How to create a Dictionary encoded string column with the Javascript interface?

[Yakov Galka <ya...@stannum.io>]

Hi All,

I have code that creates a table with string columns as follows:

for(/* each column */) {
    // ...
    column_vectors.push(Vector.new(Data.Utf8(new Utf8(), 0, element_count,
null_count, nullmap_buffer, offsets_buffer, data_buffer)));
}
const arrow_table = Table.new(column_vectors, column_names);
const data = arrow_table.serialize('binary', false).buffer;
const arrow_table2 = Table.from([new Uint8Array(data)]);

Here offsets_buffer is a Int32Array with the offsets and data_buffer is a
Uint8Array with the strings, in accordance to the Arrow format described in
https://arrow.apache.org/docs/format/Columnar.html.

I am trying to change this to use a dictionary encoding instead. I change
the producer of the data to return only the unique strings in data_buffer
and offsets_buffer, and additionally produce an interned_buffer
(Int32Array) with the indices of the strings. However I couldn't find how
to initialize the column in Javascript.

Shooting in the dark, I tried:

for(/* each column */) {
    // ...
    const dictionary = Vector.new(Data.Utf8(new Utf8(), 0,
offsets_buffer.length - 1, 0, 0, offsets_buffer, data_buffer));
    column_vectors.push(Vector.new(Data.Dictionary(new Dictionary(new
Utf8(), new Int32()), 0, element_count, null_count, nullmap_buffer, 0,
interned_buffer, dictionary)));
}
// ...

However, this causes the deserialization (Table.from) to fail with:

TypeError: undefined has no properties
    visitUtf8
    visit
    visit
    visitMany
    map
    visitMany
    _loadVectors
    _loadDictionaryBatch
    _readDictionaryBatch
    open
    open
    from

What's the correct way of creating a dictionary encoded column?

Yakov Galka
http://stannum.io/

Re: How to create a Dictionary encoded string column with the Javascript interface?

[Yakov Galka <ya...@stannum.io>]

Thank you!

Someone else pointed out that Dictionary doesn't accept an offsets buffer,
therefore it should have been:

    column_vectors.push(Vector.new(Data.Dictionary(new Dictionary(new
Utf8(), new Int32()), 0, element_count, null_count, nullmap_buffer,
interned_buffer, dictionary)));

-- i.e. without the zero after nullmap_buffer. This solved the problem.

Yakov Galka
http://stannum.io/

Re: How to create a Dictionary encoded string column with the Javascript interface?

[Micah Kornfield <em...@gmail.com>]

Hi Yakov,
You might try e-mailing the dev@ mailing list to see if anyone responds
there.  I'm not sure how many javascript devs are subscribed here.

Cheers,
Micah

On Thu, Jul 16, 2020 at 6:48 PM Yakov Galka <ya...@stannum.io> wrote:

> Hi All,
>
> I have code that creates a table with string columns as follows:
>
> for(/* each column */) {
>     // ...
>     column_vectors.push(Vector.new(Data.Utf8(new Utf8(), 0, element_count,
> null_count, nullmap_buffer, offsets_buffer, data_buffer)));
> }
> const arrow_table = Table.new(column_vectors, column_names);
> const data = arrow_table.serialize('binary', false).buffer;
> const arrow_table2 = Table.from([new Uint8Array(data)]);
>
> Here offsets_buffer is a Int32Array with the offsets and data_buffer is a
> Uint8Array with the strings, in accordance to the Arrow format described in
> https://arrow.apache.org/docs/format/Columnar.html.
>
> I am trying to change this to use a dictionary encoding instead. I change
> the producer of the data to return only the unique strings in data_buffer
> and offsets_buffer, and additionally produce an interned_buffer
> (Int32Array) with the indices of the strings. However I couldn't find how
> to initialize the column in Javascript.
>
> Shooting in the dark, I tried:
>
> for(/* each column */) {
>     // ...
>     const dictionary = Vector.new(Data.Utf8(new Utf8(), 0,
> offsets_buffer.length - 1, 0, 0, offsets_buffer, data_buffer));
>     column_vectors.push(Vector.new(Data.Dictionary(new Dictionary(new
> Utf8(), new Int32()), 0, element_count, null_count, nullmap_buffer, 0,
> interned_buffer, dictionary)));
> }
> // ...
>
> However, this causes the deserialization (Table.from) to fail with:
>
> TypeError: undefined has no properties
>     visitUtf8
>     visit
>     visit
>     visitMany
>     map
>     visitMany
>     _loadVectors
>     _loadDictionaryBatch
>     _readDictionaryBatch
>     open
>     open
>     from
>
> What's the correct way of creating a dictionary encoded column?
>
> Yakov Galka
> http://stannum.io/
>