Packages for working with Apache Arrow in Go.
Included in gomem is a DataFrame implementation. It uses Apache Arrow (Go) under the hood to store and manipulate data in a columnar format.
Tables | Description | Link |
---|---|---|
dataframe | A DataFrame implementation using Arrow. | code |
collection | Abstract access to Arrow arrays using gomem Objects. | code |
iterator | Iterators for iterating over Arrow arrays. | code |
logical | Abstract logical types. | code |
object | Abstract object type capable of automatically converting Object types. | code |
smartbuilder | Abstract Arrow array builder. | code |
A DataFrame built on Apache Arrow.
Add the package to your go.mod
file:
require github.com/gomem/gomem master
Or, clone the repository:
git clone --branch master https://github.com/gomem/gomem.git $GOPATH/src/github.com/gomem/gomem
A complete example:
mkdir my-dataframe-app && cd my-dataframe-app
cat > go.mod <<-END
module my-dataframe-app
require github.com/gomem/gomem master
END
cat > main.go <<-END
package main
import (
"fmt"
"github.com/apache/arrow/go/arrow/memory"
"github.com/gomem/gomem/pkg/dataframe
)
func main() {
pool := memory.NewGoAllocator()
df, _ := dataframe.NewDataFrameFromMem(pool, dataframe.Dict{
"col1": []int32{1, 2, 3, 4, 5},
"col2": []float64{1.1, 2.2, 3.3, 4.4, 5},
"col3": []string{"foo", "bar", "ping", "", "pong"},
"col4": []interface{}{2, 4, 6, nil, 8},
})
defer df.Release()
fmt.Printf("DataFrame:\n%s\n", df.Display(0))
}
// DataFrame:
// rec[0]["col1"]: [1 2 3 4 5]
// rec[0]["col2"]: [1.1 2.2 3.3 4.4 5]
// rec[0]["col3"]: ["foo" "bar" "ping" "" "pong"]
// rec[0]["col4"]: [2 4 6 (null) 8]
END
go run main.go
See the DataFrame tests for extensive usage examples.
From the arrow/go README...
The library makes use of reference counting so that it can track when memory buffers are no longer used. This allows Arrow to update resource accounting, pool memory such and track overall memory usage as objects are created and released. Types expose two methods to deal with this pattern. The
Retain
method will increase the reference count by 1 andRelease
method will reduce the count by 1. Once the reference count of an object is zero, any associated object will be freed.Retain
andRelease
are safe to call from multiple goroutines.
-
If you are passed an object and wish to take ownership of it, you must call
Retain
. You must later pair this with a call toRelease
when you no longer need the object. "Taking ownership" typically means you wish to access the object outside the scope of the current function call. -
You own any object you create via functions whose name begins with
New
orCopy
or any operation that results in a new immutable DataFrame being returned or when receiving an object over a channel. Therefore you must callRelease
once you no longer need the object. -
If you send an object over a channel, you must call
Retain
before sending it as the receiver is assumed to own the object and will later callRelease
when it no longer needs the object.
Note: You can write a test using memory.NewCheckedAllocator
to assert that you have
released all resources properly. See: tests
This DataFrame currently implements most of the scalar types we've come across. There is still work to be done on some of the list and struct types. Feel free to submit a PR if find you need them. This library will let you know when you do.
- Implement all Arrow DataTypes.
- Add a filter function to DataFrame.
- Add an order by function to DataFrame.
Pull requests are welcome!
(c) 2020 Nick Poorman. Licensed under the Apache License, Version 2.0.