Java Reference
In-Depth Information
This process begins by checking for any newly exported settings
by DM Lab in their repository for the associated business problem. If
settings exist, they are imported to the production system using the
executeTask operation with the appropriate ImportTask message and
the new imported settings are used to rebuild the production model
with the latest data. Otherwise, the process continues and rebuilds
the model with the current settings and the latest data available
in the system. We assume that IT automated dataset creation, used
up-to-date data collection for building, and automated testing and
applying the model. This is typically done using data warehousing
tools. The executeTask method is used to build, test, and apply the
model in the subsequent steps of the process as shown in Figure 11-6.
Before the model is applied to produce the prediction results, an
automated process will validate the test metrics with the DM Lab-
specified quality limits. If the computed test metrics are within limits
the apply operation will be performed; otherwise a notification will
be sent to DM Lab to improve the model.
Another need for the new system is getting real-time predictions
to production applications. Consider the case of a customer call
center application where an attrition model discussed in Section 7.1
is used to display customer attrition likelihood along with the customer
profile to the call center representative. The attrition likelihood mea-
sure is precomputed using batch apply with the existing customer
profile. However, when a customer calls a representative and
changes any of the attributes used to build the model, the customer
attrition likelihood displayed is no longer valid because the batch
apply does not reflect the recent changes. JDMWS record apply
allows the application to dynamically re-score the customer based on
the new customer data. Figure 11-7 shows the BPEL process used for
record apply results when the attributes are changed in the
This process begins with the retrieval of the model's signature
attributes, which are used to verify whether any of the signature
attribute values have changed in the current application session. In
the call center example, suppose the customer marital status is in the
model signature. When a customer calls the representative and says
that he was recently married and wants to add his wife as a joint
owner of his account, the process identifies that a model signature
attribute has changed and executes the record apply task with the
updated customer profile. The executeTask operation takes the newly
updated record as part of the record apply task and responds with
Search WWH ::

Custom Search