polars_core/frame/row/
dataframe.rs

1use super::*;
2
3impl DataFrame {
4    /// Get a row from a [`DataFrame`]. Use of this is discouraged as it will likely be slow.
5    pub fn get_row(&self, idx: usize) -> PolarsResult<Row> {
6        let values = self
7            .materialized_column_iter()
8            .map(|s| s.get(idx))
9            .collect::<PolarsResult<Vec<_>>>()?;
10        Ok(Row(values))
11    }
12
13    /// Amortize allocations by reusing a row.
14    /// The caller is responsible to make sure that the row has at least the capacity for the number
15    /// of columns in the [`DataFrame`]
16    pub fn get_row_amortized<'a>(&'a self, idx: usize, row: &mut Row<'a>) -> PolarsResult<()> {
17        for (s, any_val) in self.materialized_column_iter().zip(&mut row.0) {
18            *any_val = s.get(idx)?;
19        }
20        Ok(())
21    }
22
23    /// Amortize allocations by reusing a row.
24    /// The caller is responsible to make sure that the row has at least the capacity for the number
25    /// of columns in the [`DataFrame`]
26    ///
27    /// # Safety
28    /// Does not do any bounds checking.
29    #[inline]
30    pub unsafe fn get_row_amortized_unchecked<'a>(&'a self, idx: usize, row: &mut Row<'a>) {
31        self.materialized_column_iter()
32            .zip(&mut row.0)
33            .for_each(|(s, any_val)| {
34                *any_val = s.get_unchecked(idx);
35            });
36    }
37
38    /// Create a new [`DataFrame`] from rows.
39    ///
40    /// This should only be used when you have row wise data, as this is a lot slower
41    /// than creating the [`Series`] in a columnar fashion
42    pub fn from_rows_and_schema(rows: &[Row], schema: &Schema) -> PolarsResult<Self> {
43        Self::from_rows_iter_and_schema(rows.iter(), schema)
44    }
45
46    /// Create a new [`DataFrame`] from an iterator over rows.
47    ///
48    /// This should only be used when you have row wise data, as this is a lot slower
49    /// than creating the [`Series`] in a columnar fashion.
50    pub fn from_rows_iter_and_schema<'a, I>(mut rows: I, schema: &Schema) -> PolarsResult<Self>
51    where
52        I: Iterator<Item = &'a Row<'a>>,
53    {
54        if schema.is_empty() {
55            let height = rows.count();
56            let columns = Vec::new();
57            return Ok(unsafe { DataFrame::new_no_checks(height, columns) });
58        }
59
60        let capacity = rows.size_hint().0;
61
62        let mut buffers: Vec<_> = schema
63            .iter_values()
64            .map(|dtype| {
65                let buf: AnyValueBuffer = (dtype, capacity).into();
66                buf
67            })
68            .collect();
69
70        let mut expected_len = 0;
71        rows.try_for_each::<_, PolarsResult<()>>(|row| {
72            expected_len += 1;
73            for (value, buf) in row.0.iter().zip(&mut buffers) {
74                buf.add_fallible(value)?
75            }
76            Ok(())
77        })?;
78        let v = buffers
79            .into_iter()
80            .zip(schema.iter_names())
81            .map(|(b, name)| {
82                let mut c = b.into_series().into_column();
83                // if the schema adds a column not in the rows, we
84                // fill it with nulls
85                if c.is_empty() {
86                    Column::full_null(name.clone(), expected_len, c.dtype())
87                } else {
88                    c.rename(name.clone());
89                    c
90                }
91            })
92            .collect();
93        DataFrame::new(v)
94    }
95
96    /// Create a new [`DataFrame`] from an iterator over rows. This should only be used when you have row wise data,
97    /// as this is a lot slower than creating the [`Series`] in a columnar fashion
98    pub fn try_from_rows_iter_and_schema<'a, I>(mut rows: I, schema: &Schema) -> PolarsResult<Self>
99    where
100        I: Iterator<Item = PolarsResult<&'a Row<'a>>>,
101    {
102        let capacity = rows.size_hint().0;
103
104        let mut buffers: Vec<_> = schema
105            .iter_values()
106            .map(|dtype| {
107                let buf: AnyValueBuffer = (dtype, capacity).into();
108                buf
109            })
110            .collect();
111
112        let mut expected_len = 0;
113        rows.try_for_each::<_, PolarsResult<()>>(|row| {
114            expected_len += 1;
115            for (value, buf) in row?.0.iter().zip(&mut buffers) {
116                buf.add_fallible(value)?
117            }
118            Ok(())
119        })?;
120        let v = buffers
121            .into_iter()
122            .zip(schema.iter_names())
123            .map(|(b, name)| {
124                let mut c = b.into_series().into_column();
125                // if the schema adds a column not in the rows, we
126                // fill it with nulls
127                if c.is_empty() {
128                    Column::full_null(name.clone(), expected_len, c.dtype())
129                } else {
130                    c.rename(name.clone());
131                    c
132                }
133            })
134            .collect();
135        DataFrame::new(v)
136    }
137
138    /// Create a new [`DataFrame`] from rows. This should only be used when you have row wise data,
139    /// as this is a lot slower than creating the [`Series`] in a columnar fashion
140    pub fn from_rows(rows: &[Row]) -> PolarsResult<Self> {
141        let schema = rows_to_schema_first_non_null(rows, Some(50))?;
142        let has_nulls = schema
143            .iter_values()
144            .any(|dtype| matches!(dtype, DataType::Null));
145        polars_ensure!(
146            !has_nulls, ComputeError: "unable to infer row types because of null values"
147        );
148        Self::from_rows_and_schema(rows, &schema)
149    }
150}