Using neural networks with legacy streams

Version 14 of IBM® SPSS® Modeler introduced a new Neural Net node, supporting boosting and bagging techniques and optimization for very large datasets. Existing streams containing the old node may still build and score models in later releases. However, this support will be removed in a future release, so we recommend using the new version.

From version 13 onwards, fields with unknown values (that is, values not present in the training data) are no longer automatically treated as missing values, and are scored with the value $null$. Thus if you want to score fields with unknown values as non-null using an older (pre-13) Neural Net model in version 13 or later, you should mark unknown values as missing values (for example, by means of the Type node).

Note that, for compatibility, any legacy streams that still contain the old node may still be using the Limit set size option from Tools > Stream Properties > Options; this option only applies to Kohonen nets and K-Means nodes from version 14 onwards.