Skip to content

refactor: Generate predictions in parallell #324

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Siel merged 1 commit into from
Sep 15, 2026
Merged

refactor: Generate predictions in parallell #324

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
124 changes: 69 additions & 55 deletions src/estimation/nonparametric/predictions.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@ use std::path::Path;

use anyhow::{bail, Result};
use pharmsol::{prelude::simulator::Prediction, Censor, Data, Predictions as PredTrait};
use rayon::prelude::*;
use serde::{Deserialize, Serialize};

use crate::{
Expand Down Expand Up @@ -115,84 +116,91 @@ impl NPPredictions {
}

pub fn calculate(
equation: &impl pharmsol::prelude::simulator::Equation,
equation: &(impl pharmsol::prelude::simulator::Equation + Sync),
data: &Data,
theta: &Theta,
w: &Weights,
posterior: &Posterior,
idelta: f64,
tad: f64,
) -> Result<Self> {
let mut container = NPPredictions::new();

let data = data.clone().expand(idelta, tad);
let subjects = data.subjects();

if subjects.len() != posterior.matrix().nrows() {
bail!("Number of subjects and number of posterior means do not match");
};

for (subject_index, subject) in subjects.iter().enumerate() {
let mut predictions: Vec<Vec<Prediction>> = Vec::new();
let support_points: Vec<Vec<f64>> = theta
.matrix()
.row_iter()
.map(|spp| spp.iter().cloned().collect())
.collect();

let per_subject: Vec<Vec<NPPredictionRow>> = subjects
.par_iter()
.enumerate()
.map(|(subject_index, subject)| -> Result<Vec<NPPredictionRow>> {
let predictions: Vec<Vec<Prediction>> = support_points
.par_iter()
.map(|spp| {
Ok(equation
.simulate_subject_dense(subject, spp, None)?
.0
.get_predictions())
})
.collect::<Result<Vec<_>>>()?;

let Some(first_spp_preds) = predictions.first() else {
return Ok(Vec::new());
};
let n_points = first_spp_preds.len();

let mut pop_mean: Vec<f64> = vec![0.0; n_points];
for (i, outer_pred) in predictions.iter().enumerate() {
for (j, pred) in outer_pred.iter().enumerate() {
pop_mean[j] += pred.prediction() * w[i];
}
}

for spp in theta.matrix().row_iter() {
let spp_values = spp.iter().cloned().collect::<Vec<f64>>();
let pred = equation
.simulate_subject_dense(subject, &spp_values, None)?
.0
.get_predictions();
predictions.push(pred);
}
let mut pop_median: Vec<f64> = Vec::with_capacity(n_points);
for j in 0..n_points {
let mut values: Vec<f64> = Vec::new();
let mut weights: Vec<f64> = Vec::new();

if predictions.is_empty() {
continue;
}
for (i, outer_pred) in predictions.iter().enumerate() {
values.push(outer_pred[j].prediction());
weights.push(w[i]);
}

let mut pop_mean: Vec<f64> = vec![0.0; predictions.first().unwrap().len()];
for (i, outer_pred) in predictions.iter().enumerate() {
for (j, pred) in outer_pred.iter().enumerate() {
pop_mean[j] += pred.prediction() * w[i];
pop_median.push(weighted_median(&values, &weights));
}
}

let mut pop_median: Vec<f64> = Vec::new();
for j in 0..predictions.first().unwrap().len() {
let mut values: Vec<f64> = Vec::new();
let mut weights: Vec<f64> = Vec::new();

let mut posterior_mean: Vec<f64> = vec![0.0; n_points];
for (i, outer_pred) in predictions.iter().enumerate() {
values.push(outer_pred[j].prediction());
weights.push(w[i]);
for (j, pred) in outer_pred.iter().enumerate() {
posterior_mean[j] +=
pred.prediction() * posterior.matrix()[(subject_index, i)];
}
}

let median_val = weighted_median(&values, &weights);
pop_median.push(median_val);
}

let mut posterior_mean: Vec<f64> = vec![0.0; predictions.first().unwrap().len()];
for (i, outer_pred) in predictions.iter().enumerate() {
for (j, pred) in outer_pred.iter().enumerate() {
posterior_mean[j] += pred.prediction() * posterior.matrix()[(subject_index, i)];
}
}
let mut posterior_median: Vec<f64> = Vec::with_capacity(n_points);
for j in 0..n_points {
let mut values: Vec<f64> = Vec::new();
let mut weights: Vec<f64> = Vec::new();

let mut posterior_median: Vec<f64> = Vec::new();
for j in 0..predictions.first().unwrap().len() {
let mut values: Vec<f64> = Vec::new();
let mut weights: Vec<f64> = Vec::new();
for (i, outer_pred) in predictions.iter().enumerate() {
values.push(outer_pred[j].prediction());
weights.push(posterior.matrix()[(subject_index, i)]);
}

for (i, outer_pred) in predictions.iter().enumerate() {
values.push(outer_pred[j].prediction());
weights.push(posterior.matrix()[(subject_index, i)]);
posterior_median.push(weighted_median(&values, &weights));
}

let median_val = weighted_median(&values, &weights);
posterior_median.push(median_val);
}

if let Some(first_spp_preds) = predictions.first() {
for (j, p) in first_spp_preds.iter().enumerate() {
let row = NPPredictionRow {
Ok(first_spp_preds
.iter()
.enumerate()
.map(|(j, p)| NPPredictionRow {
id: subject.id().clone(),
time: p.time(),
outeq: p.outeq(),
Expand All @@ -203,9 +211,15 @@ impl NPPredictions {
pop_median: pop_median[j],
post_mean: posterior_mean[j],
post_median: posterior_median[j],
};
container.add(row);
}
})
.collect())
})
.collect::<Result<Vec<_>>>()?;

let mut container = NPPredictions::new();
for rows in per_subject {
for row in rows {
container.add(row);
}
}

Expand Down