Machine learning

Expanded PMML support

Vertica now supports additional PMML tags and attributes, including the FieldRef and LocalTransformations tags. For a list of all supported PMML model types, tags, sub-tags, and attributes, see PMML features and attributes.

Complex type support for imported TensorFlow models

The PREDICT_TENSORFLOW_SCALAR function makes predictions with imported TensorFlow models that have complex type input and output columns. This function accepts and outputs a single ROW, where each field in the ROW contains a 1D ARRAY tensor. For details, see the function reference page and TensorFlow integration and directory structure.

To use this function with an imported TensorFlow model, you must use a column-type argument of 1 when calling the freeze_tf2_model.py script. For more information, see tf_model_desc.json overview.