reshape.hhs

/**
 * @author Jianan Lin (林家南)
 * @param input - an array / matrix / tensor
 * @param sizes - an array such as [2, 3, 2] or [2, -1]
 * @returns - an 1-d array, 2-d matrix or more-d tensor with specified size
 * 
 */


function reshape(input, sizes = [-1]) {

    *import math: flatten
    *import math: ndim
    *import math: deep_copy

    // wrong argument number
    if (arguments.length === 0) {
        throw new Error('Exception occurred in reshape - no argument given');
    }
    else if (arguments.length > 2) {
        throw new Error('Exception occurred in reshape - wrong argument number');
    }

    // type check
    if (!(Array.isArray(input)) && !(input instanceof Mat) && !(input instanceof Tensor)) {
        throw new Error('Exception occurred in reshape - argument[0] is not a Mat, Tensor or JS Array');
    }

    if (!(Array.isArray(sizes)) && !(sizes instanceof Mat) && !(sizes instanceof Tensor)) {
        throw new Error('Exception occurred in reshape - argument[1] is not a Mat, Tensor or JS Array');
    }

    // dimension check
    if (ndim(sizes) !== 1) {
        throw new Error('Exception occurred in reshape - argument[1] is not 1-dimensional');
    }

    // size check
    let dimension = sizes.length;

    // the last element can be -1, others must be positive integers
    for (let i = 0; i < dimension - 1; i++) {
        if (sizes[i] <= 0 || parseInt(sizes[i]) !== sizes[i]) {
            throw new Error('Exception occurred in reshape - elements in sizes must be positive integers');
        }
    }

    if ((sizes[dimension - 1] <= 0 && sizes[dimension - 1] !== -1) || (parseInt(sizes[dimension - 1]) !== sizes[dimension - 1])) {
        throw new Error('Exception occurred in reshape - the last element in sizes must be positive integer or -1');
    }

    // process the input
    let in_type = (input instanceof Mat) || (input instanceof Tensor);
    let raw_in = (in_type) ? input.clone().val : deep_copy(input);
    raw_in = flatten(raw_in);
    let length = raw_in.length;

    // determine whether we can reshape
    let product = 1;
    for (let i = 0; i < dimension - 1; i++) {
        product = product * sizes[i];
    }
    if (length % product !== 0) {
        throw new Error('Exception occurred in reshape - the product of array sizes must be a divisor');
    }

    if (sizes[dimension - 1] !== -1 && product * sizes[dimension - 1] !== length) {
        throw new Error('Exception occurred in reshape - the product of array sizes must be a divisor');
    }
    
    sizes[dimension - 1] = length / product;

    // next we reshape the array
    if (dimension === 1) {
        return raw_in;
    }
    else {
        let result = [];
        for (let i = 0; i < sizes[0]; i++) {
            let temp = reshape_helper(raw_in, length / sizes[0] * i, length / sizes[0] * (i + 1), sizes, 1);
            result.push(temp);
        }
        if (ndim(result) == 2) {
            return mat(result);
        }
        else {
            return new Tensor(result);
        }
    }

    // use this recursive function
    function reshape_helper(input, start, end, sizes, index) {
        // the end case
        if (index === sizes.length - 1) {
            let result = [];
            for (let i = 0; i < end - start; i++) {
                result.push(input[i + start]);
            }
            return result
        }

        // the recursion
        else {
            let result = [];
            let distance = end - start;
            for (let i = 0; i < sizes[index]; i++) {
                let temp = reshape_helper(input, start + distance / sizes[index] * i, start + distance / sizes[index] * (i + 1), sizes, index + 1);
                result.push(temp)
            }
            return result;
        }
    }
}