Content
|
Airfoil
The benchmark is found on Neural Designer, there is an
internal link, but it may change one day.
The dataset is Airfoil. It is experimental data with errors and uncertainties. The physical nature of the object
is not important for accuracy test. Here is an example of several records:
10000;1.5;0.3048;39.6;0.00392107;108.991
12500;1.5;0.3048;39.6;0.00392107;106.111
400;3;0.3048;71.3;0.00425727;127.564
500;3;0.3048;71.3;0.00425727;128.454
630;3;0.3048;71.3;0.00425727;129.354
First five values are features and the last one is target.
Number of records is 1503. Our test is conducted using same training/validation concept and accuracy metrics. The dataset is
divided into three parts, 60% is training set, 20% is selection set and 20% is testing set. Several models are created
using training data, the accuracy is tested on selection data. The best model is considered selected and the final
accuracy test is conducted on testing set. The accuracy metric is Pearson correlation coefficient.
In our test it is the same as reported by Neural Designer, 95 to 96%.
The execution time is about 1.5 sec. The code is not optimized for fast execution.
|
|
|