Neural Net Evolution. A neural net typically starts out with random weights; hence
it produces essentially random predictions in the first iteration. Describe how the
neural net evolves (in JMP) to produce a more accurate prediction?
Car Sales. Consider again the data on used cars ( ) with
1436 records and details on 38 attributes, including Price, Age, KM, HP, and other
specifications. The goal is to predict the price of a used Toyota Corolla based on its
specifications.
a. Determine which variables to include, and use the neural platform in JMP Pro to
fit a model. Use the validation column for validation, and use the default values
in the Neural model launch dialog. Record the RMSE for the training data and
the validation data, and save the formula for the model to the data table (use the
Save Fast Formulas option, which will save the formula as one column in the data
table). Repeat the process, changing the number of nodes (and ouly this) to 5, 10,
and 25.
i. Using your recorded values, what happens to the RMSE for the training data
as the number of nodes increases?
ii. What happens to the RMSE for the validation data?
iii. Comment on the appropriate number of nodes for the model.
iv. Use the Model Comparison platform to compare these four models (use the
Validation column as either aByvariable or as a Group variable, and focus only
on the validation data). Here, RASE is reported rather than RMSE. Compare
RASE and AAE (average absolute error) values for these four models. Which
model has the lowest “error”?
b. Conduct a sintilar experiment to assess the effect of changing the number oflayers
in the network as well as the activation functions.
