Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 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 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 | 1x 3x 3x 6x 6x 3x 3x 1x 1x 1x 1x 1x 1x 1x 1x 1x 2x 2x 2x 1x 1x 1x 347x 347x 347x 347x 347x 5x 5x 5x 5x 5x 347x 108x 108x 108x 108x 108x 234x 234x 234x 347x 34196x 34196x 234x 234x 234x 1x 413x 413x 413x 413x 2785x 2785x 2785x 413x 413x 1x 399x 399x 399x 399x 2728x 2728x 2728x 399x 399x 1x 1x 1x 1x 1x 3x 3x 1x 1x 1x 1x 1x 1x 1x 1x 1x 74x 74x 29152x 29152x 74x 74x 1x 1x 1x 4x 4x 1x 1x 1x 165x 165x 165x 165x 57702x 57702x 165x 165x 1x 1x 1x 1x 1x 3x 3x 1x 1x 1x 166x 166x 57712x 57712x 166x 166x 1x 3x 3x 3x 10x 10x 10x 10x 10x 3x 3x 1x 2x 2x 2x 1x 1x 2x 5x 5x 1x 1x 1x 162x 20259x 20259x 20259x 20259x 142x 142x 1x 3x 7x 7x 1x 1x | /**
* Low-level tensor utility helpers.
*
* These handle operations that, in the Python code, are done by
* PyTorch/NumPy broadcasting, reshaping, and slicing. We implement them
* manually on top of Float32Array for zero external dependencies.
*/
// ---------------------------------------------------------------------------
// Reshaping / flattening
// ---------------------------------------------------------------------------
/**
* Reshape a flat array into a 2-D matrix of `[rows, cols]`.
*
* Returns an array of `rows` Float32Arrays each of length `cols`.
*/
export function reshape2D(flat: Float32Array, rows: number, cols: number): Float32Array[] {
const result: Float32Array[] = [];
for (let r = 0; r < rows; r++) {
result.push(flat.slice(r * cols, (r + 1) * cols));
}
return result;
}
/**
* Reshape a flat array into a 3-D tensor of `[d0, d1, d2]`.
*/
export function reshape3D(
flat: Float32Array,
d0: number,
d1: number,
d2: number,
): Float32Array[][] {
const result: Float32Array[][] = [];
const stride = d1 * d2;
for (let i = 0; i < d0; i++) {
const slice = flat.slice(i * stride, (i + 1) * stride);
result.push(reshape2D(slice, d1, d2));
}
return result;
}
// ---------------------------------------------------------------------------
// Padding
// ---------------------------------------------------------------------------
/**
* Left-pad an array to length `targetLen` with zeros, returning a mask
* where 1 = padding, 0 = original value.
*/
export function leftPad(
arr: Float32Array,
targetLen: number,
): { padded: Float32Array; mask: Uint8Array } {
const arrLen = arr.length;
if (arrLen > targetLen) {
return {
padded: arr.slice(arrLen - targetLen),
mask: new Uint8Array(targetLen), // all zeros (no padding)
};
}
if (arrLen === targetLen) {
// Return the array as-is to avoid unnecessary O(n) copy
return {
padded: arr,
mask: new Uint8Array(targetLen), // all zeros (no padding)
};
}
const padLen = targetLen - arrLen;
const padded = new Float32Array(targetLen);
const mask = new Uint8Array(targetLen);
// Fill padding
for (let i = 0; i < padLen; i++) {
mask[i] = 1;
}
// Copy original values
padded.set(arr, padLen);
return { padded, mask };
}
// ---------------------------------------------------------------------------
// Concatenation
// ---------------------------------------------------------------------------
/**
* Concatenate an array of Float32Arrays into a single flat Float32Array.
*/
export function concat(arrays: Float32Array[]): Float32Array {
const totalLen = arrays.reduce((sum, a) => sum + a.length, 0);
const result = new Float32Array(totalLen);
let offset = 0;
for (const arr of arrays) {
result.set(arr, offset);
offset += arr.length;
}
return result;
}
/**
* Concatenate an array of Uint8Arrays into a single flat Uint8Array.
*/
export function concatUint8(arrays: Uint8Array[]): Uint8Array {
const totalLen = arrays.reduce((sum, a) => sum + a.length, 0);
const result = new Uint8Array(totalLen);
let offset = 0;
for (const arr of arrays) {
result.set(arr, offset);
offset += arr.length;
}
return result;
}
/**
* Concatenate along a new axis (stacks arrays as rows).
* Returns a flat Float32Array where result[i * colLen + j] = arrays[i][j].
*/
export function stack(arrays: Float32Array[]): Float32Array {
if (arrays.length === 0) return new Float32Array(0);
const colLen = arrays[0].length;
const result = new Float32Array(arrays.length * colLen);
for (let i = 0; i < arrays.length; i++) {
result.set(arrays[i], i * colLen);
}
return result;
}
// ---------------------------------------------------------------------------
// Slicing
// ---------------------------------------------------------------------------
/**
* Extract a slice from each array in a batch, returning new arrays.
*/
export function sliceEach(arrays: Float32Array[], start: number, end?: number): Float32Array[] {
return arrays.map((arr) => arr.slice(start, end));
}
/**
* Take the last N elements from each array in a batch.
*/
export function takeLast(arrays: Float32Array[], n: number): Float32Array[] {
return arrays.map((arr) => arr.slice(Math.max(0, arr.length - n)));
}
// ---------------------------------------------------------------------------
// Clip / clamp
// ---------------------------------------------------------------------------
/**
* Element-wise maximum with `minVal`.
*/
export function clipMin(values: Float32Array, minVal: number): Float32Array {
const result = new Float32Array(values.length);
for (let i = 0; i < values.length; i++) {
result[i] = Math.max(values[i], minVal);
}
return result;
}
/**
* Element-wise minimum with `maxVal`.
*/
export function clipMax(values: Float32Array, maxVal: number): Float32Array {
const result = new Float32Array(values.length);
for (let i = 0; i < values.length; i++) {
result[i] = Math.min(values[i], maxVal);
}
return result;
}
// ---------------------------------------------------------------------------
// Arithmetic helpers
// ---------------------------------------------------------------------------
/**
* Element-wise mean of two arrays.
*
* @throws {RangeError} if the arrays have different lengths.
*/
export function elementwiseMean(a: Float32Array, b: Float32Array): Float32Array {
if (a.length !== b.length) {
throw new RangeError(`Length mismatch: a.length=${a.length}, b.length=${b.length}`);
}
const len = a.length;
const result = new Float32Array(len);
for (let i = 0; i < len; i++) {
result[i] = (a[i] + b[i]) / 2;
}
return result;
}
/**
* Element-wise difference: a[i] - b[i].
*
* @throws {RangeError} if the arrays have different lengths.
*/
export function elementwiseDiff(a: Float32Array, b: Float32Array): Float32Array {
if (a.length !== b.length) {
throw new RangeError(`Length mismatch: a.length=${a.length}, b.length=${b.length}`);
}
const len = a.length;
const result = new Float32Array(len);
for (let i = 0; i < len; i++) {
result[i] = a[i] - b[i];
}
return result;
}
/**
* Negate all elements.
*/
export function negate(arr: Float32Array): Float32Array {
const result = new Float32Array(arr.length);
for (let i = 0; i < arr.length; i++) {
result[i] = -arr[i];
}
return result;
}
// ---------------------------------------------------------------------------
// Statistical helpers
// ---------------------------------------------------------------------------
/**
* Mean of a 1-D array. Skips NaN and non-finite values.
*
* For numerically-stable, masked statistics consider using `computeStats()`
* from `./stats` instead.
*/
export function mean(arr: Float32Array): number {
let sum = 0;
let count = 0;
for (let i = 0; i < arr.length; i++) {
if (Number.isFinite(arr[i])) {
sum += arr[i];
count++;
}
}
return count > 0 ? sum / count : 0;
}
/**
* Population standard deviation of a 1-D array. Skips NaN and non-finite values.
*
* Uses a two-pass algorithm for numerical stability. For production
* use with weighting or masks, prefer `computeStats()` from `./stats`.
*/
export function std(arr: Float32Array): number {
let count = 0;
for (let i = 0; i < arr.length; i++) if (Number.isFinite(arr[i])) count++;
if (count <= 1) return 0;
const m = mean(arr);
let sumSq = 0;
for (let i = 0; i < arr.length; i++) {
if (Number.isFinite(arr[i])) sumSq += (arr[i] - m) ** 2;
}
return Math.sqrt(Math.max(0, sumSq / count));
}
/**
* Check if all values in a Float32Array are non-negative.
*
* Skips NaN values (NaN < 0 is false, so NaN would be silently treated
* as non-negative in a naive implementation). A series containing NaN
* cannot be assumed positive — it should be treated as having unknown sign.
*/
export function allNonNegative(arr: Float32Array): boolean {
for (let i = 0; i < arr.length; i++) {
const v = arr[i];
if (Number.isNaN(v)) return false; // NaN → unknown sign, treat as not positive
if (v < 0) return false;
}
return true;
}
/**
* Check if any value in a Float32Array is NaN or Infinity.
*/
export function hasInvalid(arr: Float32Array): boolean {
for (let i = 0; i < arr.length; i++) {
if (!Number.isFinite(arr[i])) return true;
}
return false;
}
|