polars_python/map/
lazy.rs

1use std::mem::{ManuallyDrop, MaybeUninit};
2
3use polars::prelude::*;
4use polars_ffi::version_0::SeriesExport;
5use pyo3::prelude::*;
6use pyo3::types::PyList;
7
8use crate::py_modules::{pl_series, polars, polars_rs};
9use crate::series::PySeries;
10use crate::{PyExpr, Wrap};
11
12pub(crate) trait ToSeries {
13    fn to_series(
14        &self,
15        py: Python<'_>,
16        py_polars_module: &Py<PyModule>,
17        name: &str,
18    ) -> PolarsResult<Series>;
19}
20
21impl ToSeries for PyObject {
22    fn to_series(
23        &self,
24        py: Python<'_>,
25        py_polars_module: &Py<PyModule>,
26        name: &str,
27    ) -> PolarsResult<Series> {
28        let py_pyseries = match self.getattr(py, "_s") {
29            Ok(s) => s,
30            // the lambda did not return a series, we try to create a new python Series
31            _ => {
32                let res = py_polars_module
33                    .getattr(py, "Series")
34                    .unwrap()
35                    .call1(py, (name, PyList::new(py, [self]).unwrap()));
36
37                match res {
38                    Ok(python_s) => python_s.getattr(py, "_s").unwrap(),
39                    Err(_) => {
40                        polars_bail!(ComputeError:
41                            "expected a something that could convert to a `Series` but got: {}",
42                            self.bind(py).get_type()
43                        )
44                    },
45                }
46            },
47        };
48        let s = match py_pyseries.extract::<PySeries>(py) {
49            Ok(pyseries) => pyseries.series,
50            // This happens if the executed Polars is not from this source.
51            // Currently only happens in PC-workers
52            Err(_) => {
53                let mut export: MaybeUninit<SeriesExport> = MaybeUninit::uninit();
54                py_pyseries
55                    .call_method1(py, "_export", (&raw mut export as usize,))
56                    .unwrap();
57                unsafe {
58                    let export = export.assume_init();
59                    polars_ffi::version_0::import_series(export)?
60                }
61            },
62        };
63        Ok(s)
64    }
65}
66
67pub(crate) fn call_lambda_with_series(
68    py: Python<'_>,
69    s: &Series,
70    lambda: &PyObject,
71) -> PyResult<PyObject> {
72    let pypolars = polars(py).bind(py);
73
74    // create a PySeries struct/object for Python
75    let pyseries = PySeries::new(s.clone());
76    // Wrap this PySeries object in the python side Series wrapper
77    let mut python_series_wrapper = pypolars
78        .getattr("wrap_s")
79        .unwrap()
80        .call1((pyseries,))
81        .unwrap();
82
83    if !python_series_wrapper
84        .getattr("_s")
85        .unwrap()
86        .is_instance(polars_rs(py).getattr(py, "PySeries").unwrap().bind(py))
87        .unwrap()
88    {
89        let mut export = ManuallyDrop::new(polars_ffi::version_0::export_series(s));
90        let plseries = pl_series(py).bind(py);
91
92        let s_location = &raw mut export;
93        python_series_wrapper = plseries
94            .getattr("_import")
95            .unwrap()
96            .call1((s_location as usize,))
97            .unwrap()
98    }
99    lambda.call1(py, (python_series_wrapper,))
100}
101
102/// A python lambda taking two Series
103pub(crate) fn binary_lambda(
104    lambda: &PyObject,
105    a: Series,
106    b: Series,
107) -> PolarsResult<Option<Series>> {
108    Python::with_gil(|py| {
109        // get the pypolars module
110        let pypolars = polars(py).bind(py);
111        // create a PySeries struct/object for Python
112        let pyseries_a = PySeries::new(a);
113        let pyseries_b = PySeries::new(b);
114
115        // Wrap this PySeries object in the python side Series wrapper
116        let python_series_wrapper_a = pypolars
117            .getattr("wrap_s")
118            .unwrap()
119            .call1((pyseries_a,))
120            .unwrap();
121        let python_series_wrapper_b = pypolars
122            .getattr("wrap_s")
123            .unwrap()
124            .call1((pyseries_b,))
125            .unwrap();
126
127        // call the lambda and get a python side Series wrapper
128        let result_series_wrapper =
129            match lambda.call1(py, (python_series_wrapper_a, python_series_wrapper_b)) {
130                Ok(pyobj) => pyobj,
131                Err(e) => polars_bail!(
132                    ComputeError: "custom python function failed: {}", e.value(py),
133                ),
134            };
135        let pyseries = if let Ok(expr) = result_series_wrapper.getattr(py, "_pyexpr") {
136            let pyexpr = expr.extract::<PyExpr>(py).unwrap();
137            let expr = pyexpr.inner;
138            let df = DataFrame::empty();
139            let out = df
140                .lazy()
141                .select([expr])
142                .with_predicate_pushdown(false)
143                .with_projection_pushdown(false)
144                .collect()?;
145
146            let s = out.select_at_idx(0).unwrap().clone();
147            PySeries::new(s.take_materialized_series())
148        } else {
149            return Some(result_series_wrapper.to_series(py, pypolars.as_unbound(), ""))
150                .transpose();
151        };
152
153        // Finally get the actual Series
154        Ok(Some(pyseries.series))
155    })
156}
157
158pub fn map_single(
159    pyexpr: &PyExpr,
160    lambda: PyObject,
161    output_type: Option<Wrap<DataType>>,
162    agg_list: bool,
163    is_elementwise: bool,
164    returns_scalar: bool,
165) -> PyExpr {
166    let output_type = output_type.map(|wrap| wrap.0);
167
168    let func =
169        python_dsl::PythonUdfExpression::new(lambda, output_type, is_elementwise, returns_scalar);
170    pyexpr.inner.clone().map_python(func, agg_list).into()
171}
172
173pub(crate) fn call_lambda_with_columns_slice(
174    py: Python<'_>,
175    s: &[Column],
176    lambda: &PyObject,
177    pypolars: &Py<PyModule>,
178) -> PyObject {
179    let pypolars = pypolars.bind(py);
180
181    // create a PySeries struct/object for Python
182    let iter = s.iter().map(|s| {
183        let ps = PySeries::new(s.as_materialized_series().clone());
184
185        // Wrap this PySeries object in the python side Series wrapper
186        pypolars.getattr("wrap_s").unwrap().call1((ps,)).unwrap()
187    });
188    let wrapped_s = PyList::new(py, iter).unwrap();
189
190    // call the lambda and get a python side Series wrapper
191    match lambda.call1(py, (wrapped_s,)) {
192        Ok(pyobj) => pyobj,
193        Err(e) => panic!("python function failed: {}", e.value(py)),
194    }
195}
196
197pub fn map_mul(
198    pyexpr: &[PyExpr],
199    py: Python<'_>,
200    lambda: PyObject,
201    output_type: Option<Wrap<DataType>>,
202    map_groups: bool,
203    returns_scalar: bool,
204) -> PyExpr {
205    // get the pypolars module
206    // do the import outside of the function to prevent import side effects in a hot loop.
207    let pypolars = polars(py).clone_ref(py);
208
209    let function = move |s: &mut [Column]| {
210        Python::with_gil(|py| {
211            // this is a python Series
212            let out = call_lambda_with_columns_slice(py, s, &lambda, &pypolars);
213
214            // we return an error, because that will become a null value polars lazy apply list
215            if map_groups && out.is_none(py) {
216                return Ok(None);
217            }
218
219            Ok(Some(out.to_series(py, &pypolars, "")?.into_column()))
220        })
221    };
222
223    let exprs = pyexpr.iter().map(|pe| pe.clone().inner).collect::<Vec<_>>();
224
225    let output_map = GetOutput::map_field(move |fld| {
226        Ok(match output_type {
227            Some(ref dt) => Field::new(fld.name().clone(), dt.0.clone()),
228            None => fld.clone(),
229        })
230    });
231    if map_groups {
232        polars::lazy::dsl::apply_multiple(function, exprs, output_map, returns_scalar).into()
233    } else {
234        polars::lazy::dsl::map_multiple(function, exprs, output_map).into()
235    }
236}