Revert "GH-15855: core pipeline API (#16039)"

This reverts commit c15ea1e.

h2o-admissibleml/src/main/java/hex/Infogram/Infogram.java

@@ -1,9 +1,6 @@
 package hex.Infogram;
 
 import hex.*;
-import hex.Infogram.InfogramModel.InfogramModelOutput;
-import hex.Infogram.InfogramModel.InfogramParameters;
-import hex.ModelMetrics.MetricBuilder;
 import water.*;
 import water.exceptions.H2OModelBuilderIllegalArgumentException;
 import water.fvec.Frame;
@@ -21,8 +18,8 @@
 import static water.util.ArrayUtils.sort;
 import static water.util.ArrayUtils.sum;
 
-public class Infogram extends ModelBuilder<hex.Infogram.InfogramModel, InfogramParameters,
-        InfogramModelOutput> {
+public class Infogram extends ModelBuilder<hex.Infogram.InfogramModel, hex.Infogram.InfogramModel.InfogramParameters,
+        hex.Infogram.InfogramModel.InfogramModelOutput> {
   static final double NORMALIZE_ADMISSIBLE_INDEX = 1.0/Math.sqrt(2.0);
   boolean _buildCore;       // true to find core predictors, false to find admissible predictors
   String[] _topKPredictors; // contain the names of top predictors to consider for infogram
@@ -48,14 +45,14 @@ public class Infogram extends ModelBuilder<hex.Infogram.InfogramModel, InfogramP
   Model.Parameters.FoldAssignmentScheme _foldAssignmentOrig = null;
   String _foldColumnOrig = null;
 
-  public Infogram(boolean startup_once) { super(new InfogramParameters(), startup_once);}
+  public Infogram(boolean startup_once) { super(new hex.Infogram.InfogramModel.InfogramParameters(), startup_once);}
 
-  public Infogram(InfogramParameters parms) {
+  public Infogram(hex.Infogram.InfogramModel.InfogramParameters parms) {
     super(parms);
     init(false);
   }
 
-  public Infogram(InfogramParameters parms, Key<hex.Infogram.InfogramModel> key) {
+  public Infogram(hex.Infogram.InfogramModel.InfogramParameters parms, Key<hex.Infogram.InfogramModel> key) {
     super(parms, key);
     init(false);
   }
@@ -74,23 +71,18 @@ protected int nModelsInParallel(int folds) {
    * This is called before cross-validation is carried out
    */
   @Override
-  protected void cv_init() {
-    super.cv_init();
+  public void computeCrossValidation() {
     info("cross-validation", "cross-validation infogram information is stored in frame with key" +
             " labeled as admissible_score_key_cv and the admissible features in admissible_features_cv.");
     if (error_count() > 0) {
       throw H2OModelBuilderIllegalArgumentException.makeFromBuilder(Infogram.this);
     }
-  }
-
-  @Override
-  protected MetricBuilder makeCVMetricBuilder(ModelBuilder<InfogramModel, InfogramParameters, InfogramModelOutput> cvModelBuilder, Futures fs) {
-    return null;   //infogram does not support scoring
+    super.computeCrossValidation();
   }
 
   // find the best alpha/lambda values used to build the main model moving forward by looking at the devianceValid
   @Override
-  protected void cv_computeAndSetOptimalParameters(ModelBuilder[] cvModelBuilders) {
+  public void cv_computeAndSetOptimalParameters(ModelBuilder[] cvModelBuilders) {
     int nBuilders = cvModelBuilders.length;
     double[][] cmiRaw = new double[nBuilders][];
     List<List<String>> columns = new ArrayList<>();
@@ -111,12 +103,7 @@ protected void cv_computeAndSetOptimalParameters(ModelBuilder[] cvModelBuilders)
     }
     _cvDone = true;  // cv is done and we are going to build main model next
   }
-
-  @Override
-  protected void cv_mainModelScores(int N, MetricBuilder[] mbs, ModelBuilder<InfogramModel, InfogramParameters, InfogramModelOutput>[] cvModelBuilders) {
-    //infogram does not support scoring
-  }
-
+
   public void calculateMeanInfogramInfo(double[][] cmiRaw, List<List<String>> columns,
                                         long[] nObs) {
     int nFolds = cmiRaw.length;
@@ -317,7 +304,7 @@ public final void buildModel() {
       try {
         boolean validPresent = _parms.valid() != null;
         prepareModelTrainingFrame(); // generate training frame with predictors and sensitive features (if specified)
-        InfogramModel model = new hex.Infogram.InfogramModel(dest(), _parms, new InfogramModelOutput(Infogram.this));
+        InfogramModel model = new hex.Infogram.InfogramModel(dest(), _parms, new hex.Infogram.InfogramModel.InfogramModelOutput(Infogram.this));
         _model = model.delete_and_lock(_job);
         _model._output._start_time = System.currentTimeMillis();
         _cmiRaw = new double[_numModels];
@@ -372,7 +359,7 @@ public final void buildModel() {
      * relevance >= relevance_threshold.  Derive _admissible_index as distance from point with cmi = 1 and 
      * relevance = 1.  In addition, all arrays are sorted on _admissible_index.
      */
-    private void copyCMIRelevance(InfogramModelOutput modelOutput) {
+    private void copyCMIRelevance(InfogramModel.InfogramModelOutput modelOutput) {
       modelOutput._cmi_raw = new double[_cmi.length];
       System.arraycopy(_cmiRaw, 0, modelOutput._cmi_raw, 0, modelOutput._cmi_raw.length);
       modelOutput._admissible_index = new double[_cmi.length];
@@ -388,7 +375,7 @@ private void copyCMIRelevance(InfogramModelOutput modelOutput) {
               modelOutput._admissible_index, modelOutput._admissible, modelOutput._all_predictor_names);
     }
 
-    public void copyCMIRelevanceValid(InfogramModelOutput modelOutput) {
+    public void copyCMIRelevanceValid(InfogramModel.InfogramModelOutput modelOutput) {
       modelOutput._cmi_raw_valid = new double[_cmiValid.length];
       System.arraycopy(_cmiRawValid, 0, modelOutput._cmi_raw_valid, 0, modelOutput._cmi_raw_valid.length);
       modelOutput._admissible_index_valid = new double[_cmiValid.length];

h2o-algos/src/main/java/hex/deeplearning/DeepLearning.java

@@ -165,7 +165,7 @@ static DataInfo makeDataInfo(Frame train, Frame valid, DeepLearningParameters pa
     }
   }
 
-  @Override protected void cv_computeAndSetOptimalParameters(ModelBuilder[] cvModelBuilders) {
+  @Override public void cv_computeAndSetOptimalParameters(ModelBuilder[] cvModelBuilders) {
     _parms._overwrite_with_best_model = false;
 
     if( _parms._stopping_rounds == 0 && _parms._max_runtime_secs == 0) return; // No exciting changes to stopping conditions

h2o-algos/src/main/java/hex/ensemble/StackedEnsembleModel.java

@@ -580,7 +580,7 @@ public ModelMetrics makeModelMetrics(Frame fr, Frame adaptFrm) {
     @Override
     public ModelMetrics.MetricBuilder<?> getMetricBuilder() {
       throw new UnsupportedOperationException("Stacked Ensemble model doesn't implement MetricBuilder infrastructure code, " +
-              "retrieve your metrics by calling makeModelMetrics method.");
+              "retrieve your metrics by calling getOrMakeMetrics method.");
     }
   }
 

h2o-algos/src/main/java/hex/glm/GLM.java

@@ -26,6 +26,7 @@
 import hex.util.LinearAlgebraUtils;
 import hex.util.LinearAlgebraUtils.BMulTask;
 import hex.util.LinearAlgebraUtils.FindMaxIndex;
+import jsr166y.CountedCompleter;
 import org.joda.time.format.DateTimeFormat;
 import org.joda.time.format.DateTimeFormatter;
 import water.*;
@@ -118,8 +119,7 @@ public boolean isSupervised() {
   public ModelCategory[] can_build() {
     return new ModelCategory[]{
       ModelCategory.Regression,
-      ModelCategory.Binomial, 
-      ModelCategory.Multinomial
+      ModelCategory.Binomial,
     };
   }
 
@@ -148,12 +148,13 @@ public ModelCategory[] can_build() {
    * (builds N+1 models, all have train+validation metrics, the main model has N-fold cross-validated validation metrics)
    */
   @Override
-  protected void cv_init() {
+  public void computeCrossValidation() {
     // init computes global list of lambdas
     init(true);
     _cvRuns = true;
     if (error_count() > 0)
       throw H2OModelBuilderIllegalArgumentException.makeFromBuilder(GLM.this);
+    super.computeCrossValidation();
   }
 
 
@@ -292,7 +293,7 @@ private double[] alignSubModelsAcrossCVModels(ModelBuilder[] cvModelBuilders) {
    *   4. unlock the n-folds models (they are changed here, so the unlocking happens here)
    */
   @Override
-  protected void cv_computeAndSetOptimalParameters(ModelBuilder[] cvModelBuilders) {
+  public void cv_computeAndSetOptimalParameters(ModelBuilder[] cvModelBuilders) {
       setMaxRuntimeSecsForMainModel();
       double bestTestDev = Double.POSITIVE_INFINITY;
       double[] alphasAndLambdas = alignSubModelsAcrossCVModels(cvModelBuilders);
@@ -371,6 +372,12 @@ protected void cv_computeAndSetOptimalParameters(ModelBuilder[] cvModelBuilders)
           break;
         }
       }
+      for (int i = 0; i < cvModelBuilders.length; ++i) {
+        GLM g = (GLM) cvModelBuilders[i];
+        if (g._toRemove != null)
+          for (Key k : g._toRemove)
+            Keyed.remove(k);
+      }
 
       for (int i = 0; i < cvModelBuilders.length; ++i) {
         GLM g = (GLM) cvModelBuilders[i];
@@ -1536,11 +1543,11 @@ private void buildModel() {
 
   protected static final long WORK_TOTAL = 1000000;
 
-  @Override
-  protected void cleanUp() {
-    if (_parms._lambda_search && _parms._is_cv_model)
-      keepUntilCompletion(_dinfo.getWeightsVec()._key);
-    super.cleanUp();
+  transient Key [] _toRemove;
+
+  private Key[] removeLater(Key ...k){
+    _toRemove = _toRemove == null?k:ArrayUtils.append(_toRemove,k);
+    return k;
   }
 
   @Override protected GLMDriver trainModelImpl() { return _driver = new GLMDriver(); }
@@ -1569,6 +1576,23 @@ public final class GLMDriver extends Driver implements ProgressMonitor {
     private transient GLMTask.GLMIterationTask _gramInfluence;
     private transient double[][] _cholInvInfluence;
 
+    private void doCleanup() {
+      try {
+        if (_parms._lambda_search && _parms._is_cv_model)
+          Scope.untrack(removeLater(_dinfo.getWeightsVec()._key));
+        if (_parms._HGLM) {
+          Key[] vecKeys = _toRemove;
+          for (int index = 0; index < vecKeys.length; index++) {
+            Vec tempVec = DKV.getGet(vecKeys[index]);
+            tempVec.remove();
+          }
+        }
+      } catch (Exception e) {
+        Log.err("Error while cleaning up GLM " + _result);
+        Log.err(e);
+      }
+    }
+
     private transient Cholesky _chol;
     private transient L1Solver _lslvr;
 
@@ -3540,8 +3564,9 @@ private Vec[] genGLMVectors(DataInfo dinfo, double[] nb) {
           sumExp += Math.exp(nb[i * N + P] - maxRow);
       }
       Vec[] vecs = dinfo._adaptedFrame.anyVec().makeDoubles(2, new double[]{sumExp, maxRow});
-      if (_parms._lambda_search) {
-        track(vecs[0]); track(vecs[1]);
+      if (_parms._lambda_search && _parms._is_cv_model) {
+        Scope.untrack(vecs[0]._key, vecs[1]._key);
+        removeLater(vecs[0]._key, vecs[1]._key);
       }
       return vecs;
     }
@@ -3823,7 +3848,7 @@ private void checkCoeffsBounds() {
      * - column 2: zi, intermediate values
      * - column 3: eta = X*beta, intermediate values
      */
-    private void addWdataZiEtaOld2Response() { // attach wdata, zi, eta to response for HGLM
+    public void addWdataZiEtaOld2Response() { // attach wdata, zi, eta to response for HGLM
       int moreColnum = 3 + _parms._random_columns.length;
       Vec[] vecs = _dinfo._adaptedFrame.anyVec().makeZeros(moreColnum);
       String[] colNames = new String[moreColnum];
@@ -3836,11 +3861,25 @@ private void addWdataZiEtaOld2Response() { // attach wdata, zi, eta to response
         vecs[index] = _parms.train().vec(randColIndices[index - 3]).makeCopy();
       }
       _dinfo.addResponse(colNames, vecs);
-      Frame wdataZiEta = new Frame(Key.make("wdataZiEta"+Key.rand()), colNames, vecs);
-      DKV.put(wdataZiEta);
-      track(wdataZiEta);
+      for (int index = 0; index < moreColnum; index++) {
+        Scope.untrack(vecs[index]._key);
+        removeLater(vecs[index]._key);
+      }
+    }
+
+    @Override
+    public void onCompletion(CountedCompleter caller) {
+      doCleanup();
+      super.onCompletion(caller);
     }
 
+    @Override
+    public boolean onExceptionalCompletion(Throwable t, CountedCompleter caller) {
+      doCleanup();
+      return super.onExceptionalCompletion(t, caller);
+    }
+
+
     @Override
     public boolean progress(double[] beta, GradientInfo ginfo) {
       _state._iter++;

h2o-algos/src/main/java/hex/kmeans/KMeans.java

@@ -109,7 +109,7 @@ else if( user_points.numRows() != _parms._k)
     if (expensive && error_count() == 0) checkMemoryFootPrint();
   }
 
-  protected void cv_makeAggregateModelMetrics(ModelMetrics.MetricBuilder[] mbs){
+  public void cv_makeAggregateModelMetrics(ModelMetrics.MetricBuilder[] mbs){
     super.cv_makeAggregateModelMetrics(mbs);
     ((ModelMetricsClustering.MetricBuilderClustering) mbs[0])._within_sumsqe = null;
     ((ModelMetricsClustering.MetricBuilderClustering) mbs[0])._size = null;

h2o-algos/src/main/java/hex/tree/SharedTree.java

@@ -1203,7 +1203,7 @@ public  double initialValue() {
     return _parms._parallel_main_model_building;
   }
 
-  @Override protected void cv_computeAndSetOptimalParameters(ModelBuilder<M, P, O>[] cvModelBuilders) {
+  @Override public void cv_computeAndSetOptimalParameters(ModelBuilder<M, P, O>[] cvModelBuilders) {
     // Extract stopping conditions from each CV model, and compute the best stopping answer
     if (!cv_initStoppingParameters())
       return; // No exciting changes to stopping conditions

h2o-bindings/bin/custom.py

@@ -25,9 +25,8 @@ def get_customizations_for(language, algo, property=None, default=None):
         tokens = property.split('.')
         value = customizations
         for token in tokens:
-            if token in value:
-                value = value.get(token)
-            else:        
+            value = value.get(token)
+            if value is None:
                 return default
         return value
     else:

h2o-bindings/bin/gen_R.py

@@ -145,8 +145,8 @@ def get_schema_params(pname):
                         "verbose",
                         "destination_key"] # destination_key is only for SVD
     bulk_params = list(zip(*filter(lambda t: not t[0] in bulk_pnames_skip, zip(sig_pnames, sig_params))))
-    bulk_pnames = list(bulk_params[0]) if bulk_params else []
-    sig_bulk_params = list(bulk_params[1]) if bulk_params else []
+    bulk_pnames = list(bulk_params[0])
+    sig_bulk_params = list(bulk_params[1])
     sig_bulk_params.append("segment_columns = NULL")
     sig_bulk_params.append("segment_models_id = NULL")
     sig_bulk_params.append("parallelism = 1")