ML-400 Converted back Network interface

[apphp]

No description provided.

[apphp]

All tests were passed

[Copilot]

Pull request overview

This PR reintroduces a neural-network “Network” interface and updates instantiation sites/tests to construct FeedForward networks instead of the removed Network concrete class, alongside reorganizing the NDArray-based network code under NeuralNet\Networks\Base\....

Changes:

• Replace new Network(...) with new FeedForward(...) across learners and tests.
• Introduce src/NeuralNet/Networks/Base/Contracts/Network.php and refactor the NDArray-based FeedForward network to implement it; remove the old src/NeuralNet/Networks/Network.php.
• Update snapshot/network tests and adjust one Swish layer expected value; update changelog entries.

Reviewed changes

Copilot reviewed 18 out of 18 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tests/NeuralNet/Snapshots/SnapshotTest.php	Switch snapshot test to NDArray “Base” FeedForward + new Network contract.
tests/NeuralNet/SnapshotTest.php	Switch legacy snapshot test to instantiate FeedForward instead of Network.
tests/NeuralNet/Networks/NetworkTest.php	Update NDArray network test imports and construction to use Base FeedForward.
tests/NeuralNet/NetworkTest.php	Switch legacy network test construction to FeedForward.
tests/NeuralNet/Layers/Swish/SwishTest.php	Update expected numeric output to match current computation.
tests/NeuralNet/FeedForwards/FeedForwardTest.php	Point feedforward tests at Base FeedForward and new contracts.
src/Regressors/MLPRegressor.php	Use FeedForward for the underlying network construction.
src/Regressors/Adaline.php	Use FeedForward for the underlying network construction.
src/NeuralNet/Snapshots/Snapshot.php	Update snapshot to accept Base Network contract.
src/NeuralNet/Networks/Network.php	Remove the old NDArray Network concrete class.
src/NeuralNet/Networks/Base/FeedForward/FeedForward.php	Move/rename Base FeedForward, implement Base Network contract, add sample-normalization branch.
src/NeuralNet/Networks/Base/Contracts/Network.php	Add Base Network interface (input/hidden/output/layers).
src/NeuralNet/Network.php	Convert legacy Network to an interface (currently only layers()).
src/NeuralNet/FeedForward.php	Update legacy FeedForward to implement the legacy Network interface.
src/Classifiers/SoftmaxClassifier.php	Construct FeedForward network instead of Network.
src/Classifiers/MultilayerPerceptron.php	Construct FeedForward network instead of Network.
src/Classifiers/LogisticRegression.php	Construct FeedForward network instead of Network.
CHANGELOG.md	Note NDArray conversion + “converted back Network interface” entry.

Comments suppressed due to low confidence (2)

src/NeuralNet/Networks/Base/FeedForward/FeedForward.php:214

• The empty-dataset fast path (if ($dataset->empty()) return ...) and the 1D-shape reshape are currently guarded by $this->normalizeSamples, but that flag is never enabled. If these safeguards are needed to avoid NumPower edge cases, they should be applied unconditionally (or the flag needs to be configurable) so inference on empty/non-packed samples is actually protected.
  src/NeuralNet/Networks/Base/FeedForward/FeedForward.php:101
• $this->normalizeSamples is initialized to false and the property is private, but there is no constructor parameter or method to ever set it to true (confirmed by repo-wide search). This makes the normalization/reshape logic in infer() and the normalizeSamples() helper effectively dead code. Consider either removing the flag and always normalizing (if required for NumPower), or exposing a supported way to enable it (constructor arg or public setter).

---

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