aprender 0.30.0

Next-generation ML framework in pure Rust — `cargo install aprender` for the `apr` CLI
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34

<!-- PCU: examples-dam-merge | contract: contracts/apr-page-examples-dam-merge-v1.yaml -->
<!-- Example: cargo run -p aprender-core --example dam_merge -->
<!-- Status: enforced -->

# Case Study: Differentiable Adaptive Merging (DAM)

**Ticket**: GH-446
**Module**: `aprender::online::dam`

## Overview

DAM learns per-model merge coefficients that minimize reconstruction loss on a calibration set. Unlike fixed-weight merging, DAM adapts weights per-tensor using Nelder-Mead simplex optimization.

## Key Components

- **`DamConfig`** — Learning rate, iterations, regularization, seed
- **`DamLoss`** — MSE loss, L2 regularization, gradient step
- **`softmax`** — Numerically stable logit-to-probability conversion
- **`optimize_coefficients`** — Nelder-Mead simplex optimizer
- **`DamReport`** — Final loss, convergence status, coefficients

## Run

```bash
cargo run --example dam_merge
```

## Falsification Tests

| ID | Property | Status |
|----|----------|--------|
| FALSIFY-DAM-001 | Softmax sums to 1 | Falsified (holds) |
| FALSIFY-DAM-002 | Optimization reduces loss | Falsified (holds) |
| FALSIFY-DAM-003 | MSE loss is non-negative | Falsified (holds) |