Splitting Training Files Across Multiple MLL Scripts

Help Request

Use case

split training file into multiple separate files, but do the writing part in a single file

Minimal Working Example

ackley_training.mll:

import "definitions" from de.evoal.surrogate.ml;
import "definitions" from de.evoal.surrogate.smile.ml;
import "data" from surrogate;

module ackley_training {
	prediction svr_ackley
		maps 'x:0'
	    to 'y:0'
	    using
	    	layer transfer
				with function 'gaussian-svr'
					mapping 'x:0'
					to 'y:0'
					with parameters
						'ε' := 1.0; // default 1.4
						'σ' := 2.0; // default 1.0
						'soft-margin' := 0.3; // default 0.15
						tolerance := 0.05; // default 0.1
}

all_training.mll:

import "definitions" from de.evoal.surrogate.ml;
import "definitions" from de.evoal.surrogate.smile.ml;
import "data" from surrogate;
import "svr_ackley" from ackley_training;

module all_training {
				
	predict svr_ackley from "../generated/pure/samples-00008/sigma-1/all.json"
	and measure
			'cross-validation'(10);
			'R²'();
			'rmse'();
	end
	and store to "../generated/pure/samples-00008/sigma-1/gaussian_ackley.pson"

	predict svr_rastrigin from "../generated/pure/samples-00008/sigma-1/all.json"
	and measure
			'cross-validation'(10);
			'R²'();
			'rmse'();
	end
	and store to "../generated/pure/samples-00008/sigma-1/gaussian_rastrigin.pson"

	predict svr_rosenbrock from "../generated/pure/samples-00008/sigma-1/all.json"
	and measure
			'cross-validation'(10);
			'R²'();
			'rmse'();
	end
	and store to "../generated/pure/samples-00008/sigma-1/gaussian_rosenbrock.pson"

	predict svr_weightedsphere from "../generated/pure/samples-00008/sigma-1/all.json"
	and measure
			'cross-validation'(10);
			'R²'();
			'rmse'();
	end
	and store to "../generated/pure/samples-00008/sigma-1/gaussian_weightedsphere.pson"
}

What is your question or problem

It was mentioned this would be possible, but i am not sure right now how to do this properly, this are the relevant error messages i get:

13:22:09.450 [main] ERROR d.e.s.m.c.MachineLearningModuleProducer - Error while processing rule 'all_training.mll': XtextLinkingDiagnostic: null:8 Couldn't resolve reference to SurrogateDefinition 'svr_ackley'.
13:22:09.454 [main] ERROR d.e.s.m.c.MachineLearningModuleProducer - Error while processing rule 'all_training.mll': XtextLinkingDiagnostic: null:16 Couldn't resolve reference to SurrogateDefinition 'svr_rastrigin'.
13:22:09.454 [main] ERROR d.e.s.m.c.MachineLearningModuleProducer - Error while processing rule 'all_training.mll': XtextLinkingDiagnostic: null:24 Couldn't resolve reference to SurrogateDefinition 'svr_rosenbrock'.
13:22:09.454 [main] ERROR d.e.s.m.c.MachineLearningModuleProducer - Error while processing rule 'all_training.mll': XtextLinkingDiagnostic: null:32 Couldn't resolve reference to SurrogateDefinition 'svr_weightedsphere'.

Why would i want to do this

i have the following prediction generator file:

import "definitions" from de.evoal.core.math;
import "definitions" from de.evoal.generator.generator;
import "definitions" from de.evoal.surrogate.generator;
import "data" from surrogate;


module surrogate {
	/**
	 * Introduce a simple pipeline that generates some test data using
	 *   the ackley function.
	 */
	pipeline 'main-pipeline' [
		/**
		 * First, we generate some normally distributed data.
		 */
		step {
			component 'normal-distribution' {
				'μ' := 0.0;
				'σ' := 1;
			}		
			
			writes [data 'x:0', data 'x:1', data 'x:2', data 'x:3', data 'x:4'];

		}
	    step {
	        component 'surrogate' {
	        }
	
	        reads [data 'x:0'];
	        writes [data 'y:0'];
	    }
		step {
	        component 'surrogate' {
	        }
	
	        reads [data 'x:1'];
	        writes [data 'y:1'];
	    }
		step {
	        component 'surrogate' {
	        }
	
	        reads [data 'x:2', data 'x:3'];
	        writes [data 'y:2'];
	    }
		step {
	        component 'surrogate' {
	        }
	
	        reads [data 'x:4'];
	        writes [data 'y:3'];
	    }
	]
	
	write "../generated/pure/samples-16384/sigma-1/predicted.json" with 16384 samples from executing [ pipeline 'main-pipeline'];
}

this was the old training file:

import "definitions" from de.evoal.surrogate.ml;
import "definitions" from de.evoal.surrogate.smile.ml;
import "data" from surrogate;

module all_training {
	prediction svr_ackley
		maps 'x:0'
	    to 'y:0'
	    using
	    	layer transfer
				with function 'gaussian-svr'
					mapping 'x:0'
					to 'y:0'
					with parameters
						'ε' := 1.0; // default 1.4
						'σ' := 2.0; // default 1.0
						'soft-margin' := 0.3; // default 0.15
						tolerance := 0.05; // default 0.1

	prediction svr_rastrigin
		maps 'x:1'
	    to 'y:1'
	    using
	    	layer transfer
				with function 'gaussian-svr'
					mapping 'x:1'
					to 'y:1'
					with parameters
						'ε' := 1.0; // default 0.5
						'σ' := 2.0; // default 1.0
						'soft-margin' := 0.3; // default 0.15
						tolerance := 0.05; // default 0.1

	prediction svr_rosenbrock
		maps 'x:2', 'x:3'
	    to 'y:2'
	    using
	    	layer transfer
				with function 'gaussian-svr'
					mapping 'x:2', 'x:3'
					to 'y:2'
					with parameters
						'ε' := 1.0; // default 0.5
						'σ' := 2.0; // default 1.0
						'soft-margin' := 0.3; // default 0.15
						tolerance := 0.05; // default 0.1

	prediction svr_weightedsphere
		maps 'x:4'
	    to 'y:3'
	    using
	    	layer transfer
				with function 'gaussian-svr'
					mapping 'x:4'
					to 'y:3'
					with parameters
						'ε' := 1.0; // default 0.5
						'σ' := 2.0; // default 1.0
						'soft-margin' := 0.3; // default 0.15
						tolerance := 0.05; // default 0.1

						
	predict svr_ackley from "../generated/pure/samples-00008/sigma-1/all.json"
	and measure
			'cross-validation'(10);
			'R²'();
			'rmse'();
	end
	and store to "../generated/pure/samples-00008/sigma-1/gaussian_ackley.pson"

	predict svr_rastrigin from "../generated/pure/samples-00008/sigma-1/all.json"
	and measure
			'cross-validation'(10);
			'R²'();
			'rmse'();
	end
	and store to "../generated/pure/samples-00008/sigma-1/gaussian_rastrigin.pson"

	predict svr_rosenbrock from "../generated/pure/samples-00008/sigma-1/all.json"
	and measure
			'cross-validation'(10);
			'R²'();
			'rmse'();
	end
	and store to "../generated/pure/samples-00008/sigma-1/gaussian_rosenbrock.pson"

	predict svr_weightedsphere from "../generated/pure/samples-00008/sigma-1/all.json"
	and measure
			'cross-validation'(10);
			'R²'();
			'rmse'();
	end
	and store to "../generated/pure/samples-00008/sigma-1/gaussian_weightedsphere.pson"
}

and this is the shell file:

#!/bin/bash

export EVOAL_HOME=$( cd -- "$(dirname $0)/../../../" >/dev/null 2>&1 ; pwd -P )

$SHELL $EVOAL_HOME/bin/evoal-generator.sh . prediction.generator  -Bsurrogate:configuration-file=all_training.mll -Bsurrogate:pre-trained=../generated/pure/samples-00008/sigma-1/gaussian_ackley.pson

after using the shell file, it correctly predicts the values for the first model (ackley), but the other values are empty in the json file