frame

a DataFrame for Javascript.

crunch numbers in Node or the Browser

features

• Interactive performance (<100ms) on millions of rows
• Syntax similar to SQL and Pandas
• Compatible with PapaParse and BabyParse

examples

Parse the Iris dataset (with BabyParse) and create a Frame from the result.

var baby = require('babyparse'),
    Frame = require('frame');

// parse the csv file
config = {"header" :true, "dynamicTyping" : true, "skipEmptyLines" : true};
iris = baby.parseFiles('iris.csv', config).data;

// create a frame from the parsed results
frame = new Frame(iris);

groupby

Group on Species and find the average value (mean) for Sepal.Length.

g = frame.groupby("Species");
g.mean("Sepal.Length");

{ "virginica": 6.58799, "versicolor": 5.9360, "setosa": 5.006 }

Using the same grouping, find the average value for Sepal.Width.

g.mean("Sepal.Width");

{ "virginica": 2.97399, "versicolor": 2.770, "setosa": 3.4279 }

where

Filter by Species value virginica then find the average.

f = frame.where("Species", "virginica");
f.mean("Sepal.Length");

6.58799

Get the number of rows that match the filter.

f.count();

50

Columns can also be accessed directly (with the filter applied).

f["Species"]

["virginica", "virginica", "virginica", ..., "virginica"]

tests

Hundreds of tests verify correctness on millions of data points (against a Pandas reference).

npm run data && npm run test

benchmarks

npm run bench

typical performance on one million rows

operation | time ----------|------ groupby | 54ms where | 29ms sum | 5ms

design goals and inspiration

• compatibility with feather

interface

• pandas
• R
• Linq
• rethinkDB
• Matlab

performance

• datavore