serde_arrow - convert sequences of structs / maps to and from arrow tables

[Crate info] | [API docs] | Changes | Example | Performance | How does it work? | Status | Development | License

Warning: this package is in an experiment at the moment.

The arrow in-memory format is a powerful way to work with data frame like structures. The surrounding ecosystem includes a rich set of libraries, ranging from data frames via Polars to query engines via DataFusion. However, the API of the underlying rust crates can be at times cumbersome to use directly due to the statically typed nature of Rust. This package, serde_arrow, tries to bridge this gap by offering a simple way to convert Rust objects into Arrow objects and vice versa. serde_arrow relies on the Serde package to interpret Rust objects. Therefore, adding support for serde_arrow to custom types is as easy as using Serde's derive macros.

In the Rust ecosystem there are two competing implemenetations of the arrow in-memory format: arrow and arrow2. serde_arrow supports both. Schema tracing and serialization from Rust structs to arrays is implemented for both. Deserialization from arrays to Rust structs is currently only implemented for arrow2.

Example

```rust

[derive(Serialize)]

struct Item { a: f32, b: i32, point: Point, }

[derive(Serialize)]

struct Point(f32, f32);

let items = vec![ Item { a: 1.0, b: 1, point: Point(0.0, 1.0) }, Item { a: 2.0, b: 2, point: Point(2.0, 3.0) }, // ... ];

// detect the field types and convert the items to arrays use serdearrow::arrow2::{serializeintofields, serializeinto_arrays};

let fields = serializeintofields(&items, TracingOptions::default())?; let arrays = serializeintoarrays(&fields, &items)?; ```

These arrays can now be written to disk using the helper method defined in the arrow2 guide. For parquet:

```rust,ignore use arrow2::{chunk::Chunk, datatypes::Schema};

// see https://jorgecarleitao.github.io/arrow2/io/parquetwrite.html writechunk( "example.pq", Schema::from(fields), Chunk::new(arrays), )?; ```

The written file can now be read in Python via

```python

using polars

import polars as pl pl.read_parquet("example.pq")

using pandas

import pandas as pd pd.read_parquet("example.pq") ```

Performance

See the implementation notes for details on how it is implemented.

This package is optimized for ease of use, not performance. Depending on the complexity of the types, a performance penality of 4x - 7x compared to manually building the arrays can be expected. More complex types incur a smaller performance penalty. See the benches for details.

Development

All common tasks are bundled in the x.py script:

```bash

format the code and run tests

python x.py precommit ```

Run python x.py --help for details. The script only uses standard Python modules can can be run without installing further packages.

License

```text Copyright (c) 2021 - 2023 Christopher Prohm

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. ```