polars-python 0.53.0

Enable running Polars workloads in Python
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
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140

use arrow::array::Array;
use polars::prelude::*;
use polars_core::with_match_physical_numeric_polars_type;
use pyo3::prelude::*;

use super::PySeries;
use crate::utils::EnterPolarsExt;

#[pymethods]
impl PySeries {
    fn scatter(&self, py: Python<'_>, idx: PySeries, values: PySeries) -> PyResult<()> {
        py.enter_polars(|| {
            // We take the value because we want a ref count of 1 so that we can
            // have mutable access cheaply via _get_inner_mut().
            let mut lock = self.series.write();
            let s = std::mem::take(&mut *lock);
            let result = scatter(s, &idx.series.into_inner(), &values.series.into_inner());
            match result {
                Ok(out) => {
                    *lock = out;
                    Ok(())
                },
                Err((s, e)) => {
                    *lock = s; // Restore original series.
                    Err(e)
                },
            }
        })
    }
}

fn scatter(s: Series, idx: &Series, values: &Series) -> Result<Series, (Series, PolarsError)> {
    let logical_dtype = s.dtype().clone();
    let converted_values;
    let values = if logical_dtype.is_categorical() || logical_dtype.is_enum() {
        if matches!(
            values.dtype(),
            DataType::Categorical(_, _) | DataType::Enum(_, _) | DataType::String | DataType::Null
        ) {
            converted_values = values.strict_cast(&logical_dtype);
            match converted_values {
                Ok(ref values) => values,
                Err(err) => return Err((s, err)),
            }
        } else {
            return Err((
                s,
                polars_err!(InvalidOperation: "invalid values dtype '{}' for scattering into dtype '{}'", values.dtype(), logical_dtype),
            ));
        }
    } else if logical_dtype.is_decimal() {
        if values.dtype().is_numeric() {
            converted_values = values.strict_cast(&logical_dtype);
            match converted_values {
                Ok(ref values) => values,
                Err(err) => return Err((s, err)),
            }
        } else {
            return Err((
                s,
                polars_err!(InvalidOperation: "invalid values dtype '{}' for scattering into dtype '{}'", values.dtype(), logical_dtype),
            ));
        }
    } else {
        values
    };

    let null_on_oob = false;
    let idx = match polars_ops::prelude::convert_and_bound_index(idx, s.len(), null_on_oob) {
        Ok(idx) => idx,
        Err(err) => return Err((s, err)),
    };
    let idx = idx.rechunk();
    let idx = idx.downcast_as_array();
    if idx.has_nulls() {
        return Err((
            s,
            PolarsError::ComputeError("index values should not be null".into()),
        ));
    }
    let idx = idx.values().as_slice();

    let mut values = match values.to_physical_repr().cast(&s.dtype().to_physical()) {
        Ok(values) => values,
        Err(err) => return Err((s, err)),
    };

    // Broadcast values input.
    if values.len() == 1 && idx.len() > 1 {
        values = values.new_from_index(0, idx.len());
    }

    let mut phys = s.to_physical_repr().into_owned();
    drop(s); // Reduce refcount to make use of in-place mutation of possible.
    let ret = scatter_impl(&mut phys, &logical_dtype, idx, &values);
    match ret {
        Ok(s) => Ok(unsafe { s.from_physical_unchecked(&logical_dtype).unwrap() }),
        Err(e) => Err((
            unsafe { phys.from_physical_unchecked(&logical_dtype).unwrap() },
            e,
        )),
    }
}

fn scatter_impl(
    s: &mut Series,
    logical_dtype: &DataType,
    idx: &[IdxSize],
    values: &Series,
) -> PolarsResult<Series> {
    let mutable_s = s._get_inner_mut();

    match mutable_s.dtype() {
        dt if dt.is_primitive_numeric() => {
            with_match_physical_numeric_polars_type!(dt, |$T| {
                let ca: &mut ChunkedArray<$T> = mutable_s.as_mut();
                let values: &ChunkedArray<$T> = values.as_ref().as_ref();
                ca.scatter(idx, values)
            })
        },
        DataType::Boolean => {
            let ca: &mut ChunkedArray<BooleanType> = mutable_s.as_mut();
            let values = values.bool()?;
            ca.scatter(idx, values)
        },
        DataType::Binary => {
            let ca: &mut ChunkedArray<BinaryType> = mutable_s.as_mut();
            let values = values.binary()?;
            ca.scatter(idx, values)
        },
        DataType::String => {
            let ca: &mut ChunkedArray<StringType> = mutable_s.as_mut();
            let values = values.str()?;
            ca.scatter(idx, values)
        },
        _ => Err(PolarsError::ComputeError(
            format!("not yet implemented for dtype: {logical_dtype}").into(),
        )),
    }
}