Clean up with Copilot

.github/workflows/ci_meson.yml

@@ -3,45 +3,24 @@ on:
   push:
     branches:
       - main
-    paths:
-      - '**.c'
-      - '**.h'
-      - 'meson.build'
-      - 'meson_options.txt'
-
   pull_request:
-    branches:
-      - main
-    paths:
-      - '**.c'
-      - '**.h'
-      - 'meson.build'
-      - 'meson_options.txt'
-
 jobs:
   build:
-    name: Build and Test on ${{ matrix.os }} with Meson v${{ matrix.meson_version }}
+    name: Build and Test on ${{ matrix.os }}
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
         os: [ubuntu-latest, macos-latest, windows-latest]
-        # meson_version: ["1.2.0", "1.3.0", "1.4.0"]
         meson_version: ["1.4.0"]
     steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-      - name: Set up Python
-        uses: actions/setup-python@v5
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
         with:
           python-version: '3.x'
-      - name: Install dependencies
-        run: python -m pip install meson==${{ matrix.meson_version }} ninja
-      - name: Configure Project
-        run: meson setup builddir/
-      - name: Run Tests
-        run: meson test -C builddir/ -v
-      - name: Upload Test Log
-        uses: actions/upload-artifact@v4
+      - run: python -m pip install meson==${{ matrix.meson_version }} ninja
+      - run: meson setup builddir/
+      - run: meson test -C builddir/ -v
+      - uses: actions/upload-artifact@v4
         if: failure()
         with:
           name: ${{ matrix.os }}_Meson_Testlog

README.md

@@ -1,6 +1,6 @@
 # Epsilon
-Epsilon is a library with small functions for machine learning and statistics
-written in plain C. The functions are decoupled and well tested.
+
+Epsilon is a C library of small, decoupled, and well-tested functions for machine learning and statistics. It is designed for microcontrollers and resource-constrained environments.
 
 [![CI Meson](https://github.com/breuderink/epsilon/actions/workflows/ci_meson.yml/badge.svg)](https://github.com/breuderink/epsilon/actions/workflows/ci_meson.yml)
 
@@ -31,40 +31,50 @@ To allow machine learning to run on microcontrollers, the implementations:
 - work with fixed-point math when realistic, and
 - are easy to tune.
 
-# Building
-Epsilon uses Meson for building. Install Meson and Ninja, create a build
 directory, and configure the project. In the repository root, configure and
 build the project, then run the unit tests and examples:
 
+# Folder Structure
+
+```
+/           # Project root
+├── LICENSE
+├── README.md
+├── meson.build
+├── docs/           # Documentation and references
+├── src/            # Source code (.c/.h)
+├── tests/          # Unit tests
+├── examples/       # Example programs
+├── subprojects/    # External dependencies (e.g., Unity)
+```
+
+# Building & Testing
+Epsilon uses Meson for building. In the repository root:
+
 ```bash
-$ meson setup builddir
-$ meson test -C builddir
+meson setup builddir
+meson compile -C builddir
+meson test -C builddir
 ```
 
+
 # Algorithms
+
 ## Pseudo-random number generation
-- [Xorshift](docs/marsaglia2003xrn.pdf) is a fast and simple
-pseudo-random number generator by George Marsaglia that has good statistical
-properties. See the [xorshift example](examples/example_rng.c).
+- [Xorshift](docs/marsaglia2003xrn.pdf): Fast, simple PRNG with good statistical properties. See `examples/example_rng.c`.
 
 ## Hashing
-- The [FNV hash](https://tools.ietf.org/html/draft-eastlake-fnv-17) is a fast
-hash function that maps variable length input to a fixed output
-([example](examples/example_hash.c)). It can be used for [feature
-hashing](https://en.wikipedia.org/wiki/Feature_hashing).
+- [FNV hash](https://tools.ietf.org/html/draft-eastlake-fnv-17): Fast hash for feature hashing. See `examples/example_hash.c`.
 
 ## Statistics
-- Welford's method computes mean and variance in a single pass. See the
-[example of Welford's method](examples/example_stats.c).
+- Welford's method: Online mean and variance in one pass. See `examples/example_stats.c`.
 
 ## Transformations
-- Fast Walsh-Hadamard transform (FWHT) implements the Walsh-Hadamard
-transform in O(n log n) time. FWHT is similar to the fast Fourier transform
-and the Haar transform. See the [FWHT example](examples/example_transform.c).
+- Fast Walsh-Hadamard Transform (FWHT): O(n log n) transform, similar to FFT. See `examples/example_transform.c`.
 
 ## Passive-aggressive learning
-- [Online passive-aggressive (PA)](docs/crammer2006opa.pdf) regression solves a
-regression problem by only updating the model on prediction mistakes. 
+- [Online passive-aggressive (PA)](docs/crammer2006opa.pdf): Regression with updates only on mistakes.
+
 
 # Other solutions for Tiny ML or Edge AI
 - [TensorFlow Lite](https://www.tensorflow.org/lite/)

examples/example_hash.c

@@ -1,4 +1,4 @@
-#include "epsilon/hash.h"
+#include "hash.h"
 #include <stdint.h>
 #include <stdio.h>
 #include <string.h>

examples/example_rng.c

@@ -1,4 +1,4 @@
-#include "epsilon/rng.h"
+#include "rng.h"
 #include <inttypes.h>
 #include <stdio.h>
 

examples/example_stats.c

@@ -1,4 +1,4 @@
-#include "epsilon/stats.h"
+#include "stats.h"
 #include <stdio.h>
 
 int main(void) {

examples/example_transform.c

@@ -1,4 +1,4 @@
-#include "epsilon/transform.h"
+#include "transform.h"
 #include <stdio.h>
 
 #define LOG2_DIMS 3

meson.build

@@ -3,71 +3,87 @@ project('epsilon', 'c',
   default_options: ['c_std=c99', 'warning_level=3', 'werror=true']
 )
 
+# Dependencies
 cc = meson.get_compiler('c')
 math_dep = cc.find_library('m', required: false)
 unity_dep = dependency('unity', required: true)
 
-# Library sources
-epsilon_sources = [
-  'epsilon/rng.c',
-  'epsilon/stats.c',
-  'epsilon/hash.c',
-  'epsilon/transform.c'
-]
+# Library configuration
 
-# Build the library
+epsilon_sources = files(
+  'src/rng.c',
+  'src/stats.c',
+  'src/hash.c',
+  'src/transform.c'
+)
+
+
+epsilon_headers = files(
+  'src/rng.h',
+  'src/stats.h',
+  'src/hash.h',
+  'src/transform.h',
+  'src/pa.h'
+)
+
+# Build library
 epsilon_lib = library('epsilon',
   sources: epsilon_sources,
   dependencies: math_dep,
   install: true,
   version: meson.project_version()
 )
 
-# Dependency object for downstream use
+# Declare dependency for downstream use
+
 epsilon_dep = declare_dependency(
   link_with: epsilon_lib,
+  include_directories: include_directories('src'),
   dependencies: math_dep
 )
 
 # Install headers
-epsilon_headers = [
-  'epsilon/rng.h',
-  'epsilon/stats.h',
-  'epsilon/hash.h',
-  'epsilon/transform.h',
-  'epsilon/pa.h'
-]
+
 install_headers(epsilon_headers, subdir: 'epsilon')
 
-# Examples
-example_exes = [
-  ['example_hash', 'examples/example_hash.c'],
-  ['example_transform', 'examples/example_transform.c'],
-  ['example_stats', 'examples/example_stats.c'],
-  ['example_rng', 'examples/example_rng.c']
-]
-
-foreach pair : example_exes
-  exe = executable(pair[0], pair[1], dependencies: epsilon_dep, install: false)
-  test(pair[0], exe)
+# Examples and tests configuration
+examples = {
+  'example_hash': 'examples/example_hash.c',
+  'example_transform': 'examples/example_transform.c',
+  'example_stats': 'examples/example_stats.c',
+  'example_rng': 'examples/example_rng.c'
+}
+
+tests = {
+  'hash_test': 'tests/hash_test.c',
+  'rng_test': 'tests/rng_test.c',
+  'stats_test': 'tests/stats_test.c',
+  'transform_test': 'tests/transform_test.c',
+}
+
+# Build and register examples
+foreach name, source : examples
+  exe = executable(name, source,
+    dependencies: epsilon_dep,
+    install: false
+  )
+  test(name, exe)
 endforeach
 
-# Tests
-test_exes = [
-  ['hash_test', 'tests/hash_test.c'],
-  ['rng_test', 'tests/rng_test.c'],
-  ['stats_test', 'tests/stats_test.c'],
-  ['transform_test', 'tests/transform_test.c']
-]
-
-foreach pair : test_exes
-  exe = executable(pair[0], pair[1], dependencies: [epsilon_dep, unity_dep], install: false)
-  test(pair[0], exe)
+# Build and register tests
+foreach name, source : tests
+  exe = executable(name, source,
+    dependencies: [epsilon_dep, unity_dep],
+    install: false
+  )
+  test(name, exe)
 endforeach
 
-# Documentation
-install_data('README.md', install_dir: 'share/doc/epsilon')
-install_data('LICENSE', install_dir: 'share/licenses/epsilon')
+# Documentation installation
+install_data(
+  files('README.md', 'LICENSE'),
+  install_dir: 'share/doc/epsilon'
+)
 
-# Optional metadata
+# Distribution metadata
 meson.add_dist_script('echo', 'Package contact: boris@cortext.nl')

epsilon/hash.c → src/hash.c

@@ -1,3 +1,5 @@
+
+// FNV-1a 32-bit hash implementation.
 #include <stddef.h>
 #include <stdint.h>
 

src/hash.h

@@ -0,0 +1,12 @@
+
+#ifndef EPSILON_HASH_H
+#define EPSILON_HASH_H
+
+// FNV-1a 32-bit hash interface
+#include <stddef.h>
+#include <stdint.h>
+
+uint32_t FNV1a32_update(uint32_t hash, uint8_t data);
+uint32_t FNV1a32_hash(const void *data, size_t n);
+
+#endif // EPSILON_HASH_H

src/pa.h

@@ -0,0 +1,6 @@
+
+// Passive-aggressive regression parameters.
+typedef struct {
+	float C;
+	float eps;
+} PA_t;