ceh-pmml-models

Example Jupyter notebook can be viewed here. It can be executed online here.

Models

Models are in the models folder. The two PMML files are:

• rf_segmentation.pmml - a "large" model, with 1000 trees (18Mb)
• rf_segmentation_small.pmml - a "smaller" model, with 162 trees (2.7Mb)
• rf_segmentation_sklearn.pmml - an sklearn model using the additional events feature

Dataset

A sample dataset is available here.

Inputs

For the rf_segmentation_sklearn.pmml model the expected inputs are:

• age as a double
• income as a double (in terms of thousands/year, e.g. 100.0)
• response as an integer, representing (0 no response, 1 response)
• events as an integer, the campaign number coded as:
  • Airlines, 0
  • Merchandise, 1
  • Hotel, 2
  • Online purchase, 3
  • Utilities, 4
  • Restaurantes, 5
  • Others, 6

The outputs are the probabilities of each segment (low, medium, high).

Predictions

Sample output from jpmml with the rf_segmentation_small.pmml model:

------------------------------------------------------------------------------
{age=ContinuousDouble{opType=continuous, dataType=double, value=32.21556819580098}, income=ContinuousDouble{opType=continuous, dataType=double, value=5.370116396577485}, response=ContinuousDouble{opType=continuous, dataType=double, value=0.0}}
{segment=0, probability_0=0.9032258064516129, probability_1=0.0967741935483871, probability_2=0.0, predicted_segment=0}
------------------------------------------------------------------------------
{age=ContinuousDouble{opType=continuous, dataType=double, value=30.23771420174174}, income=ContinuousDouble{opType=continuous, dataType=double, value=116.99344386773195}, response=ContinuousDouble{opType=continuous, dataType=double, value=1.0}}
{segment=1, probability_0=0.0, probability_1=1.0, probability_2=0.0, predicted_segment=1}
------------------------------------------------------------------------------
{age=ContinuousDouble{opType=continuous, dataType=double, value=16.658638021134774}, income=ContinuousDouble{opType=continuous, dataType=double, value=221.54952628874256}, response=ContinuousDouble{opType=continuous, dataType=double, value=1.0}}
{segment=2, probability_0=0.0, probability_1=0.04838709677419355, probability_2=0.9516129032258065, predicted_segment=2}
------------------------------------------------------------------------------

Confidence can be extracted by matching with the predicted output label, e.g. for the first item above, predicted_segment=0, therefore the "confidence" is probability_0=0.9032258064516129.