foxhunt/crates/ml/tests/kan_integration.rs

//! KAN (Kolmogorov-Arnold Network) Integration Tests
//!
//! Validates the KAN trainable adapter end-to-end:
//! construction, forward pass, training loop, checkpoint save/load.

#![allow(unused_crate_dependencies)]

use candle_core::{Device, Tensor};
use ml::kan::config::KANConfig;
use ml::kan::trainable::KANTrainableAdapter;
use ml::training::unified_trainer::UnifiedTrainable;

fn small_kan_config() -> KANConfig {
    KANConfig {
        layer_widths: vec![10, 8, 4, 1],
        grid_size: 3,
        spline_order: 3,
        learning_rate: 1e-3,
        weight_decay: 1e-5,
        grad_clip: 1.0,
    }
}

#[test]
fn test_kan_construction() {
    let config = small_kan_config();
    let adapter = KANTrainableAdapter::new(config, &Device::Cpu);
    assert!(
        adapter.is_ok(),
        "KAN construction failed: {:?}",
        adapter.err()
    );
    let adapter = adapter.unwrap();
    assert_eq!(adapter.model_type(), "KAN");
    assert_eq!(adapter.get_step(), 0);
}

#[test]
fn test_kan_forward_pass() {
    let config = small_kan_config();
    let mut adapter = KANTrainableAdapter::new(config, &Device::Cpu).unwrap();

    // [batch=4, input_dim=10]
    let input = Tensor::randn(0f32, 1.0, &[4, 10], &Device::Cpu).unwrap();
    let output = adapter.forward(&input);
    assert!(output.is_ok(), "Forward failed: {:?}", output.err());

    let output = output.unwrap();
    assert_eq!(
        output.dims(),
        &[4, 1],
        "Expected [4, 1], got {:?}",
        output.dims()
    );
}

#[test]
fn test_kan_training_loop_loss_decreases() {
    let config = small_kan_config();
    let mut adapter = KANTrainableAdapter::new(config, &Device::Cpu).unwrap();

    // Synthetic regression: target = mean of inputs
    let batch_size = 16;
    let input_dim = 10;
    let input = Tensor::randn(0f32, 0.5, &[batch_size, input_dim], &Device::Cpu).unwrap();
    let target = input.mean_keepdim(1).unwrap();

    let mut first_loss = None;
    let mut last_loss = 0.0;

    for epoch in 0..50 {
        // Forward
        let predictions = adapter.forward(&input).unwrap();

        // Loss
        let loss = adapter.compute_loss(&predictions, &target).unwrap();
        let loss_val = loss.to_scalar::<f32>().unwrap() as f64;

        if first_loss.is_none() {
            first_loss = Some(loss_val);
        }
        last_loss = loss_val;

        // Backward
        let _grad_norm = adapter.backward(&loss).unwrap();

        // Optimizer step
        adapter.optimizer_step().unwrap();
        adapter.zero_grad().unwrap();

        if epoch % 10 == 0 {
            println!("KAN epoch {}: loss = {:.6}", epoch, loss_val);
        }
    }

    let first = first_loss.unwrap();
    println!(
        "KAN training: first_loss={:.6}, last_loss={:.6}, reduction={:.1}%",
        first,
        last_loss,
        (1.0 - last_loss / first) * 100.0
    );

    assert!(
        last_loss < first,
        "Loss should decrease: first={}, last={}",
        first,
        last_loss
    );
}

#[test]
fn test_kan_checkpoint_roundtrip() {
    let config = small_kan_config();
    let mut adapter = KANTrainableAdapter::new(config.clone(), &Device::Cpu).unwrap();

    // Do a few training steps to change weights
    let input = Tensor::randn(0f32, 1.0, &[4, 10], &Device::Cpu).unwrap();
    let target = Tensor::randn(0f32, 0.1, &[4, 1], &Device::Cpu).unwrap();
    for _ in 0..5 {
        let pred = adapter.forward(&input).unwrap();
        let loss = adapter.compute_loss(&pred, &target).unwrap();
        adapter.backward(&loss).unwrap();
        adapter.optimizer_step().unwrap();
    }

    // Save checkpoint
    let tmp_dir = std::env::temp_dir().join("kan_test_checkpoint");
    std::fs::create_dir_all(&tmp_dir).unwrap();
    let checkpoint_path = tmp_dir.join("kan_ckpt");
    let save_result = adapter.save_checkpoint(checkpoint_path.to_str().unwrap());
    assert!(save_result.is_ok(), "Save failed: {:?}", save_result.err());

    // Load into fresh adapter
    let mut adapter2 = KANTrainableAdapter::new(config, &Device::Cpu).unwrap();
    let load_result = adapter2.load_checkpoint(checkpoint_path.to_str().unwrap());
    assert!(load_result.is_ok(), "Load failed: {:?}", load_result.err());

    // Verify same predictions
    let pred1 = adapter.forward(&input).unwrap();
    let pred2 = adapter2.forward(&input).unwrap();

    let diff = (pred1 - pred2)
        .unwrap()
        .abs()
        .unwrap()
        .sum_all()
        .unwrap()
        .to_scalar::<f32>()
        .unwrap();
    assert!(
        diff < 1e-5,
        "Checkpoint roundtrip predictions differ by {}",
        diff
    );

    // Cleanup
    let _ = std::fs::remove_dir_all(&tmp_dir);
}

#[test]
fn test_kan_metrics_collection() {
    let config = small_kan_config();
    let mut adapter = KANTrainableAdapter::new(config, &Device::Cpu).unwrap();

    let input = Tensor::randn(0f32, 1.0, &[4, 10], &Device::Cpu).unwrap();
    let target = Tensor::randn(0f32, 0.1, &[4, 1], &Device::Cpu).unwrap();

    let pred = adapter.forward(&input).unwrap();
    let loss = adapter.compute_loss(&pred, &target).unwrap();
    adapter.backward(&loss).unwrap();
    adapter.optimizer_step().unwrap();

    let metrics = adapter.collect_metrics();
    assert!(metrics.learning_rate > 0.0);
    assert!(metrics.custom_metrics.contains_key("training_steps"));
    assert!(metrics.custom_metrics.contains_key("grid_size"));
    assert!(metrics.custom_metrics.contains_key("spline_order"));
}