df-derive 0.3.0

Derive fast conversions from Rust structs into Polars DataFrames.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48

//! Demonstrates deep `Vec<Vec<Vec<T>>>` nesting.
//!
//! Each `Vec` layer becomes a `LargeListArray` wrap, so a triple-Vec column
//! materializes as `List<List<List<T>>>` in the output frame. The derive
//! fuses the wrapper layers into a single bulk emission, so per-row work is
//! O(total leaf count) rather than O(layer count * leaf count).
//!
//! Uses the default `df-derive` facade runtime.

use df_derive::ToDataFrame;
use df_derive::dataframe;
use df_derive::dataframe::ToDataFrameVec;

#[derive(ToDataFrame)]
struct Tensor {
    label: String,
    // List<List<List<f64>>> after derive: outer batches, sequences, frames.
    values: Vec<Vec<Vec<f64>>>,
    // List<List<i32>> for a smaller comparison column.
    counts: Vec<Vec<i32>>,
}

fn main() -> polars::prelude::PolarsResult<()> {
    let tensors = vec![
        Tensor {
            label: "alpha".into(),
            values: vec![vec![vec![1.0, 2.0], vec![3.0]], vec![vec![4.0, 5.0, 6.0]]],
            counts: vec![vec![1, 2, 3], vec![]],
        },
        Tensor {
            label: "beta".into(),
            values: vec![],
            counts: vec![vec![10]],
        },
    ];

    let df = tensors.as_slice().to_dataframe()?;
    println!("Deep Vec<Vec<Vec<T>>> DataFrame:");
    println!("{df}");

    let schema = <Tensor as dataframe::ToDataFrame>::schema()?;
    println!("\nSchema (each Vec layer becomes a List wrap):");
    for (name, dtype) in schema {
        println!("  {name}: {dtype:?}");
    }

    Ok(())
}