torch.js has not been released yet.

torch.bincount

function bincount(input: Tensor, options?: BincountOptions): Tensorfunction bincount(input: Tensor, minlength: number, weights: Tensor, options?: BincountOptions): Tensor

Count the frequency of each value in an array of non-negative integers.

Computes a histogram for integer values in range [0, max(input)]. Each position in the output contains the number of times that index value appears in the input (optionally weighted). Efficient for small integer datasets and frequency counting.

Commonly used for:

Histogram computation for integer data
Frequency tables and value counting
Building weighted histograms
Label distribution analysis
Data validation and sanity checks

\begin{aligned} \\text{output}[i] = \\sum_{j=0}^{n-1} \\mathbb{1}[\\text{input}[j] = i] \\cdot w[j] \end{aligned}

Input tensor must have non-negative integer values only
Output length = max(max(input) + 1, minlength)
If all inputs are 0, output is [n] where n = number of zeros
Weights must have the same shape as input
Efficient O(n + max(input)) algorithm
Input values must be in range [0, 2^31-1] for practical use

Parameters

inputTensor: 1D tensor of non-negative integer values (dtype must support integers)
optionsBincountOptionsoptional

Returns

Tensor– 1D tensor of counts/weights, where output[i] = number of times i appears in input (or sum of weights where input == i if weights provided)

Examples

// Basic frequency counting
const x = torch.tensor([1, 2, 1, 0, 3, 1]);
torch.bincount(x);  // [1, 3, 1, 1] - count of 0, 1, 2, 3

// With minimum length (padding)
const x = torch.tensor([0, 1, 2]);
torch.bincount(x, 5);  // [1, 1, 1, 0, 0] - padded to length 5

// Weighted histogram (soft assignments)
const x = torch.tensor([0, 1, 1, 2]);
const w = torch.tensor([0.5, 1.0, 2.0, 1.5]);
torch.bincount(x, 0, w);  // [0.5, 3.0, 1.5] - weighted counts

// Label distribution in classification
const labels = torch.tensor([0, 1, 1, 2, 0, 1, 0]);  // Class labels
const distribution = torch.bincount(labels);  // [3, 3, 1] - count per class
const weights = distribution / labels.shape[0];  // [3/7, 3/7, 1/7] - class weights

See Also

PyTorch torch.bincount()
histogram - Continuous histogram with float bin edges
histc - Histogram with specified bin range and count
unique - Get unique elements and their counts
searchsorted - Binary search for binning continuous values

BincountOptions