dataframe-go

⭐   the project to show your appreciation. :arrow_upper_right:

Dataframes are used for statistics, machine-learning, and data manipulation/exploration. You can think of a Dataframe as an excel spreadsheet. This package is designed to be light-weight and intuitive.

⚠️ The package is production ready but the API is not stable yet. Once Go 1.18 (Generics) is introduced, the ENTIRE package will be rewritten. For example, there will only be 1 generic Series type. After that, version 1.0.0 will be tagged.

It is recommended your package manager locks to a commit id instead of the master branch directly. ⚠️

Features

1. Importing from CSV, JSONL, Parquet, MySQL & PostgreSQL
2. Exporting to CSV, JSONL, Excel, Parquet, MySQL & PostgreSQL
3. Developer Friendly
4. Flexible - Create custom Series (custom data types)
5. Performant
6. Interoperability with gonum package.
7. pandas sub-package
8. Fake data generation
9. Interpolation (ForwardFill, BackwardFill, Linear, Spline, Lagrange)
10. Time-series Forecasting (SES, Holt-Winters)
11. Math functions
12. Plotting (cross-platform)

See Tutorial here.

Installation

go get -u github.com/rocketlaunchr/dataframe-go

import dataframe "github.com/rocketlaunchr/dataframe-go"

DataFrames

Creating a DataFrame

…

Insert and Remove Row

df.Append(nil, 9, 123.6)

df.Append(nil, map[string]interface{}{
	"day":   10,
	"sales": nil,
})

df.Remove(0)

OUTPUT:
+-----+-------+---------+
|     |  DAY  |  SALES  |
+-----+-------+---------+
| 0:  |   2   |  23.4   |
| 1:  |   3   |  56.2   |
| 2:  |   4   |   NaN   |
| 3:  |   5   |   NaN   |
| 4:  |   6   |  84.2   |
| 5:  |   7   |   72    |
| 6:  |   8   |   89    |
| 7:  |   9   |  123.6  |
| 8:  |  10   |   NaN   |
+-----+-------+---------+
| 9X2 | INT64 | FLOAT64 |
+-----+-------+---------+

Update Row

df.UpdateRow(0, nil, map[string]interface{}{
	"day":   3,
	"sales": 45,
})

Sorting

sks := []dataframe.SortKey{
	{Key: "sales", Desc: true},
	{Key: "day", Desc: true},
}

df.Sort(ctx, sks)

OUTPUT:
+-----+-------+---------+
|     |  DAY  |  SALES  |
+-----+-------+---------+
| 0:  |   9   |  123.6  |
| 1:  |   8   |   89    |
| 2:  |   6   |  84.2   |
| 3:  |   7   |   72    |
| 4:  |   3   |  56.2   |
| 5:  |   2   |  23.4   |
| 6:  |  10   |   NaN   |
| 7:  |   5   |   NaN   |
| 8:  |   4   |   NaN   |
+-----+-------+---------+
| 9X2 | INT64 | FLOAT64 |
+-----+-------+---------+

Iterating

You can change the step and starting row. It may be wise to lock the DataFrame before iterating.

The returned value is a map containing the name of the series (string) and the index of the series (int) as keys.

iterator := df.ValuesIterator(dataframe.ValuesOptions{0, 1, true}) // Don't apply read lock because we are write locking from outside.

df.Lock()
for {
	row, vals, _ := iterator()
	if row == nil {
		break
	}
	fmt.Println(*row, vals)
}
df.Unlock()

OUTPUT:
0 map[day:1 0:1 sales:50.3 1:50.3]
1 map[sales:23.4 1:23.4 day:2 0:2]
2 map[day:3 0:3 sales:56.2 1:56.2]
3 map[1: day:4 0:4 sales:]
4 map[day:5 0:5 sales: 1:]
5 map[sales:84.2 1:84.2 day:6 0:6]
6 map[day:7 0:7 sales:72 1:72]
7 map[day:8 0:8 sales:89 1:89]

Statistics

You can easily calculate statistics for a Series using the gonum or montanaflynn/stats package.

SeriesFloat64 and SeriesTime provide access to the exported Values field to seamlessly interoperate with external math-based packages.

Example

Some series provide easy conversion using the ToSeriesFloat64 method.

import "gonum.org/v1/gonum/stat"

s := dataframe.NewSeriesInt64("random", nil, 1, 2, 3, 4, 5, 6, 7, 8)
sf, _ := s.ToSeriesFloat64(ctx)

Mean

mean := stat.Mean(sf.Values, nil)

Median

import "github.com/montanaflynn/stats"
median, _ := stats.Median(sf.Values)

Standard Deviation

std := stat.StdDev(sf.Values, nil)

Plotting (cross-platform)

import (
	chart "github.com/wcharczuk/go-chart"
	"github.com/rocketlaunchr/dataframe-go/plot"
	wc "github.com/rocketlaunchr/dataframe-go/plot/wcharczuk/go-chart"
)

sales := dataframe.NewSeriesFloat64("sales", nil, 50.3, nil, 23.4, 56.2, 89, 32, 84.2, 72, 89)
cs, _ := wc.S(ctx, sales, nil, nil)

graph := chart.Chart{Series: []chart.Series{cs}}

plt, _ := plot.Open("Monthly sales", 450, 300)
graph.Render(chart.SVG, plt)
plt.Display(plot.None)
<-plt.Closed

Output:

Math Functions

import "github.com/rocketlaunchr/dataframe-go/math/funcs"

res := 24
sx := dataframe.NewSeriesFloat64("x", nil, utils.Float64Seq(1, float64(res), 1))
sy := dataframe.NewSeriesFloat64("y", &dataframe.SeriesInit{Size: res})
df := dataframe.NewDataFrame(sx, sy)

fn := funcs.RegFunc("sin(2**x/24)")
funcs.Evaluate(ctx, df, fn, 1)

Output:

Importing Data

The imports sub-package has support for importing csv, jsonl, parquet, and directly from a SQL database. The DictateDataType option can be set to specify the true underlying data type. Alternatively, InferDataTypes option can be set.

CSV

…

Exporting Data

The exports sub-package has support for exporting to csv, jsonl, parquet, Excel and directly to a SQL database.

Optimizations

• If you know the number of rows in advance, you can set the capacity of the underlying slice of a series using SeriesInit{}. This will preallocate memory and provide speed improvements.

Generic Series

Out of the box, there is support for string, time.Time, float64 and int64. Automatic support exists for float32 and all types of integers. There is a convenience function provided for dealing with bool. There is also support for complex128 inside the xseries subpackage.

There may be times that you want to use your own custom data types. You can either implement your own Series type (more performant) or use the Generic Series (more convenient).

civil.Date

…

Tutorial

Create some fake data

Let's create a list of 8 "fake" employees with a name, title and base hourly wage rate.

import "golang.org/x/exp/rand"
import "rocketlaunchr/dataframe-go/utils/faker"

src := rand.NewSource(uint64(time.Now().UTC().UnixNano()))
df := faker.NewDataFrame(8, src, faker.S("name", 0, "Name"), faker.S("title", 0.5, "JobTitle"), faker.S("base rate", 0, "Number", 15, 50))

…

Apply Function

Let's give a promotion to everyone by doubling their salary.

s := df.Series[2]

applyFn := dataframe.ApplySeriesFn(func(val interface{}, row, nRows int) interface{} {
	return 2 * val.(int64)
})

dataframe.Apply(ctx, s, applyFn, dataframe.FilterOptions{InPlace: true})

…

Create a Time series

Let's inform all employees separately on sequential days.

import "rocketlaunchr/dataframe-go/utils/utime"

mts, _ := utime.NewSeriesTime(ctx, "meeting time", "1D", time.Now().UTC(), false, utime.NewSeriesTimeOptions{Size: &[]int{8}[0]})
df.AddSeries(mts, nil)

…

Filtering

Let's filter out our senior employees (they have titles) for no reason.

filterFn := dataframe.FilterDataFrameFn(func(vals map[interface{}]interface{}, row, nRows int) (dataframe.FilterAction, error) {
	if vals["title"] == nil {
		return dataframe.DROP, nil
	}
	return dataframe.KEEP, nil
})

seniors, _ := dataframe.Filter(ctx, df, filterFn)

…

Other useful packages

• awesome-svelte - Resources for killing react
• dbq - Zero boilerplate database operations for Go
• electron-alert - SweetAlert2 for Electron Applications
• google-search - Scrape google search results
• igo - A Go transpiler with cool new syntax such as fordefer (defer for for-loops)
• mysql-go - Properly cancel slow MySQL queries
• react - Build front end applications using Go
• remember-go - Cache slow database queries
• testing-go - Testing framework for unit testing

Legal Information

The license is a modified MIT license. Refer to LICENSE file for more details.

© 2018-21 PJ Engineering and Business Solutions Pty. Ltd.