Skip to main content

Math and Statistics

Each valid number in Excel can be represented as an "IEEE754 double"1.

With full support for IEEE754 doubles and singles, JavaScript is an excellent language for mathematics and statistical analysis. It has also proven to be a viable platform for machine learning.

Demos

Demos for various libraries are included in separate pages:

Typed Arrays

Modern JavaScript math and statistics libraries typically use Float64Array or Float32Array objects to efficiently store data variables.

Technical details (click to show)

Under the hood, ArrayBuffer objects represent raw binary data. "Typed arrays" such as Float64Array and Float32Array are objects designed for efficient interpretation and mutation of ArrayBuffer data.

ArrayBuffer objects are roughly analogous to heap-allocated memory. Typed arrays behave like typed pointers.

JavaScript

const buf = new ArrayBuffer(16);
const dbl = new Float64Array(buf);
dbl[1] = 3.14159;
const u8 = new Uint8Array(buf);
for(let i = 0; i < 8; ++i)
console.log(u8[i+8]);

Equivalent C

void *const buf = malloc(16);
double *const dbl = (double *)buf;
dbl[1] = 3.14159;
uint8_t *const u8 = (uint8_t *)buf;
for(uint8_t i = 0; i < 8; ++i)
printf("%u\n", u8[i+8]);

Reading from Sheets

Each typed array class has a from static method for converting data into a typed array. Float64Array.from returns a double typed array (8 bytes per value) and Float32Array.from generates a float typed array (4 bytes).

const column_f32 = Float32Array.from(arr); // 4-byte floats
const column_f64 = Float64Array.from(arr); // 8-byte doubles

Values in the array will be coerced to the relevant data type. Unsupported entries will be converted to quiet NaN values.

Extracting Worksheet Data

The SheetJS sheet_to_json2 method with the option header: 1 generates an array of arrays from a worksheet object. The result is in row-major order:

const aoa = XLSX.utils.sheet_to_json(worksheet, {header: 1});

Categorical Variables

Dichotomous variables are commonly represented as spreadsheet TRUE or FALSE. The SheetJS sheet_to_json method will translate these values to true and false. Typed array methods will interpret values as 1 and 0 respectively.

Polychotomous variables must be manually mapped to numeric values. For example, using the Iris dataset:

Iris dataset

[
["sepal length", "sepal width", "petal length", "petal width", "class"],
[5.1, 3.5, 1.4, 0.2, "Iris-setosa"],
[4.9, 3, 1.4, 0.2, "Iris-setosa"],
]

Column E (class) is a polychotomous variable and must be manually translated:

const aoa = XLSX.utils.sheet_to_json(worksheet, {header: 1});

/* index_to_class will be needed to recover the values later */
const index_to_class = [];

/* map from class name to number */
const class_to_index = new Map();

/* loop over the data */
for(let R = 1; R < aoa.length; ++R) {
/* Column E = SheetJS row 4 */
const category = aoa[R][4];
const val = class_to_index.get(category);
if(val == null) {
/* assign a new index */
class_to_index.set(category, index_to_class.length);
aoa[R][4] = index_to_class.length;
index_to_class.push(category);
} else aoa[R][4] = val;
}
Live Demo (click to show)

This example fetches and parses iris.xlsx. The first worksheet is processed and the new data and mapping are printed.

Result
Loading...
Live Editor

One Variable per Column

It is common to store datasets where each row represents an observation and each column represents a variable:

Iris dataset

var aoa = [
["sepal length", "sepal width", "petal length", "petal width", "class"],
[5.1, 3.5, 1.4, 0.2, "Iris-setosa"],
[4.9, 3, 1.4, 0.2, "Iris-setosa"],
]

An array map operation can pull data from an individual column. After mapping, a slice can remove the header label. For example, the following snippet pulls column C ("petal length") into a Float64Array:

const C = XLSX.utils.decode_col("C"); // Column "C" = SheetJS index 2
const petal_length = Float64Array.from(aoa.map(row => row[C]).slice(1));

One Variable per Row

Some datasets are stored in tables where each row represents a variable and each column represents an observation:

JavaScriptSpreadsheet
var aoa = [
["sepal length", 5.1, 4.9],
["sepal width", 3.5, 3],
["petal length", 1.4, 1.4],
["petal width", 0.2, 0.2],
["class", "setosa", "setosa"]
]

Single column of data

From the row-major array of arrays, each entry of the outer array is a row.

Many sheets include header columns. The slice method can remove the header. After removing the header, Float64Array.from can generate a typed array. For example, this snippet pulls row 3 ("petal length") into a Float64Array:

const petal_length = Float64Array.from(aoa[2].slice(1));

Writing to Sheets

The SheetJS aoa_to_sheet3 method can generate a worksheet from an array of arrays. Similarly, sheet_add_aoa4 can add an array of arrays of data into an existing worksheet object. The origin option5 controls where data will be written in the worksheet.

Neither method understands typed arrays, so data columns must be converted to arrays of arrays.

One Variable per Row

A single typed array can be converted to a pure JS array with Array.from:

const arr = Array.from(row);

An array of arrays can be created from the array:

const aoa = [
arr // this array is the first element of the array literal
];

aoa_to_sheet and sheet_add_aoa treat this as one row. By default, data will be written to cells in the first row of the worksheet.

Titles can be added to data rows with an unshift operation, but it is more efficient to build up the worksheet with aoa_to_sheet:

/* sample data */
const data = new Float64Array([54337.95, 3.14159, 2.718281828]);
const title = "Values";

/* convert sample data to array */
const arr = Array.from(data);
/* create worksheet from title (array of arrays) */
const ws = XLSX.utils.aoa_to_sheet([ [ "Values" ] ]);
/* add data starting at B1 */
XLSX.utils.sheet_add_aoa(ws, [ arr ], { origin: "B1" });

Typed Array to single row with title

Live Demo (click to hide)

In this example, two typed arrays are exported. aoa_to_sheet creates the worksheet and sheet_add_aoa will add the data to the sheet.

Result
Loading...
Live Editor

One Variable per Column

A single typed array can be converted to a pure JS array with Array.from. For columns, each value should be individually wrapped in an array:

JavaScriptSpreadsheet
var data = [
[54337.95],
[3.14159],
[2.718281828]
];

Single column of data

Array.from takes a second argument. If it is a function, the function will be called on each element and the value will be used in place of the original value (in effect, mapping over the data). To generate a data column, each element must be wrapped in an array literal:

var arr = Array.from(column, (value) => ([ value ]));

aoa_to_sheet and sheet_add_aoa treat this as rows with one column of data per row. By default, data will be written to cells in column "A".

Titles can be added to data columns with an unshift operation, but it is more efficient to build up the worksheet with aoa_to_sheet:

/* sample data */
const data = new Float64Array([54337.95, 3.14159, 2.718281828]);
const title = "Values";

/* convert sample data to array */
const arr = Array.from(data, (value) => ([value]));
/* create worksheet from title (array of arrays) */
const ws = XLSX.utils.aoa_to_sheet([ [ "Values" ] ]);
/* add data starting at B1 */
XLSX.utils.sheet_add_aoa(ws, arr, { origin: "A2" });

Typed Array to single column with title

Live Demo (click to hide)

In this example, two typed arrays are exported. aoa_to_sheet creates the worksheet and sheet_add_aoa will add the data to the sheet.

Result
Loading...
Live Editor