private RecommendationModel train()

in modules/ml-ext/ml/src/main/java/org/apache/ignite/ml/recommendation/RecommendationTrainer.java [157:208]
39 lines of code
12 McCabe index (conditional complexity)

    private <O extends Serializable, S extends Serializable> RecommendationModel<O, S> train(
        Dataset<EmptyContext, RecommendationDatasetData<O, S>> dataset, RecommendationModel<O, S> mdl) {
        // Collect total set of objects and subjects (their identifiers).
        Set<O> objects = dataset.compute(RecommendationDatasetData::getObjects, RecommendationTrainer::join);
        Set<S> subjects = dataset.compute(RecommendationDatasetData::getSubjects, RecommendationTrainer::join);

        // Generate initial model (object and subject matrices) initializing them with random values.
        Map<O, Vector> objMatrix = mdl == null ?
            generateRandomVectorForEach(objects, trainerEnvironment.randomNumbersGenerator()) :
            new HashMap<>(mdl.getObjMatrix());
        Map<S, Vector> subjMatrix = mdl == null ?
            generateRandomVectorForEach(subjects, trainerEnvironment.randomNumbersGenerator()) :
            new HashMap<>(mdl.getSubjMatrix());

        if (mdl != null) {
            for (O o : objects) {
                if (!objMatrix.containsKey(o))
                    objMatrix.put(o, randomVector(k, trainerEnvironment.randomNumbersGenerator()));
            }

            for (S s : subjects) {
                if (!subjMatrix.containsKey(s))
                    subjMatrix.put(s, randomVector(k, trainerEnvironment.randomNumbersGenerator()));
            }
        }

        // SGD steps.
        for (int i = 0; maxIterations == -1 || i < maxIterations; i++) {
            int seed = i;

            // Calculate gradient on reach partition and aggregate results.
            MatrixFactorizationGradient<O, S> grad = dataset.compute(
                (data, env) -> data.calculateGradient(
                    objMatrix,
                    subjMatrix,
                    batchSize,
                    seed ^ env.partition(),
                    regParam,
                    learningRate
                ),
                RecommendationTrainer::sum
            );

            if (minMdlImprovement != 0 && calculateImprovement(grad) < minMdlImprovement)
                break;

            // Apply aggregated gradient.
            grad.applyGradient(objMatrix, subjMatrix);
        }

        return new RecommendationModel<>(objMatrix, subjMatrix);
    }