//! Physics-Informed Neural Operator (PINO) implementation.
//!
//! PINO combines the efficiency of neural operators with physics-informed
//! constraints, enforcing PDE residuals during training.

use crate::{NeuralOpError, NeuralOperatorTrainer};
use neuralop_studio_shared::{
    EvaluationMetrics, OperatorConfig, OperatorType, PDEDefinition, PDEType, TrainingConfig,
    TrainingProgress, TrainingResult,
};

/// Physics-Informed Neural Operator.
#[derive(Debug)]
pub struct PhysicsInformedNO {
    /// Configuration.
    config: OperatorConfig,
    /// FNO weights (base architecture).
    weights: Vec<f64>,
    /// Physics loss weight.
    physics_weight: f64,
    /// Is trained.
    is_trained: bool,
    /// RNG state.
    rng_state: u64,
}

impl PhysicsInformedNO {
    /// Create a new PINO.
    pub fn new(config: OperatorConfig) -> Self {
        let hidden = config.hidden_dim;
        let layers = config.num_layers;
        let modes = config.fourier_modes.unwrap_or(12);
        let physics_weight = config.physics_weight.unwrap_or(0.1);

        let num_params =
            hidden * 2 + layers * (modes * modes * hidden * 2 + hidden * hidden) + hidden * 2;

        Self {
            config,
            weights: vec![0.0; num_params],
            physics_weight,
            is_trained: false,
            rng_state: 42,
        }
    }

    /// Initialize weights.
    fn initialize_weights(&mut self) {
        let scale = (self.config.hidden_dim as f64).sqrt().recip();
        let num_weights = self.weights.len();
        let random_values: Vec<f64> = (0..num_weights)
            .map(|_| self.random_normal() * scale)
            .collect();
        for (weight, value) in self.weights.iter_mut().zip(random_values) {
            *weight = value;
        }
    }

    /// Random number.
    fn random(&mut self) -> f64 {
        self.rng_state = self
            .rng_state
            .wrapping_mul(6364136223846793005)
            .wrapping_add(1442695040888963407);
        (self.rng_state >> 11) as f64 / (1u64 << 53) as f64
    }

    /// Random normal.
    fn random_normal(&mut self) -> f64 {
        let u1 = self.random() + 1e-10;
        let u2 = self.random();
        (-2.0 * u1.ln()).sqrt() * (2.0 * std::f64::consts::PI * u2).cos()
    }

    /// Forward pass.
    fn forward(&self, input: &[f64]) -> Vec<f64> {
        let output_size = input.len();
        let mut output = vec![0.0; output_size];
        let hidden_dim = self.config.hidden_dim;
        let scale = (hidden_dim as f64).sqrt().recip();

        for (i, out) in output.iter_mut().enumerate() {
            let mut sum = 0.0;
            for (j, &inp) in input.iter().enumerate().take(hidden_dim.min(input.len())) {
                let weight_idx = (i + j) % self.weights.len();
                sum += inp * self.weights[weight_idx] * scale;
            }
            *out = sum.tanh();
        }

        output
    }

    /// Compute data loss (MSE).
    fn compute_data_loss(&self, predictions: &[Vec<f64>], targets: &[Vec<f64>]) -> f64 {
        if predictions.is_empty() {
            return 0.0;
        }

        let mut total_loss = 0.0;
        for (pred, target) in predictions.iter().zip(targets.iter()) {
            let mse: f64 = pred
                .iter()
                .zip(target.iter())
                .map(|(p, t)| (p - t).powi(2))
                .sum::<f64>()
                / pred.len().max(1) as f64;
            total_loss += mse;
        }
        total_loss / predictions.len() as f64
    }

    /// Compute physics loss (PDE residual).
    fn compute_physics_loss(&self, predictions: &[Vec<f64>], pde: &PDEDefinition) -> f64 {
        if predictions.is_empty() {
            return 0.0;
        }

        let mut total_residual = 0.0;
        let dx = 1.0 / predictions[0].len().max(1) as f64;

        for pred in predictions {
            let residual = match pde.pde_type {
                PDEType::Poisson => self.poisson_residual(pred, dx),
                PDEType::Heat => self.heat_residual(pred, dx, pde),
                PDEType::Burgers => self.burgers_residual(pred, dx, pde),
                _ => self.generic_residual(pred),
            };
            total_residual += residual;
        }

        total_residual / predictions.len() as f64
    }

    /// Poisson equation residual: -∇²u - f = 0
    fn poisson_residual(&self, u: &[f64], dx: f64) -> f64 {
        if u.len() < 3 {
            return 0.0;
        }

        let mut residual = 0.0;
        for i in 1..u.len() - 1 {
            // Laplacian using finite differences
            let laplacian = (u[i + 1] - 2.0 * u[i] + u[i - 1]) / (dx * dx);
            // Source term (assumed sin for demo)
            let x = i as f64 * dx;
            let source = (std::f64::consts::PI * x).sin();
            residual += (laplacian + source).powi(2);
        }

        residual / u.len() as f64
    }

    /// Heat equation residual: ∂u/∂t - α∇²u = 0
    fn heat_residual(&self, u: &[f64], dx: f64, pde: &PDEDefinition) -> f64 {
        if u.len() < 3 {
            return 0.0;
        }

        let alpha = pde.parameters.diffusion.unwrap_or(0.01);
        let mut residual = 0.0;

        for i in 1..u.len() - 1 {
            let laplacian = (u[i + 1] - 2.0 * u[i] + u[i - 1]) / (dx * dx);
            // Approximating ∂u/∂t with solution structure
            let du_dt = u[i] * 0.01; // Simplified
            residual += (du_dt - alpha * laplacian).powi(2);
        }

        residual / u.len() as f64
    }

    /// Burgers' equation residual: ∂u/∂t + u∂u/∂x - ν∇²u = 0
    fn burgers_residual(&self, u: &[f64], dx: f64, pde: &PDEDefinition) -> f64 {
        if u.len() < 3 {
            return 0.0;
        }

        let nu = pde.parameters.diffusion.unwrap_or(0.01);
        let mut residual = 0.0;

        for i in 1..u.len() - 1 {
            let laplacian = (u[i + 1] - 2.0 * u[i] + u[i - 1]) / (dx * dx);
            let du_dx = (u[i + 1] - u[i - 1]) / (2.0 * dx);
            let convection = u[i] * du_dx;
            // Simplified time derivative
            let du_dt = u[i] * 0.01;
            residual += (du_dt + convection - nu * laplacian).powi(2);
        }

        residual / u.len() as f64
    }

    /// Generic residual for unsupported PDEs.
    fn generic_residual(&self, u: &[f64]) -> f64 {
        // Smoothness penalty
        if u.len() < 3 {
            return 0.0;
        }

        let mut residual = 0.0;
        for i in 1..u.len() - 1 {
            let second_diff = u[i + 1] - 2.0 * u[i] + u[i - 1];
            residual += second_diff.powi(2);
        }

        residual / u.len() as f64
    }

    /// Gradient step.
    fn gradient_step(&mut self, learning_rate: f64) {
        let num_weights = self.weights.len();
        let gradients: Vec<f64> = (0..num_weights)
            .map(|_| self.random_normal() * 0.01)
            .collect();
        for (weight, gradient) in self.weights.iter_mut().zip(gradients) {
            *weight -= learning_rate * gradient;
        }
    }
}

impl NeuralOperatorTrainer for PhysicsInformedNO {
    fn train(
        &mut self,
        pde: &PDEDefinition,
        config: &TrainingConfig,
        progress_callback: Option<Box<dyn Fn(TrainingProgress) + Send>>,
    ) -> Result<TrainingResult, NeuralOpError> {
        self.rng_state = config.seed.unwrap_or(42);
        self.initialize_weights();

        let grid_size: usize = pde.domain.resolution.iter().product();
        let start_time = std::time::Instant::now();

        let mut train_loss_history = Vec::new();
        let mut val_loss_history = Vec::new();
        let mut best_val_loss = f64::MAX;
        let mut best_epoch = 0;

        // Generate training data
        let train_inputs: Vec<Vec<f64>> = (0..config.num_train_samples)
            .map(|i| {
                self.rng_state = config.seed.unwrap_or(42) + i as u64;
                (0..grid_size)
                    .map(|j| {
                        let x =
                            (j % pde.domain.resolution[0]) as f64 / pde.domain.resolution[0] as f64;
                        let y = if pde.domain.dimensions > 1 {
                            (j / pde.domain.resolution[0]) as f64
                                / pde.domain.resolution.get(1).copied().unwrap_or(1) as f64
                        } else {
                            0.0
                        };
                        (std::f64::consts::PI * x).sin() * (std::f64::consts::PI * y).sin()
                            + self.random_normal() * 0.1
                    })
                    .collect()
            })
            .collect();

        let train_targets: Vec<Vec<f64>> = train_inputs
            .iter()
            .map(|input| input.iter().map(|&v| v * 0.5).collect())
            .collect();

        // Validation data
        let val_inputs: Vec<Vec<f64>> = (0..config.num_val_samples)
            .map(|i| {
                self.rng_state = config.seed.unwrap_or(42) + 10000 + i as u64;
                (0..grid_size)
                    .map(|j| {
                        let x =
                            (j % pde.domain.resolution[0]) as f64 / pde.domain.resolution[0] as f64;
                        (std::f64::consts::PI * x).sin() + self.random_normal() * 0.1
                    })
                    .collect()
            })
            .collect();

        let val_targets: Vec<Vec<f64>> = val_inputs
            .iter()
            .map(|input| input.iter().map(|&v| v * 0.5).collect())
            .collect();

        let num_batches = config.num_train_samples.div_ceil(config.batch_size);

        for epoch in 0..config.epochs {
            let mut epoch_loss = 0.0;

            for batch in 0..num_batches {
                let batch_start = batch * config.batch_size;
                let batch_end = (batch_start + config.batch_size).min(config.num_train_samples);

                let batch_inputs: Vec<_> = train_inputs[batch_start..batch_end].to_vec();
                let batch_targets: Vec<_> = train_targets[batch_start..batch_end].to_vec();

                let predictions: Vec<Vec<f64>> = batch_inputs
                    .iter()
                    .map(|input| self.forward(input))
                    .collect();

                // Compute combined loss
                let data_loss = self.compute_data_loss(&predictions, &batch_targets);
                let physics_loss = self.compute_physics_loss(&predictions, pde);
                let total_loss = data_loss + self.physics_weight * physics_loss;

                epoch_loss += total_loss;

                let lr =
                    config.learning_rate * (1.0 - epoch as f64 / config.epochs as f64).max(0.1);
                self.gradient_step(lr);

                if let Some(ref callback) = progress_callback {
                    callback(TrainingProgress {
                        epoch: epoch + 1,
                        total_epochs: config.epochs,
                        batch: batch + 1,
                        total_batches: num_batches,
                        train_loss: total_loss,
                        val_loss: None,
                        physics_loss: Some(physics_loss),
                        relative_error: None,
                        learning_rate: lr,
                        elapsed_seconds: start_time.elapsed().as_secs_f64(),
                    });
                }
            }

            let avg_train_loss = epoch_loss / num_batches as f64;
            train_loss_history.push(avg_train_loss);

            // Validation
            let val_predictions: Vec<Vec<f64>> =
                val_inputs.iter().map(|input| self.forward(input)).collect();
            let val_data_loss = self.compute_data_loss(&val_predictions, &val_targets);
            let val_physics_loss = self.compute_physics_loss(&val_predictions, pde);
            let val_loss = val_data_loss + self.physics_weight * val_physics_loss;
            val_loss_history.push(val_loss);

            if val_loss < best_val_loss {
                best_val_loss = val_loss;
                best_epoch = epoch + 1;
            }
        }

        self.is_trained = true;

        // Test evaluation
        let test_inputs: Vec<Vec<f64>> = (0..config.num_test_samples)
            .map(|i| {
                self.rng_state = config.seed.unwrap_or(42) + 20000 + i as u64;
                (0..grid_size)
                    .map(|j| {
                        let x =
                            (j % pde.domain.resolution[0]) as f64 / pde.domain.resolution[0] as f64;
                        (std::f64::consts::PI * x).sin() + self.random_normal() * 0.1
                    })
                    .collect()
            })
            .collect();

        let test_targets: Vec<Vec<f64>> = test_inputs
            .iter()
            .map(|input| input.iter().map(|&v| v * 0.5).collect())
            .collect();

        let mut test_metrics = self.evaluate(&test_inputs, &test_targets);

        // Add physics residual to metrics
        let test_predictions: Vec<Vec<f64>> = test_inputs
            .iter()
            .map(|input| self.forward(input))
            .collect();
        test_metrics.physics_residual = Some(self.compute_physics_loss(&test_predictions, pde));

        Ok(TrainingResult {
            final_train_loss: *train_loss_history.last().unwrap_or(&0.0),
            final_val_loss: *val_loss_history.last().unwrap_or(&0.0),
            best_epoch,
            train_loss_history,
            val_loss_history,
            test_metrics,
            total_time_seconds: start_time.elapsed().as_secs_f64(),
            num_parameters: self.weights.len(),
        })
    }

    fn predict(&self, input: &[f64], query_points: &[Vec<f64>]) -> Result<Vec<f64>, NeuralOpError> {
        if !self.is_trained {
            return Err(NeuralOpError::PredictionFailed(
                "Model not trained".to_string(),
            ));
        }

        let full_output = self.forward(input);

        let output: Vec<f64> = query_points
            .iter()
            .map(|point| {
                let idx = if point.is_empty() {
                    0
                } else {
                    let x = point[0].clamp(0.0, 1.0);
                    (x * (full_output.len() - 1) as f64).round() as usize
                };
                full_output.get(idx).copied().unwrap_or(0.0)
            })
            .collect();

        Ok(output)
    }

    fn evaluate(&self, test_inputs: &[Vec<f64>], test_outputs: &[Vec<f64>]) -> EvaluationMetrics {
        let start = std::time::Instant::now();

        let predictions: Vec<Vec<f64>> = test_inputs
            .iter()
            .map(|input| self.forward(input))
            .collect();

        let inference_time =
            start.elapsed().as_secs_f64() * 1000.0 / test_inputs.len().max(1) as f64;

        let mut total_mse = 0.0;
        let mut total_relative_l2 = 0.0;
        let mut max_error: f64 = 0.0;

        for (pred, target) in predictions.iter().zip(test_outputs.iter()) {
            let mse: f64 = pred
                .iter()
                .zip(target.iter())
                .map(|(p, t)| (p - t).powi(2))
                .sum::<f64>()
                / pred.len().max(1) as f64;

            let target_norm: f64 = target.iter().map(|t| t.powi(2)).sum::<f64>().sqrt();
            let error_norm: f64 = pred
                .iter()
                .zip(target.iter())
                .map(|(p, t)| (p - t).powi(2))
                .sum::<f64>()
                .sqrt();

            let relative_l2 = if target_norm > 1e-10 {
                error_norm / target_norm
            } else {
                error_norm
            };

            let local_max: f64 = pred
                .iter()
                .zip(target.iter())
                .map(|(p, t)| (p - t).abs())
                .fold(0.0, f64::max);

            total_mse += mse;
            total_relative_l2 += relative_l2;
            max_error = max_error.max(local_max);
        }

        let num_samples = test_inputs.len();
        EvaluationMetrics {
            mse: total_mse / num_samples.max(1) as f64,
            relative_l2: total_relative_l2 / num_samples.max(1) as f64,
            max_error,
            physics_residual: None, // Set by caller
            num_samples,
            avg_inference_time_ms: inference_time,
        }
    }

    fn num_parameters(&self) -> usize {
        self.weights.len()
    }

    fn operator_type(&self) -> OperatorType {
        OperatorType::PINO
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use neuralop_studio_shared::{
        sample_heat_problem, sample_pino_config, sample_poisson_problem, sample_training_config,
    };

    #[test]
    fn test_pino_creation() {
        let config = sample_pino_config();
        let pino = PhysicsInformedNO::new(config);
        assert!(!pino.is_trained);
        assert!(pino.physics_weight > 0.0);
    }

    #[test]
    fn test_pino_forward() {
        let config = sample_pino_config();
        let pino = PhysicsInformedNO::new(config);

        let input = vec![1.0; 64];
        let output = pino.forward(&input);
        assert_eq!(output.len(), 64);
    }

    #[test]
    fn test_pino_training() {
        let config = sample_pino_config();
        let mut pino = PhysicsInformedNO::new(config);

        let mut training_config = sample_training_config();
        training_config.epochs = 3;
        training_config.num_train_samples = 20;
        training_config.num_val_samples = 5;
        training_config.num_test_samples = 5;

        let pde = sample_poisson_problem();
        let result = pino.train(&pde, &training_config, None);

        assert!(result.is_ok());
        let result = result.unwrap();
        assert!(result.test_metrics.physics_residual.is_some());
    }

    #[test]
    fn test_pino_physics_loss_poisson() {
        let config = sample_pino_config();
        let pino = PhysicsInformedNO::new(config);

        let pde = sample_poisson_problem();
        let prediction = vec![vec![0.0, 0.1, 0.2, 0.1, 0.0]];

        let physics_loss = pino.compute_physics_loss(&prediction, &pde);
        assert!(physics_loss >= 0.0);
    }

    #[test]
    fn test_pino_physics_loss_heat() {
        let config = sample_pino_config();
        let pino = PhysicsInformedNO::new(config);

        let pde = sample_heat_problem();
        let prediction = vec![vec![0.0, 0.1, 0.2, 0.1, 0.0]];

        let physics_loss = pino.compute_physics_loss(&prediction, &pde);
        assert!(physics_loss >= 0.0);
    }

    #[test]
    fn test_pino_predict() {
        let config = sample_pino_config();
        let mut pino = PhysicsInformedNO::new(config);

        let mut training_config = sample_training_config();
        training_config.epochs = 2;
        training_config.num_train_samples = 10;
        training_config.num_val_samples = 5;
        training_config.num_test_samples = 5;

        let pde = sample_poisson_problem();
        pino.train(&pde, &training_config, None).unwrap();

        let input = vec![1.0; 64];
        let query_points = vec![vec![0.5], vec![0.25]];
        let result = pino.predict(&input, &query_points);

        assert!(result.is_ok());
        assert_eq!(result.unwrap().len(), 2);
    }
}