Module aprender_compat 

Aprender Compatibility Layer

Re-exports from the aprender crate for users who need direct access to aprender’s ML primitives without adding a separate dependency.

Architecture Boundary

Entrenar owns training orchestration (autograd, optimizers, LoRA, training loop). Aprender owns ML primitives (loss functions, metrics, pruning algorithms, HF Hub client).

Entrenar delegates to aprender internally (e.g., regression metrics) and re-exports aprender’s APIs here for convenience.

Loss Functions

Aprender provides standalone loss functions that operate on Vector<f32>. For training with autograd backward passes, use entrenar’s train::LossFn trait instead.

use entrenar::aprender_compat::loss;
use entrenar::aprender_compat::primitives::Vector;

let y_pred = Vector::from_slice(&[0.9, 0.1, 0.8]);
let y_true = Vector::from_slice(&[1.0, 0.0, 1.0]);
let error = loss::mse_loss(&y_pred, &y_true);

Metrics

Aprender provides standalone metric functions. Entrenar’s train::Metric trait wraps these for integration with the training loop.

use entrenar::aprender_compat::metrics;
use entrenar::aprender_compat::primitives::Vector;

let y_pred = Vector::from_slice(&[1.1, 2.0, 3.2]);
let y_true = Vector::from_slice(&[1.0, 2.0, 3.0]);
let r2 = metrics::r_squared(&y_pred, &y_true);

Pruning

Aprender provides low-level pruning algorithms (magnitude, WANDA, SparseGPT). Entrenar’s prune module wraps these with training-loop integration.

Modules

estimators

sklearn estimator coverage via aprender’s ML algorithms (CP-05)

loss

Re-export aprender’s loss functions

metrics

Re-export aprender’s metrics

primitives

Re-export aprender’s primitive types

pruning

Re-export aprender’s pruning primitives