This example combines the PSI capabilities with Vertical Partitioning
protocols to create a model across a subset of records found in three distinct
datasets. The model is then used to infer a result locally.

The Santander Customer Transaction Prediction dataset is split into three
independent datasets. The first dataset contains the first 40 columns, the
second dataset contains the next 60 columns, and the third contains the
remaining 100 columns. Each dataset also contains the ID_code column. These
get placed under the three example Access Points.

PSI is used to select the common records from these three datasets. Then three
models are defined for the subset of data in each dataset and combined to
produce a vertically trained model. This PyTorch model is downloaded for local
usage.

Finally, the trained model is applied to a local data to perform inference.
