#157 Next try on training data

b0n541 · Sep 9, 2024 · a1f6639 · a1f6639
1 parent 15b167c
commit a1f6639
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 9 deletions.
diff --git a/jskat-base/src/main/java/org/jskat/control/iss/IssGameExtractor.java b/jskat-base/src/main/java/org/jskat/control/iss/IssGameExtractor.java
@@ -45,7 +45,7 @@ private void filterGameDatabase(final Predicate<SkatGameData> predicate, final S
         try (final Stream<String> stream = Files.lines(Paths.get(sourceFileName))) {
             final AtomicInteger count = new AtomicInteger();
             final var filteredGames = stream
-                    .skip(9_000_000)
+                    .skip(6_000_000)
                     .peek(logProgress(count))
                     //.peek(System.out::println)
                     // TODO: fix parsing of filtered games
@@ -78,7 +78,7 @@ private void filterGameDatabase(final Predicate<SkatGameData> predicate, final S
                     .filter(predicate)
                     //.map(SkatGameData::toString)
                     .map(NETWORK_INPUTS)
-                    .limit(100)
+                    .limit(100_000)
                     .collect(Collectors.toList());
 
             final var lines = new ArrayList<String>();

diff --git a/jskat-base/src/main/kotlin/org/jskat/ai/deeplearning/BidNetTrainer.kt b/jskat-base/src/main/kotlin/org/jskat/ai/deeplearning/BidNetTrainer.kt
@@ -11,7 +11,7 @@ import ai.djl.nn.core.Linear
 import ai.djl.training.DefaultTrainingConfig
 import ai.djl.training.EasyTrain
 import ai.djl.training.evaluator.Accuracy
-import ai.djl.training.initializer.NormalInitializer
+import ai.djl.training.initializer.UniformInitializer
 import ai.djl.training.listener.TrainingListener
 import ai.djl.training.loss.Loss
 import ai.djl.training.optimizer.Optimizer
@@ -25,7 +25,7 @@ fun main() {
     val batchSize = 10_000
     val builder = DataFrameDataSet.Builder()
     builder.filePath = "data/kermit_games.csv"
-    builder.setSampling(10, true)
+    builder.setSampling(batchSize, true)
     builder.addCategoricalFeature("declarer", true)
     listOf(
         "♣A", "♣T", "♣K", "♣Q", "♣J", "♣9", "♣8", "♣7",
@@ -51,13 +51,13 @@ fun main() {
 
     val featurizer2 = dataSet.labels[0].featurizer
     println("Featurizer gameType: ${featurizer2.dataRequired()}")
-    val defeature2 = featurizer2.deFeaturize(floatArrayOf(1.0f, 0.0f, 0.0f, 0.0f, 0.0f, 0.7f))
+    val defeature2 = featurizer2.deFeaturize(floatArrayOf(1.0f, 0.1f, 0.2f, 0.3f, 0.4f, 0.7f))
     println("DeFeature: $defeature2")
     if (defeature2 is Classifications) {
         println(defeature2.best<Classifications.Classification?>().className)
     }
 
-    val trainTest = dataSet.randomSplit(8, 2)
+    val trainTest = dataSet.randomSplit(80, 20)
     val training = trainTest[0]
     val test = trainTest[1]
 
@@ -81,12 +81,16 @@ fun main() {
 //    block.add(Activation::relu)
 //    block.add(Linear.builder().setUnits(512).build())
 //    block.add(Activation::relu)
-    block.add(Linear.builder().setUnits(256).build())
-    block.add(Activation::relu)
+//    block.add(Linear.builder().setUnits(256).build())
+//    block.add(Activation::relu)
     block.add(Linear.builder().setUnits(128).build())
     block.add(Activation::relu)
+    block.add(Linear.builder().setUnits(64).build())
+    block.add(Activation::relu)
+    block.add(Linear.builder().setUnits(32).build())
+    block.add(Activation::relu)
     block.add(Linear.builder().setUnits(outputSize).build())
-    block.setInitializer(NormalInitializer(), Parameter.Type.WEIGHT)
+    block.setInitializer(UniformInitializer(), Parameter.Type.WEIGHT)
 
     val model = Model.newInstance("bidnet")
     model.block = block