Fix PUB-1118 - AutoEncoder POJO scoring.

Also fixed a bug in initializing the activation layer for categoricals.

R/tests/Utils/shared_javapredict_DL.R

@@ -7,7 +7,7 @@ iris_train.hex <- h2o.uploadFile(conn, train)
 
 heading("Creating DL model in H2O")
 balance_classes <- if (exists("balance_classes")) balance_classes else FALSE
-iris.dl.h2o <- h2o.deeplearning(x = x, y = y, data = iris_train.hex, hidden = hidden, balance_classes = balance_classes, classification = classification, activation = activation, epochs = epochs)
+iris.dl.h2o <- h2o.deeplearning(x = x, y = y, data = iris_train.hex, seed = 1234, reproducible = T, hidden = hidden, balance_classes = balance_classes, classification = classification, activation = activation, epochs = epochs, autoencoder = autoencoder)
 print(iris.dl.h2o)
 
 heading("Downloading Java prediction model code from H2O")

R/tests/testdir_javapredict/PredictCSV.java

@@ -102,10 +102,14 @@ public static void main(String[] args) throws Exception{
         }
 
         // Print outputCSV column names.
-        output.write("predict");
-        for (int i = 0; i < model.getNumResponseClasses(); i++) {
-            output.write(",");
-            output.write(model.getDomainValues(model.getResponseIdx())[i]);
+        if (model.isAutoEncoder()) {
+          output.write(model.getHeader());
+        } else {
+            output.write("predict");
+            for (int i = 0; i < model.getNumResponseClasses(); i++) {
+                output.write(",");
+                output.write(model.getDomainValues(model.getResponseIdx())[i]);
+            }
         }
         output.write("\n");
 
@@ -132,7 +136,7 @@ public static void main(String[] args) throws Exception{
             // Parse the CSV line.  Don't handle quoted commas.  This isn't a parser test.
             String trimmedLine = line.trim();
             String[] inputColumnsArray = trimmedLine.split(",");
-            int numInputColumns = model.getNames().length-1; // we do not need response !
+            int numInputColumns = model.isAutoEncoder() ? model.getNames().length : model.getNames().length-1; // we do not need response !
             if (inputColumnsArray.length != numInputColumns) {
                 System.out.println("WARNING: Line " + lineno + " has " + inputColumnsArray.length + " columns (expected " + numInputColumns + ")");
             }
@@ -201,7 +205,7 @@ public static void main(String[] args) throws Exception{
                 } else {
                     if (i > 0) output.write(",");
                     output.write(Double.toHexString(preds[i]));
-                    if (!model.isClassifier()) break;
+                    if (!model.isClassifier() && !model.isAutoEncoder()) break;
                 }
             }
             output.write("\n");

R/tests/testdir_javapredict/runit_DL_javapredict_iris_large.R

@@ -21,14 +21,17 @@ train <- locate("smalldata/iris/iris_train.csv")
 test <- locate("smalldata/iris/iris_test.csv")
 x = c("sepal_len","sepal_wid","petal_len","petal_wid");
 y = "species"
-classification = T
 activation = "Tanh"
 epochs = 2
+autoencoder = F
 
 #----------------------------------------------------------------------
 # Run the tests
 #----------------------------------------------------------------------
 
+# CLASSIFICATION
+classification = T
+
 # large network
 hidden = c(500,500,500)
 source('../Utils/shared_javapredict_DL.R')
@@ -58,8 +61,11 @@ source('../Utils/shared_javapredict_DL.R')
 activation = "MaxoutWithDropout"
 source('../Utils/shared_javapredict_DL.R')
 
-# regression
+
+
+# REGRESSION
 classification = F
+
 activation = "Tanh"
 x = c("species","sepal_len","sepal_wid","petal_len")
 y = c("petal_wid")
@@ -84,3 +90,37 @@ source('../Utils/shared_javapredict_DL.R')
 
 activation = "MaxoutWithDropout"
 source('../Utils/shared_javapredict_DL.R')
+
+
+
+# AUTOENCODER
+autoencoder = T
+
+activation = "Rectifier"
+hidden = c(5,3,2)
+epochs = 3
+
+x = c("species","sepal_len","sepal_wid","petal_len","petal_wid");
+y = c("petal_wid") #ignored
+source('../Utils/shared_javapredict_DL.R')
+
+# only numericals
+x = c("sepal_len","sepal_wid","petal_len","petal_wid");
+y = c("petal_wid") #ignored
+source('../Utils/shared_javapredict_DL.R')
+
+# mixed numericals and categoricals
+x = c("species","sepal_len","sepal_wid","petal_len","petal_wid");
+y = c("petal_wid") #ignored
+source('../Utils/shared_javapredict_DL.R')
+
+activation = "Tanh"
+x = c("species","sepal_len","sepal_wid","petal_len","petal_wid");
+y = c("petal_wid") #ignored
+source('../Utils/shared_javapredict_DL.R')
+
+hidden = c(3)
+activation = "Tanh"
+x = c("species","sepal_len","sepal_wid","petal_len","petal_wid");
+y = c("petal_wid") #ignored
+source('../Utils/shared_javapredict_DL.R')

src/main/java/hex/deeplearning/DeepLearningModel.java

@@ -1050,6 +1050,21 @@ public String toStringAll() {
     return sb.toString();
   }
 
+  public String getHeader() {
+    assert get_params().autoencoder;
+    StringBuilder sb = new StringBuilder();
+    final int len = model_info().data_info().fullN();
+    String prefix = "reconstr_";
+    assert (model_info().data_info()._responses == 0);
+    String[] coefnames = model_info().data_info().coefNames();
+    assert (len == coefnames.length);
+    for (int c = 0; c < len; c++) {
+      if (c>0) sb.append(",");
+      sb.append(prefix + coefnames[c]);
+    }
+    return sb.toString();
+  }
+
   /**
    * This is an overridden version of Model.score(). Make either a prediction or a reconstruction.
    * @param frame Test dataset
@@ -1248,6 +1263,10 @@ private double score_autoencoder(double[] data, float[] preds, Neurons[] neurons
     DeepLearningTask.step(-1, neurons, model_info, false, null); // reconstructs data in expanded space
     float[] in  = neurons[0]._a.raw(); //input (expanded)
     float[] out = neurons[neurons.length - 1]._a.raw(); //output (expanded)
+    // DEBUGGING
+//    Log.info(Arrays.toString(data));
+//    Log.info(Arrays.toString(in));
+//    Log.info(Arrays.toString(out));
     assert(in.length == out.length);
 
     // First normalize categorical reconstructions to be probabilities
@@ -1265,6 +1284,9 @@ private double score_autoencoder(double[] data, float[] preds, Neurons[] neurons
       model_info().data_info().unScaleNumericals(out, out); //only modifies the numericals
       System.arraycopy(out, 0, preds, 0, out.length); //copy reconstruction into preds
     }
+    // DEBUGGING
+//    Log.info(Arrays.toString(preds));
+//    Log.info("");
     return l2;
   }
 
@@ -1673,6 +1695,17 @@ else if(e.num_folds > 0) {
     return true;
   }
 
+  @Override
+  protected SB toJavaNCLASSES(SB sb) {
+    return !get_params().autoencoder ? super.toJavaNCLASSES(sb) :
+            JCodeGen.toStaticVar(sb, "NCLASSES", model_info.units[model_info.units.length-1], "Number of output features (same as features of training data).");
+  }
+
+  @Override
+  protected void toJavaFillPreds0(SB bodySb) {
+    if (!get_params().autoencoder) super.toJavaFillPreds0(bodySb);
+  }
+
   public void toJavaHtml(StringBuilder sb) {
     sb.append("<br /><br /><div class=\"pull-right\"><a href=\"#\" onclick=\'$(\"#javaModel\").toggleClass(\"hide\");\'" +
             "class=\'btn btn-inverse btn-mini\'>Java Model</a></div><br /><div class=\"hide\" id=\"javaModel\">");
@@ -1733,6 +1766,13 @@ else if( model_info().size() > 100000 ) {
       layers[i] = neurons[i].units;
     JCodeGen.toStaticVar(sb, "NEURONS", layers, "Number of neurons for each layer.");
 
+    if (get_params().autoencoder) {
+      sb.i(1).p("@Override public int getPredsSize() { return " + model_info.units[model_info.units.length-1] + "; }").nl();
+      sb.i(1).p("@Override public boolean isAutoEncoder() { return true; }").nl();
+      sb.i(1).p("@Override public String getHeader() { return \"" + getHeader() + "\"; }").nl();
+
+    }
+
     // activation storage
     sb.i(1).p("// Storage for neuron activation values.").nl();
     sb.i(1).p("public static final float[][] ACTIVATION = new float[][] {").nl();
@@ -1826,6 +1866,7 @@ else if( model_info().size() > 100000 ) {
       }
       bodySb.i(0).p("}").nl();
     }
+    bodySb.i().p("java.util.Arrays.fill(ACTIVATION[0],0);").nl();
     if (cats > 0) {
       bodySb.i().p("for (i=0; i<ncats; ++i) ACTIVATION[0][CATS[i]] = 1f;").nl();
     }
@@ -1839,6 +1880,8 @@ else if( model_info().size() > 100000 ) {
     boolean relu=(get_params().activation == DeepLearning.Activation.Rectifier || get_params().activation == DeepLearning.Activation.RectifierWithDropout);
     boolean maxout=(get_params().activation == DeepLearning.Activation.Maxout || get_params().activation == DeepLearning.Activation.MaxoutWithDropout);
 
+    final String stopping = get_params().autoencoder ? "(i<=ACTIVATION.length-1)" : "(i<ACTIVATION.length-1)";
+
     // make prediction: forward propagation
     bodySb.i().p("for (i=1; i<ACTIVATION.length; ++i) {").nl();
     bodySb.i(1).p("java.util.Arrays.fill(ACTIVATION[i],0f);").nl();
@@ -1854,20 +1897,20 @@ else if( model_info().size() > 100000 ) {
     if (!maxout) {
       bodySb.i(3).p("ACTIVATION[i][r] += ACTIVATION[i-1][c] * WEIGHT[i][r*cols+c];").nl();
     } else {
-      bodySb.i(3).p("if (i<ACTIVATION.length-1) cmax = Math.max(ACTIVATION[i-1][c] * WEIGHT[i][r*cols+c], cmax);").nl();
+      bodySb.i(3).p("if " + stopping + " cmax = Math.max(ACTIVATION[i-1][c] * WEIGHT[i][r*cols+c], cmax);").nl();
       bodySb.i(3).p("else ACTIVATION[i][r] += ACTIVATION[i-1][c] * WEIGHT[i][r*cols+c];").nl();
     }
     bodySb.i(2).p("}").nl();
     if (maxout) {
-      bodySb.i(2).p("if (i<ACTIVATION.length-1) ACTIVATION[i][r] = Float.isInfinite(cmax) ? 0f : cmax;").nl();
+      bodySb.i(2).p("if "+ stopping +" ACTIVATION[i][r] = Float.isInfinite(cmax) ? 0f : cmax;").nl();
     }
     bodySb.i(2).p("ACTIVATION[i][r] += BIAS[i][r];").nl();
     if (maxout) {
-      bodySb.i(2).p("if (i<ACTIVATION.length-1) rmax = Math.max(rmax, ACTIVATION[i][r]);").nl();
+      bodySb.i(2).p("if " + stopping + " rmax = Math.max(rmax, ACTIVATION[i][r]);").nl();
     }
     bodySb.i(1).p("}").nl();
 
-    if (!maxout) bodySb.i(1).p("if (i<ACTIVATION.length-1) {").nl();
+    if (!maxout) bodySb.i(1).p("if " + stopping + " {").nl();
     bodySb.i(2).p("for (int r=0; r<ACTIVATION[i].length; ++r) {").nl();
     if (tanh) {
       bodySb.i(3).p("ACTIVATION[i][r] = 1f - 2f / (1f + (float)Math.exp(2*ACTIVATION[i][r]));").nl();
@@ -1877,7 +1920,7 @@ else if( model_info().size() > 100000 ) {
       bodySb.i(3).p("if (rmax > 1 ) ACTIVATION[i][r] /= rmax;").nl();
     }
     if (get_params().hidden_dropout_ratios != null) {
-      if (maxout) bodySb.i(1).p("if (i<ACTIVATION.length-1) {").nl();
+      if (maxout) bodySb.i(1).p("if " + stopping + " {").nl();
       bodySb.i(3).p("ACTIVATION[i][r] *= HIDDEN_DROPOUT_RATIOS[i-1];").nl();
       if (maxout) bodySb.i(1).p("}").nl();
     }
@@ -1903,18 +1946,40 @@ else if( model_info().size() > 100000 ) {
       bodySb.i(2).p("}").nl();
       bodySb.i(1).p("}").nl();
       bodySb.i().p("}").nl();
-    } else {
+    } else if (!get_params().autoencoder) { //Regression
       bodySb.i().p("}").nl();
+      bodySb.i().p("if (i == ACTIVATION.length-1) {").nl();
       // regression: set preds[1], FillPreds0 will put it into preds[0]
       if (model_info().data_info()._normRespMul != null) {
-        bodySb.i().p("preds[1] = (float) (ACTIVATION[ACTIVATION.length-1][0] / NORMRESPMUL[0] + NORMRESPSUB[0]);").nl();
+        bodySb.i().p("preds[1] = (float) (ACTIVATION[i][0] / NORMRESPMUL[0] + NORMRESPSUB[0]);").nl();
       }
       else {
-        bodySb.i().p("preds[1] = ACTIVATION[ACTIVATION.length-1][0];").nl();
+        bodySb.i().p("preds[1] = ACTIVATION[i][0];").nl();
       }
       bodySb.i().p("if (Float.isNaN(preds[1])) throw new RuntimeException(\"Predicted regression target NaN!\");").nl();
+      bodySb.i().p("}").nl();
+    } else { //AutoEncoder
+      bodySb.i(1).p("if (i == ACTIVATION.length-1) {").nl();
+      bodySb.i(2).p("for (int r=0; r<ACTIVATION[i].length; r++) {").nl();
+      bodySb.i(3).p("if (Float.isNaN(ACTIVATION[i][r]))").nl();
+      bodySb.i(4).p("throw new RuntimeException(\"Numerical instability, reconstructed NaN.\");").nl();
+      bodySb.i(3).p("preds[r] = ACTIVATION[i][r];").nl();
+      bodySb.i(2).p("}").nl();
+      if (model_info().data_info()._nums > 0) {
+        int ns = model_info().data_info().numStart();
+        bodySb.i(2).p("for (int k=" + ns + "; k<" + model_info().data_info().fullN() + "; ++k) {").nl();
+        bodySb.i(3).p("preds[k] = preds[k] / (float)NORMMUL[k-" + ns + "] + (float)NORMSUB[k-" + ns + "];").nl();
+        bodySb.i(2).p("}").nl();
+      }
+      bodySb.i(1).p("}").nl();
+      bodySb.i().p("}").nl();
+      // DEBUGGING
+//      bodySb.i().p("System.out.println(java.util.Arrays.toString(data));").nl();
+//      bodySb.i().p("System.out.println(java.util.Arrays.toString(ACTIVATION[0]));").nl();
+//      bodySb.i().p("System.out.println(java.util.Arrays.toString(ACTIVATION[ACTIVATION.length-1]));").nl();
+//      bodySb.i().p("System.out.println(java.util.Arrays.toString(preds));").nl();
+//      bodySb.i().p("System.out.println(\"\");").nl();
     }
-
     fileCtxSb.p(model);
     toJavaUnifyPreds(bodySb);
     toJavaFillPreds0(bodySb);

src/main/java/water/Model.java

@@ -661,7 +661,7 @@ protected SB toJavaSuper( SB sb ) {
     return sb;
   }
   private SB toJavaNAMES( SB sb ) { return JCodeGen.toStaticVar(sb, "NAMES", _names, "Names of columns used by model."); }
-  private SB toJavaNCLASSES( SB sb ) { return isClassifier() ? JCodeGen.toStaticVar(sb, "NCLASSES", nclasses(), "Number of output classes included in training data response column.") : sb; }
+  protected SB toJavaNCLASSES( SB sb ) { return isClassifier() ? JCodeGen.toStaticVar(sb, "NCLASSES", nclasses(), "Number of output classes included in training data response column.") : sb; }
   private SB toJavaDOMAINS( SB sb, SB fileContextSB ) {
     sb.nl();
     sb.ii(1);

src/main/java/water/api/DeepLearningModelView.java

@@ -43,8 +43,7 @@ public static Response redirect(Request req, Key modelKey) {
   }
   @Override protected String serveJava() {
     deeplearning_model = UKV.get(_modelKey);
-    if (deeplearning_model!=null
-            && !deeplearning_model.get_params().autoencoder) //not yet implemented
+    if (deeplearning_model!=null)
       return deeplearning_model.toJava();
     else
       return "";

src/main/java/water/genmodel/GeneratedModel.java

@@ -34,6 +34,8 @@ public static boolean grpContains(byte[] gcmp, int offset, int idx) {
     return (gcmp[idx >> 3] & ((byte)1 << (idx % 8))) != 0;
   }
 
+  public String getHeader() { return null; }
+  public boolean isAutoEncoder() { return false; }
   @Override public int getColIdx(String name) {
     String[] names = getNames();
     for (int i=0; i<names.length; i++) if (names[i].equals(name)) return i;