Example models with PBJ Runtimes

The library cml includes the package, models_v1. This package includes the cml_model decorator that can be used to allow a function to work as a model in a PBJ Runtime. It can also be used to enable gathering of model metrics.

The package has an optional boolean argument, called metrics, that when set to True, enables the model_metric decorator functionality from the deprecated cdsw lib. The default value of the metrics parameter is False. Therefore, this example does not have model metric gathering enabled, but the R example does. In both examples the function to be used for the model is called predict.

Python example🔗

Example #1

import cml.models_v1 as models
    
@models.cml_model
def predict(args):
    return args["x"]*2

Example #2:

import cml.models_v1 as models
    
@models.cml_model(metrics=True)
def predict(args):
    return args["x"]*2

R example🔗

Almost all of the functionality in the cdsw library for R has also been migrated to the cml library and is available automatically for every R Runtime workload. It includes a new function wrapper, called cml_model, to be used for model entry point functions in PBJ Runtimes. The function to be used for the model in this case is called predict.

library(cml)
    
predict <- cml_model(function(args) {
    return(args$x*2)
    })