IBM Data WH KNN Nugget - Settings Tab

On the Settings tab, you can set options for scoring the model.

Distance measure. The method to be used for measuring the distance between data points; greater distances indicate greater dissimilarities. The options are:

• Euclidean. (default) The distance between two points is computed by joining them with a straight line.
• Manhattan. The distance between two points is calculated as the sum of the absolute differences between their co-ordinates.
• Canberra. Similar to Manhattan distance, but more sensitive to data points closer to the origin.
• Maximum. The distance between two points is calculated as the greatest of their differences along any coordinate dimension.

Number of Nearest Neighbors (k). The number of nearest neighbors for a particular case. Note that using a greater number of neighbors will not necessarily result in a more accurate model.

The choice of k controls the balance between the prevention of overfitting (this may be important, particularly for "noisy" data) and resolution (yielding different predictions for similar instances). You will usually have to adjust the value of k for each data set, with typical values ranging from 1 to several dozen.

Include input fields. If selected, this option passes all the original input fields downstream, appending the extra modeling field or fields to each row of data. If you clear this check box, only the Record ID field and the extra modeling fields are passed on, and so the stream runs more quickly.

Standardize measurements before calculating distance. If selected, this option standardizes the measurements for continuous input fields before calculating the distance values.

Use coresets to increase performance for large datasets. If selected, this option uses core set sampling to speed up the calculation when large data sets are involved.