Java Local Regression Engine

Configuration

Data Source

This sample regression engine reads data from file system.

Edit the file path in engine.json, change filepath of datasource to an absolute path that points to lr_data.py

$ cat engine.json
...
"datasource": {
  "filepath": <absolute_path_to_lr_data.txt>
}
...

Algorithms

This engine comes with two algorithms.

1. OLS. It is the standard ordinary least square algorithm. It takes no parameter.
2. Default. This algorithm always return the same value v defined in the parameter.

Below is an example of algorithms defines in engine.json, it invokes three algorithms,

1. The standard OLS algorithm,
2. A default algorithm which always return 2.0,
3. A default algorithm which always return 4.0.

"algorithms": [
  { "name": "OLS", "params": {} },
  { "name": "Default", "params": { "v": 2.0 } },
  { "name": "Default", "params": { "v": 4.0 } }
]

The Serving class returns the average of all these prediction to the user.

Register engine, train, and deploy.

$ pio build
$ pio train
$ pio deploy --port 9997

Query the Engine Instance

$ curl -X POST http://localhost:9997/queries.json -d \
  '[1.80,0.87,2.41,0.35,-0.21,1.35,0.51,1.55,-0.20,1.32]'
2.2434392991944025

2.24... is the prediction result.

Evaluation

You can evaluate the engine instance too.

$  pio eval --batch JavaRegressionEval \
--metrics-class io.prediction.examples.java.regression.MeanSquareMetrics

...
2014-09-24 03:23:07,170 INFO  spark.SparkContext - Job finished: collect at Workflow.scala:695, took 0.092829 s
2014-09-24 03:23:07,284 WARN  workflow.CoreWorkflow$ - java.lang.String is not a NiceRendering instance.
2014-09-24 03:23:07,296 INFO  workflow.CoreWorkflow$ - Saved engine instance with ID: OCCUucs7QBOOG--9kIWFEw

And you will see you result in PredictionIO dashboard. To start dashboard:

$ pio dashboard