Add after method for drop(n).next(); easier for non-Scala folks

README.md

@@ -114,7 +114,7 @@ Training the model reduces to a termination condition on this iterator; here aft
 A model checkpointer serializes the final train state object.
 
 ```scala
-val finalState = trainTrajectory.drop(numIterations).next()
+val finalState = trainTrajectory.after(numIterations)
 
 TensorTreeCheckpointer.newIn(checkpointRoot).save(finalState, numIterations)
 ```
@@ -151,7 +151,7 @@ The user code composes these core modules into custom architectures given the us
 | `deepwit.init` | Xavier/Glorot normal and uniform, for matrices and vectors |
 | `deepwit.regularization` | `Perturbation` — thinning (dropout) as a mutation of the weights that *read* a feature |
 | `deepwit.optimizer` | `LearningRateSchedule` (constant, linear warmup, cosine decay), `LearningRateScheduler`, `clipGlobalNorm` |
-| `deepwit.training` | `Monitor` (step, loss, throughput, learning rate), `tapEvery` |
+| `deepwit.training` | `Monitor` (step, loss, throughput, learning rate), `tapEvery`, `after` |
 | `deepwit.checkpointing` | `TensorTreeCheckpointer` — save and load any `TensorTree` by iteration |
 
 ## Relationship to DimWit

core/src/main/scala/deepwit/training/package.scala

@@ -10,11 +10,7 @@ extension [T](it: Iterator[T])
         if id > 0 && id % n == 0 then f(t, id)
       .map(_._1)
 
-extension [T](it: LazyList[T])
-
-  def tapEvery(n: Int)(f: (T, Int) => Unit): LazyList[T] =
-    it
-      .zipWithIndex
-      .tapEach: (t, id) =>
-        if id > 0 && id % n == 0 then f(t, id)
-      .map(_._1)
+  /** The state after n iterations: Advances the iterator n steps and returns the resulting element */
+  def after(n: Int): T =
+    require(n >= 0, s"A number of steps must not be negative, but was $n.")
+    it.drop(n).next()

core/src/test/scala/deepwit/training/TapEverySuite.scala

@@ -22,15 +22,21 @@ class TapEverySuite extends AnyFunSpec with Matchers:
       Iterator.from(0).tapEvery(1)((_, id) => seen += id).take(3).toList
       seen.toList shouldBe List(1, 2)
 
-  describe("LazyList.tapEvery"):
+  describe("Iterator.after"):
 
-    it("fires at every n-th index but not at zero"):
-      val seen = ListBuffer.empty[(String, Int)]
-      LazyList.from(0).map(i => s"e$i").tapEvery(3)((t, id) => seen += ((t, id))).take(10).toList
-      seen.toList shouldBe List(("e3", 3), ("e6", 6), ("e9", 9))
+    it("counts from zero, so the first element is the state after no steps"):
+      Iterator.from(0).after(0) shouldBe 0
 
-    it("stays lazy until the elements are forced"):
-      val seen = ListBuffer.empty[Int]
-      val tapped = LazyList.from(0).tapEvery(1)((_, id) => seen += id)
-      seen.toList shouldBe empty
-      tapped.take(3).toList shouldBe List(0, 1, 2)
+    it("returns the element that many steps in"):
+      Iterator.from(0).after(3) shouldBe 3
+
+    it("advances the iterator past what it returns"):
+      val trajectory = Iterator.from(0)
+      trajectory.after(3) shouldBe 3
+      trajectory.next() shouldBe 4
+
+    it("throws when the iterator ends first"):
+      a[NoSuchElementException] should be thrownBy Iterator(0, 1).after(5)
+
+    it("rejects a negative number of steps"):
+      an[IllegalArgumentException] should be thrownBy Iterator.from(0).after(-1)

examples/src/main/scala/deepwit/examples/autoencoder/AutoEncoderTrain.scala

@@ -6,7 +6,7 @@ import dimwit.Conversions.given
 import deepwit.examples.dataset.MNISTLoader
 import MNISTLoader.TestSample
 
-import deepwit.training.{Monitor, tapEvery}
+import deepwit.training.{Monitor, after, tapEvery}
 import deepwit.checkpointing.TensorTreeCheckpointer
 import deepwit.loss.BinaryCrossEntropy
 import dimwit.optimizer.{Adam, AdamState}
@@ -83,7 +83,6 @@ def train(): Unit =
       case (state, step) =>
         checkpointer.save(state, step)
         println(s"Checkpoint saved at epoch $step")
-    .drop(numIterations)
-    .next()
+    .after(numIterations)
 
   println(s"Done. Wrote ${checkpointer.rootPath}.")

examples/src/main/scala/deepwit/examples/gpt/GPTTrain.scala

@@ -4,7 +4,7 @@ import deepwit.loss.CategoricalCrossEntropy
 
 import dimwit.*
 import dimwit.Conversions.given
-import deepwit.training.{Monitor, tapEvery}
+import deepwit.training.{Monitor, after, tapEvery}
 import deepwit.optimizer.*
 import dimwit.optimizer.{AdamW, Adam, AdamState}
 import dimwit.TreeOf.ops.*
@@ -186,5 +186,4 @@ import Config.*
         logger.save(state, step)
         println(s"Checkpoint saved")
         println("-" * 30)
-    .drop(1_000_000_000)
-    .next()
+    .after(1_000_000_000)

examples/src/main/scala/deepwit/examples/mnistClassification/MNistCNNTrain.scala

@@ -8,7 +8,7 @@ import deepwit.loss.CategoricalCrossEntropy
 
 import deepwit.examples.dataset.{MNISTLoader, MNISTBatchSample}
 import dimwit.optimizer.GradientDescentState
-import deepwit.training.{Monitor, tapEvery}
+import deepwit.training.{Monitor, after, tapEvery}
 import deepwit.checkpointing.TensorTreeCheckpointer
 
 case class TrainState(
@@ -81,7 +81,6 @@ def train(): Unit =
       case (state, step) =>
         checkpointer.save(state, step)
         println(s"Checkpoint saved at epoch $step")
-    .drop(numIterations)
-    .next()
+    .after(numIterations)
 
   println(s"Done. Wrote ${checkpointer.rootPath}.")

examples/src/main/scala/deepwit/examples/neuralImage/NeuralImageTrain.scala

@@ -4,7 +4,7 @@ import dimwit.*
 import dimwit.Conversions.given
 import dimwit.optimizer.{Adam, AdamState}
 
-import deepwit.training.{Monitor, tapEvery}
+import deepwit.training.{Monitor, after, tapEvery}
 import deepwit.checkpointing.TensorTreeCheckpointer
 import deepwit.loss.SquaredError
 
@@ -103,8 +103,7 @@ def train(): Unit =
   val finalState = trainTrajectory
     .tapEvery(100):
       case (state, step) => println(trainMonitor.report(step, state))
-    .drop(numIterations)
-    .next()
+    .after(numIterations)
 
   // -- Save final state --
 

examples/src/main/scala/deepwit/examples/regression/Regression.scala

@@ -9,6 +9,7 @@ import io.circe.Json
 import plotwit.*
 import plotwit.PlotTargets.desktopBrowser
 
+import deepwit.training.after
 import deepwit.activation.gelu
 import deepwit.base.{AffineFormLayer, AffineLayer}
 import deepwit.checkpointing.TensorTreeCheckpointer
@@ -104,8 +105,7 @@ def train(): Unit =
   // -- Run train trajectory --
 
   val finalState = trainTrajectory
-    .drop(numIterations)
-    .next()
+    .after(numIterations)
 
   // -- Save the fitted state --
 

examples/src/main/scala/deepwit/examples/thinning/MoonsMLPTrain.scala

@@ -8,7 +8,7 @@ import dimwit.Conversions.given
 import dimwit.optimizer.{Adam, AdamState}
 
 import deepwit.loss.CategoricalCrossEntropy
-import deepwit.training.{Monitor, tapEvery}
+import deepwit.training.{Monitor, after, tapEvery}
 import deepwit.checkpointing.TensorTreeCheckpointer
 
 case class TrainState(
@@ -90,8 +90,7 @@ def train(): Unit =
       case (state, step) =>
         checkpointer.save(state, step)
         println(s"Checkpoint saved at step $step")
-    .drop(numIterations)
-    .next()
+    .after(numIterations)
 
   println(f"Final cost: ${finalState.lastCost.item}%.6f")
   println(s"Done. Wrote ${checkpointer.rootPath}.")

examples/src/main/scala/deepwit/examples/variationalAutoencoder/VariationalAutoencoderTrain.scala

@@ -8,7 +8,7 @@ import deepwit.examples.dataset.MNISTLoader
 
 import deepwit.checkpointing.TensorTreeCheckpointer
 import deepwit.loss.BinaryCrossEntropy
-import deepwit.training.{Monitor, tapEvery}
+import deepwit.training.{Monitor, after, tapEvery}
 
 case class TrainState(
     params: VariationalAutoencoder.Params,
@@ -101,8 +101,7 @@ def train(): Unit =
       case (state, step) =>
         checkpointer.save(state, step)
         println(s"Checkpoint saved at step $step")
-    .drop(numIterations)
-    .next()
+    .after(numIterations)
 
   println(f"Final cost: ${finalState.lastCost.item}%.6f")
   println(s"Done. Wrote ${checkpointer.rootPath}.")

mdocs/README.md

@@ -7,6 +7,7 @@ import deepwit.activation.gelu
 import deepwit.base.{AffineFormLayer, AffineLayer}
 import deepwit.checkpointing.TensorTreeCheckpointer
 import deepwit.loss.SquaredError
+import deepwit.training.after
 
 dimwit.initialize()
 
@@ -142,7 +143,7 @@ Training the model reduces to a termination condition on this iterator; here aft
 A model checkpointer serializes the final train state object.
 
 ```scala mdoc:compile-only
-val finalState = trainTrajectory.drop(numIterations).next()
+val finalState = trainTrajectory.after(numIterations)
 
 TensorTreeCheckpointer.newIn(checkpointRoot).save(finalState, numIterations)
 ```
@@ -179,7 +180,7 @@ The user code composes these core modules into custom architectures given the us
 | `deepwit.init` | Xavier/Glorot normal and uniform, for matrices and vectors |
 | `deepwit.regularization` | `Perturbation` — thinning (dropout) as a mutation of the weights that *read* a feature |
 | `deepwit.optimizer` | `LearningRateSchedule` (constant, linear warmup, cosine decay), `LearningRateScheduler`, `clipGlobalNorm` |
-| `deepwit.training` | `Monitor` (step, loss, throughput, learning rate), `tapEvery` |
+| `deepwit.training` | `Monitor` (step, loss, throughput, learning rate), `tapEvery`, `after` |
 | `deepwit.checkpointing` | `TensorTreeCheckpointer` — save and load any `TensorTree` by iteration |
 
 ## Relationship to DimWit