-left box and the number of samples classified correctly as female are shown in the bottom-right box. The other two boxes show the samples that were classified incorrectly. This report also includes a table listing various performance measures. Notice the last measure gives the percentage of samples classified correctly for each class.",
"Button": "None",
"Next": "None"
},
{
"Heading": "Testing Data Classification",
"Body": "In the last slide, you found that the network did a good job classifying the samples that were used to train the network. However, the true test of a network is how well it can classify samples that it has not seen before. In this step we will test the networks performance on the 50 rows of data which were tagged as \"Testing\". Click the \"Test on Testing Data\" button now to test the network. A dialog box will be displayed with the correct options pre-set such that the testing process will load the saved best weights (found in the last demo), run the testing data set through the network, and produce a report summarizing the network classification performance. Examine the settings then click OK.",
"Button": "Test on Testing Data",
"Next": "None"
},
{
"Heading": "Classification Results",
"Body": "Again a classification report has been generated within the active worksheet. This report summarizes the classification results for the testing data set. Looking at the results in the confusion matrix, you can see that we have developed a good model for this data. If you want to examine the actual network output, you can go to the data sheet created by this testing process. This data sheet can be accessed via a button (Test1 Testing IOData) on the top right hand corner of the report or you can simply click on the tab with this same name. Click \"Next\" to return to the main demo panel.",
