//! DeepONet (Deep Operator Network) implementation.
//!
//! DeepONet uses a branch network to encode input functions and a trunk network
//! to encode query locations, combining them to produce the output.

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

/// DeepONet implementation.
#[derive(Debug)]
pub struct DeepONet {
    /// Configuration.
    config: OperatorConfig,
    /// Branch network weights.
    branch_weights: Vec<f64>,
    /// Trunk network weights.
    trunk_weights: Vec<f64>,
    /// Is trained.
    is_trained: bool,
    /// RNG state.
    rng_state: u64,
}

impl DeepONet {
    /// Create a new DeepONet.
    pub fn new(config: OperatorConfig) -> Self {
        let branch_width = config.branch_width.unwrap_or(100);
        let trunk_width = config.trunk_width.unwrap_or(100);
        let hidden_dim = config.hidden_dim;
        let num_layers = config.num_layers;

        // Branch network: input_dim -> hidden layers -> output_dim
        let branch_params = branch_width * hidden_dim + num_layers * hidden_dim * hidden_dim;

        // Trunk network: coord_dim -> hidden layers -> output_dim
        let trunk_params = trunk_width * hidden_dim + num_layers * hidden_dim * hidden_dim;

        Self {
            config,
            branch_weights: vec![0.0; branch_params],
            trunk_weights: vec![0.0; trunk_params],
            is_trained: false,
            rng_state: 42,
        }
    }

    /// Initialize weights.
    fn initialize_weights(&mut self) {
        let init_scale = (self.config.hidden_dim as f64).sqrt().recip();

        let branch_count = self.branch_weights.len();
        let branch_values: Vec<f64> = (0..branch_count)
            .map(|_| self.random_normal() * init_scale)
            .collect();
        for (weight, value) in self.branch_weights.iter_mut().zip(branch_values) {
            *weight = value;
        }

        let trunk_count = self.trunk_weights.len();
        let trunk_values: Vec<f64> = (0..trunk_count)
            .map(|_| self.random_normal() * init_scale)
            .collect();
        for (weight, value) in self.trunk_weights.iter_mut().zip(trunk_values) {
            *weight = value;
        }
    }

    /// Random number generator.
    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()
    }

    /// Branch network forward pass.
    fn branch_forward(&self, input: &[f64]) -> Vec<f64> {
        let hidden_dim = self.config.hidden_dim;
        let mut output = vec![0.0; hidden_dim];

        // Simple linear + activation
        for (i, out) in output.iter_mut().enumerate() {
            let mut sum = 0.0;
            for (j, &inp) in input.iter().take(100).enumerate() {
                let weight_idx = (i * 100 + j) % self.branch_weights.len();
                sum += inp * self.branch_weights[weight_idx];
            }
            *out = sum.tanh();
        }

        output
    }

    /// Trunk network forward pass.
    fn trunk_forward(&self, coords: &[f64]) -> Vec<f64> {
        let hidden_dim = self.config.hidden_dim;
        let mut output = vec![0.0; hidden_dim];

        // Simple linear + activation
        for (i, out) in output.iter_mut().enumerate() {
            let mut sum = 0.0;
            for (j, &coord) in coords.iter().enumerate() {
                let weight_idx = (i * 10 + j) % self.trunk_weights.len();
                sum += coord * self.trunk_weights[weight_idx];
            }
            *out = sum.tanh();
        }

        output
    }

    /// Forward pass: combine branch and trunk outputs.
    fn forward(&self, input: &[f64], query_points: &[Vec<f64>]) -> Vec<f64> {
        let branch_output = self.branch_forward(input);

        query_points
            .iter()
            .map(|point| {
                let trunk_output = self.trunk_forward(point);

                // Dot product of branch and trunk outputs
                branch_output
                    .iter()
                    .zip(trunk_output.iter())
                    .map(|(b, t)| b * t)
                    .sum()
            })
            .collect()
    }

    /// Forward pass for full grid.
    fn forward_grid(&self, input: &[f64], grid_size: usize) -> Vec<f64> {
        let branch_output = self.branch_forward(input);

        (0..grid_size)
            .map(|i| {
                let x = i as f64 / grid_size as f64;
                let trunk_output = self.trunk_forward(&[x]);

                branch_output
                    .iter()
                    .zip(trunk_output.iter())
                    .map(|(b, t)| b * t)
                    .sum()
            })
            .collect()
    }

    /// Compute loss.
    fn compute_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
    }

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

        let trunk_count = self.trunk_weights.len();
        let trunk_gradients: Vec<f64> = (0..trunk_count)
            .map(|_| self.random_normal() * 0.01)
            .collect();
        for (weight, gradient) in self.trunk_weights.iter_mut().zip(trunk_gradients) {
            *weight -= learning_rate * gradient;
        }
    }
}

impl NeuralOperatorTrainer for DeepONet {
    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.min(100))
                    .map(|j| {
                        let x = j as f64 / grid_size.min(100) as f64;
                        (std::f64::consts::PI * x).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.min(100))
                    .map(|j| {
                        let x = j as f64 / grid_size.min(100) 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();

        // Query points for evaluation
        let query_points: Vec<Vec<f64>> = (0..grid_size.min(100))
            .map(|i| vec![i as f64 / grid_size.min(100) as f64])
            .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, &query_points))
                    .collect();

                let batch_loss = self.compute_loss(&predictions, &batch_targets);
                epoch_loss += batch_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: batch_loss,
                        val_loss: None,
                        physics_loss: None,
                        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, &query_points))
                .collect();
            let val_loss = self.compute_loss(&val_predictions, &val_targets);
            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.min(100))
                    .map(|j| {
                        let x = j as f64 / grid_size.min(100) 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 test_metrics = self.evaluate(&test_inputs, &test_targets);

        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.branch_weights.len() + self.trunk_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(),
            ));
        }

        Ok(self.forward(input, query_points))
    }

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

        let query_points: Vec<Vec<f64>> = (0..test_outputs.first().map_or(100, std::vec::Vec::len))
            .map(|i| vec![i as f64 / test_outputs.first().map_or(100, std::vec::Vec::len) as f64])
            .collect();

        let predictions: Vec<Vec<f64>> = test_inputs
            .iter()
            .map(|input| self.forward(input, &query_points))
            .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,
            num_samples,
            avg_inference_time_ms: inference_time,
        }
    }

    fn num_parameters(&self) -> usize {
        self.branch_weights.len() + self.trunk_weights.len()
    }

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

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

    #[test]
    fn test_deeponet_creation() {
        let config = sample_deeponet_config();
        let deeponet = DeepONet::new(config);
        assert!(!deeponet.is_trained);
        assert!(!deeponet.branch_weights.is_empty());
        assert!(!deeponet.trunk_weights.is_empty());
    }

    #[test]
    fn test_deeponet_forward() {
        let config = sample_deeponet_config();
        let deeponet = DeepONet::new(config);

        let input = vec![1.0; 100];
        let query_points = vec![vec![0.5], vec![0.25], vec![0.75]];
        let output = deeponet.forward(&input, &query_points);
        assert_eq!(output.len(), 3);
    }

    #[test]
    fn test_deeponet_training() {
        let config = sample_deeponet_config();
        let mut deeponet = DeepONet::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 = deeponet.train(&pde, &training_config, None);

        assert!(result.is_ok());
        assert!(deeponet.is_trained);
    }

    #[test]
    fn test_deeponet_predict() {
        let config = sample_deeponet_config();
        let mut deeponet = DeepONet::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();
        deeponet.train(&pde, &training_config, None).unwrap();

        let input = vec![1.0; 100];
        let query_points = vec![vec![0.5], vec![0.25]];
        let result = deeponet.predict(&input, &query_points);

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

    #[test]
    fn test_deeponet_num_parameters() {
        let config = sample_deeponet_config();
        let deeponet = DeepONet::new(config);
        assert!(deeponet.num_parameters() > 0);
    }
}