Implementation:LaurentMazare Tch rs ModuleT Batch Accuracy

Knowledge Sources	tch-rs
Domains	Deep_Learning, Model_Evaluation
Last Updated	2026-02-08 14:00 GMT

Overview

Concrete tool for computing batched classification accuracy over a test set provided by the tch nn module.

Description

ModuleT::batch_accuracy_for_logits is a default method on the ModuleT trait that evaluates a model over a full dataset in mini-batches. It creates a no_grad_guard, iterates over the dataset using Iter2 with return_smaller_last_batch(), computes accuracy_for_logits for each batch, and returns the weighted average accuracy as an f64 in [0.0, 1.0].

Usage

Call on any model implementing ModuleT to evaluate test accuracy. Pass the full test images, labels, device, and evaluation batch size.

Code Reference

Source Location

• Repository: tch-rs
• File: src/nn/module.rs
• Lines: 16-33

Signature

fn batch_accuracy_for_logits(
    &self,
    xs: &Tensor,
    ys: &Tensor,
    d: Device,
    batch_size: i64,
) -> f64

Import

use tch::nn::ModuleT;

I/O Contract

Inputs

Name	Type	Required	Description
xs	&Tensor	Yes	Full test set images tensor
ys	&Tensor	Yes	Full test set labels tensor
d	Device	Yes	Device for computation (should match model device)
batch_size	i64	Yes	Evaluation mini-batch size

Outputs

Name	Type	Description
f64	f64	Classification accuracy in [0.0, 1.0]

Usage Examples

use tch::nn::ModuleT;

// After training
let test_accuracy = model.batch_accuracy_for_logits(
    &dataset.test_images,
    &dataset.test_labels,
    vs.device(),
    256,  // eval batch size
);
println!("Test accuracy: {:.2}%", 100.0 * test_accuracy);

Related Pages

Implements Principle

• Principle:LaurentMazare_Tch_rs_Batch_Evaluation