fix: add SetModel method to IOptimizer to fix NullReferenceException in InitializeRandomSolution

[ooples]

Summary

• Fixes Issue fix(us-ci-004): refine TransferLearning cross-domain exception messages #225: [Fix] Kill Time Series Stubs & Implement Real Training
• Resolves NullReferenceException at OptimizerBase.InitializeRandomSolution line 1239 where _model.ParameterCount was accessed but _model was null

Root Cause

The optimizer's _model field was never being set after configuration. The model was only passed during optimizer construction, but many constructors passed null for the model parameter. When Optimize() was called, InitializeRandomSolution() would try to access _model.ParameterCount and throw a NullReferenceException.

Changes

1. IOptimizer interface - Added SetModel(IFullModel<T, TInput, TOutput> model) method
2. OptimizerBase - Implemented SetModel to set the _model field
3. ShardedOptimizerBase - Implemented SetModel to delegate to wrapped optimizer
4. AiModelBuilder - Added SetModel() calls before Optimize() in all 4 training paths:
   • Regular training path
   • Hyperparameter optimization trial
   • Knowledge distillation fallback
   • Deep ensemble member optimizer

Test plan

• Build passes for net10.0 and net471
• All existing optimizer and time series tests pass (811 tests)
• TDAB integration tests pass after fix

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features

  • Exposed a SetModel API across optimizers to allow explicit model assignment before training.

• Bug Fixes

  • Ensured optimizers are bound to models before training, fallback, and ensemble paths to prevent training/runtime issues.
  • Fixed ARIMA prediction handling for absent AR or MA terms to avoid out-of-range writes.

• Chores

  • Added review/architecture configuration and updated example kernel parameter naming.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
aidotnet-playground-api	Ready	Preview, Comment	Feb 6, 2026 6:07pm

[coderabbitai]

Walkthrough

Adds IOptimizer.SetModel and implementations in OptimizerBase and ShardedOptimizerBase; updates AiModelBuilder to call SetModel after Reset and before Optimize across training, KD-fallback, ensemble, and deep-ensemble paths; adds no-op SetModel stubs in tests; fixes ARIMA Predict index/length handling; adds .coderabbit.yaml and a small playground rename.

Changes

Cohort / File(s)	Summary
Optimizer Interface & Base src/Interfaces/IOptimizer.cs, src/Optimizers/OptimizerBase.cs	Adds void SetModel(IFullModel<..., ..., ...> model) to the optimizer interface and a concrete implementation in OptimizerBase that stores the model, exposes RequireModel(), and calls OnModelChanged(...).
Distributed Optimizer Delegation src/DistributedTraining/ShardedOptimizerBase.cs	Adds SetModel(...) that delegates to WrappedOptimizer.SetModel(...) so sharded/wrapped optimizers propagate model binding.
AiModelBuilder call sites src/AiModelBuilder.cs	Calls to SetModel(model) inserted after Reset() and/or before Optimize() across supervised/trial, regular training, KD-fallback, ensemble, and deep-ensemble member construction flows to ensure optimizer→model binding.
Test optimizers (no-op SetModel) tests/** tests/AiDotNet.Serving.Tests/SinglePassTestOptimizer.cs, tests/AiDotNet.Tests/.../Federated*.cs, tests/.../TestOptimizers.cs, tests/.../MetaLearningTestInfrastructure.cs, tests/.../CheckpointManagerTests.cs	Adds no-op SetModel(...) implementations to multiple test optimizer classes and mocks to satisfy the new interface requirement; behavior unchanged.
Time Series Predict Fixes src/TimeSeries/ARIMAModel.cs	Fixes AR/MA vector lengths to use coefficient lengths and guards writes to last-observed/last-error arrays when P or Q is zero.
Playground minor rename src/AiDotNet.Playground/Services/ExampleService.cs	Renames Matérn kernel named parameter from length to lengthScale at call site.
Repo review config .coderabbit.yaml	Adds review/architecture guidance configuration (non-runtime).

Sequence Diagram(s)

sequenceDiagram
  participant Builder as AiModelBuilder
  participant Optim as IOptimizer
  participant Model as IFullModel
  participant Wrapped as Sharded/WrappedOptimizer

  Builder->>Optim: Reset()
  Note right of Optim: internal state cleared
  Builder->>Optim: SetModel(Model)
  Note right of Optim: optimizer bound to model
  Builder->>Optim: Optimize(...)
  Optim->>Model: Read/Modify parameters

  %% Sharded/wrapped delegation
  Builder->>Wrapped: SetModel(Model)
  Wrapped->>Optim: WrappedOptimizer.SetModel(Model)

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• fix: eliminate 372 build errors - production readiness improvements #220 — Related optimizer↔model binding refactor and constructor/binding changes.
• fix(us-bf-057): improve step/calculateupdate exception messages in optimizerbase #236 — Overlaps with changes to OptimizerBase internals and lifecycle semantics.
• test: add MetaLearning integration tests #678 — Related updates to test optimizers and meta-learning test infrastructure.

Poem

I bind the optimizers, one by one,
After Reset the model dance begun.
Ensembles and students line the row,
Parameters found — the updates flow.
Tests nod gently: "No-op, we know."

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 35.14% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The PR title accurately and specifically describes the primary change: adding a SetModel method to IOptimizer to resolve a NullReferenceException issue.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch fix/optimizer-model-association

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[Copilot]

Pull request overview

This PR addresses a NullReferenceException in OptimizerBase.InitializeRandomSolution by ensuring optimizers have their target model assigned before optimization runs, primarily through a new SetModel(...) API on IOptimizer and call sites in AiModelBuilder.

Changes:

• Added SetModel(IFullModel<...> model) to IOptimizer and implemented it in OptimizerBase.
• Implemented SetModel in ShardedOptimizerBase by delegating to the wrapped optimizer.
• Updated AiModelBuilder training flows to call SetModel(...) before Optimize().

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.

File	Description
src/Optimizers/OptimizerBase.cs	Adds SetModel(...) implementation to assign the internal _model.
src/Interfaces/IOptimizer.cs	Extends the optimizer public interface with a new SetModel(...) method.
src/DistributedTraining/ShardedOptimizerBase.cs	Delegates SetModel(...) to the wrapped optimizer for sharded/distributed cases.
src/AiModelBuilder.cs	Ensures the model is set on optimizers before optimization across several training paths.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[coderabbitai]

Actionable comments posted: 1

🤖 Fix all issues with AI agents

In `@src/AiModelBuilder.cs`:
- Around line 5268-5270: The optimizer's internal state must be cleared before
running the fallback training to avoid carrying over partial KD updates: before
calling optimizer.SetModel(studentModel) / optimizer.Optimize(...) reset the
optimizer state (e.g., call optimizer.ResetState() or reinitialize a fresh
optimizer instance and assign it to the same variable) so Optimize runs from a
clean state; if ResetState() does not exist, construct a new optimizer of the
same type (or use a provided factory/CloneWithoutState) and then call
SetModel(studentModel) and Optimize with OptimizerHelper<T, TInput,
TOutput>.CreateOptimizationInputData(...).

🧹 Nitpick comments (1)

src/Optimizers/OptimizerBase.cs (1)

1220-1245: Consider adding a guard clause for _model at the method entry.

InitializeRandomSolution(TInput) accesses _model! with the null-forgiving operator at lines 1245, 1325, 1337, and 1354. If a derived class or a new code path calls this before SetModel, the original NRE will resurface with no clear message. A single guard at the top converts a cryptic NRE into an actionable InvalidOperationException.

🛡️ Proposed guard clause

 protected virtual IFullModel<T, TInput, TOutput> InitializeRandomSolution(TInput trainingData)
 {
     if (trainingData == null) throw new ArgumentNullException(nameof(trainingData));
+    if (_model == null)
+        throw new InvalidOperationException(
+            "Model has not been set. Call SetModel() before optimization.");

     // Compute lower and upper bounds from the training data

[Copilot]

Pull request overview

Copilot reviewed 11 out of 11 changed files in this pull request and generated 1 comment.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 13 out of 13 changed files in this pull request and generated 1 comment.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[coderabbitai]

Actionable comments posted: 0

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

src/Optimizers/OptimizerBase.cs (1)

608-615: ⚠️ Potential issue | 🟡 Minor

Model!.DeepCopy() still uses the null-forgiving operator — missed RequireModel() replacement.

This call site can produce the same opaque NullReferenceException the PR aims to eliminate. Updating it to RequireModel().DeepCopy() would be consistent with the changes elsewhere in the file.

Proposed fix

-        var solution = Model!.DeepCopy();
+        var solution = RequireModel().DeepCopy();

🧹 Nitpick comments (2)

src/Optimizers/OptimizerBase.cs (2)

151-161: SetModel expands the public API surface on optimizer types. As per the coding guidelines, users should interact through AiModelBuilder, and optimizer internals should prefer internal access. Since SetModel is driven by the IOptimizer interface contract, consider using explicit interface implementation to keep the method off the optimizer's public type surface:

void IOptimizer<T, TInput, TOutput>.SetModel(IFullModel<T, TInput, TOutput> model) { ... }

This would still allow AiModelBuilder to call it through the IOptimizer reference while not adding a new public method visible to users who might hold a concrete optimizer type. This is a minor concern given that optimizers are typically accessed via the interface already. As per coding guidelines: "Prefer internal over public for all classes, methods, and properties unless they are part of the facade API."

---

136-175: Well-designed lifecycle — consider sealing SetModel to protect the invariant.

RequireModel() and the OnModelChanged hook are clean additions. One architectural concern: SetModel is public virtual, which means a derived optimizer can override it and forget to call base.SetModel(...), leaving _model silently null. Since OnModelChanged already provides the extensibility hook, removing virtual from SetModel (making it sealed for override) would close this gap while keeping the same extension surface.

Proposed diff

-    public virtual void SetModel(IFullModel<T, TInput, TOutput> model)
+    public void SetModel(IFullModel<T, TInput, TOutput> model)

If IOptimizer.SetModel must remain overridable for distributed wrappers like ShardedOptimizerBase, consider the Template Method pattern: keep the assignment + hook in a non-virtual public method, and only expose OnModelChanged for override.

#!/bin/bash
# Check if any derived class overrides SetModel (would break if we remove virtual)
rg -n --type=cs 'override\s+.*\bSetModel\b' -C 2

[sonarqubecloud]

Quality Gate failed

Failed conditions
0.0% Coverage on New Code (required ≥ 80%)
B Reliability Rating on New Code (required ≥ A)

See analysis details on SonarQube Cloud

Catch issues before they fail your Quality Gate with our IDE extension SonarQube for IDE