feat: Add Series::{first, last, approx_n_unique} (#19093)

crates/polars-compute/Cargo.toml

@@ -26,4 +26,5 @@ version_check = { workspace = true }
 [features]
 nightly = []
 simd = ["arrow/simd"]
+approx_unique = []
 dtype-array = []

crates/polars-ops/src/series/ops/approx_algo/hyperloglogplus.rs → crates/polars-compute/src/hyperloglogplus.rs

@@ -9,7 +9,7 @@
 //! # Examples
 //!
 //! ```
-//!     # use polars_ops::prelude::*;
+//!     # use polars_compute::hyperloglogplus::*;
 //!     let mut hllp = HyperLogLog::new();
 //!     hllp.add(&12345);
 //!     hllp.add(&23456);

crates/polars-compute/src/lib.rs

@@ -13,6 +13,8 @@ pub mod bitwise;
 pub mod comparisons;
 pub mod filter;
 pub mod float_sum;
+#[cfg(feature = "approx_unique")]
+pub mod hyperloglogplus;
 pub mod if_then_else;
 pub mod min_max;
 pub mod size;

crates/polars-core/Cargo.toml

@@ -76,6 +76,7 @@ fmt_no_tty = ["comfy-table"]
 rows = []
 
 # operations
+approx_unique = ["polars-compute/approx_unique"]
 bitwise = ["algorithm_group_by"]
 zip_with = []
 round_series = []

crates/polars-core/src/chunked_array/ops/approx_n_unique.rs

@@ -0,0 +1,20 @@
+use std::hash::Hash;
+
+use polars_compute::hyperloglogplus::HyperLogLog;
+use polars_utils::total_ord::{ToTotalOrd, TotalEq, TotalHash};
+use polars_utils::IdxSize;
+
+use super::{ChunkApproxNUnique, ChunkedArray, PolarsDataType};
+
+impl<T> ChunkApproxNUnique for ChunkedArray<T>
+where
+    T: PolarsDataType,
+    for<'a> T::Physical<'a>: TotalHash + TotalEq + Copy + ToTotalOrd,
+    for<'a> <Option<T::Physical<'a>> as ToTotalOrd>::TotalOrdItem: Hash + Eq,
+{
+    fn approx_n_unique(&self) -> IdxSize {
+        let mut hllp = HyperLogLog::new();
+        self.iter().for_each(|item| hllp.add(&item.to_total_ord()));
+        hllp.count() as IdxSize
+    }
+}

crates/polars-core/src/chunked_array/ops/mod.rs

@@ -7,6 +7,8 @@ pub(crate) mod aggregate;
 pub(crate) mod any_value;
 pub(crate) mod append;
 mod apply;
+#[cfg(feature = "approx_unique")]
+mod approx_n_unique;
 pub mod arity;
 mod bit_repr;
 #[cfg(feature = "bitwise")]
@@ -375,6 +377,11 @@ pub trait ChunkUnique {
     }
 }
 
+#[cfg(feature = "approx_unique")]
+pub trait ChunkApproxNUnique {
+    fn approx_n_unique(&self) -> IdxSize;
+}
+
 /// Sort operations on `ChunkedArray`.
 pub trait ChunkSort<T: PolarsDataType> {
     #[allow(unused_variables)]

crates/polars-core/src/frame/column/mod.rs

@@ -1031,6 +1031,18 @@ impl Column {
             .into()),
         }
     }
+
+    #[cfg(feature = "approx_unique")]
+    pub fn approx_n_unique(&self) -> PolarsResult<IdxSize> {
+        match self {
+            Column::Series(s) => s.approx_n_unique(),
+            Column::Scalar(s) => {
+                // @NOTE: We do this for the error handling.
+                s.as_single_value_series().approx_n_unique()?;
+                Ok(1)
+            },
+        }
+    }
 }
 
 impl Default for Column {

crates/polars-core/src/series/implementations/binary.rs

@@ -214,6 +214,11 @@ impl SeriesTrait for SeriesWrap<BinaryChunked> {
         ChunkUnique::arg_unique(&self.0)
     }
 
+    #[cfg(feature = "approx_unique")]
+    fn approx_n_unique(&self) -> PolarsResult<IdxSize> {
+        Ok(ChunkApproxNUnique::approx_n_unique(&self.0))
+    }
+
     fn is_null(&self) -> BooleanChunked {
         self.0.is_null()
     }

crates/polars-core/src/series/implementations/boolean.rs

@@ -370,6 +370,11 @@ impl SeriesTrait for SeriesWrap<BooleanChunked> {
         ))
     }
 
+    #[cfg(feature = "approx_unique")]
+    fn approx_n_unique(&self) -> PolarsResult<IdxSize> {
+        Ok(ChunkApproxNUnique::approx_n_unique(&self.0))
+    }
+
     fn clone_inner(&self) -> Arc<dyn SeriesTrait> {
         Arc::new(SeriesWrap(Clone::clone(&self.0)))
     }

crates/polars-core/src/series/implementations/floats.rs

@@ -391,6 +391,11 @@ macro_rules! impl_dyn_series {
                 Ok(Scalar::new(dt, av))
             }
 
+            #[cfg(feature = "approx_unique")]
+            fn approx_n_unique(&self) -> PolarsResult<IdxSize> {
+                Ok(ChunkApproxNUnique::approx_n_unique(&self.0))
+            }
+
             fn clone_inner(&self) -> Arc<dyn SeriesTrait> {
                 Arc::new(SeriesWrap(Clone::clone(&self.0)))
             }

crates/polars-core/src/series/implementations/mod.rs

@@ -497,6 +497,11 @@ macro_rules! impl_dyn_series {
                 Ok(Scalar::new(dt, av))
             }
 
+            #[cfg(feature = "approx_unique")]
+            fn approx_n_unique(&self) -> PolarsResult<IdxSize> {
+                Ok(ChunkApproxNUnique::approx_n_unique(&self.0))
+            }
+
             fn clone_inner(&self) -> Arc<dyn SeriesTrait> {
                 Arc::new(SeriesWrap(Clone::clone(&self.0)))
             }

crates/polars-core/src/series/implementations/string.rs

@@ -252,6 +252,12 @@ impl SeriesTrait for SeriesWrap<StringChunked> {
     fn min_reduce(&self) -> PolarsResult<Scalar> {
         Ok(ChunkAggSeries::min_reduce(&self.0))
     }
+
+    #[cfg(feature = "approx_unique")]
+    fn approx_n_unique(&self) -> PolarsResult<IdxSize> {
+        Ok(ChunkApproxNUnique::approx_n_unique(&self.0))
+    }
+
     fn clone_inner(&self) -> Arc<dyn SeriesTrait> {
         Arc::new(SeriesWrap(Clone::clone(&self.0)))
     }

crates/polars-core/src/series/series_trait.rs

@@ -147,20 +147,23 @@ pub(crate) mod private {
         unsafe fn agg_list(&self, groups: &GroupsProxy) -> Series {
             Series::full_null(self._field().name().clone(), groups.len(), self._dtype())
         }
+
         /// # Safety
         ///
         /// Does no bounds checks, groups must be correct.
         #[cfg(feature = "bitwise")]
         unsafe fn agg_and(&self, groups: &GroupsProxy) -> Series {
             Series::full_null(self._field().name().clone(), groups.len(), self._dtype())
         }
+
         /// # Safety
         ///
         /// Does no bounds checks, groups must be correct.
         #[cfg(feature = "bitwise")]
         unsafe fn agg_or(&self, groups: &GroupsProxy) -> Series {
             Series::full_null(self._field().name().clone(), groups.len(), self._dtype())
         }
+
         /// # Safety
         ///
         /// Does no bounds checks, groups must be correct.
@@ -504,15 +507,45 @@ pub trait SeriesTrait:
     }
     /// Get the bitwise AND of the Series as a new Series of length 1,
     fn and_reduce(&self) -> PolarsResult<Scalar> {
-        polars_bail!(opq = sum, self._dtype());
+        polars_bail!(opq = and_reduce, self._dtype());
     }
     /// Get the bitwise OR of the Series as a new Series of length 1,
     fn or_reduce(&self) -> PolarsResult<Scalar> {
-        polars_bail!(opq = sum, self._dtype());
+        polars_bail!(opq = or_reduce, self._dtype());
     }
     /// Get the bitwise XOR of the Series as a new Series of length 1,
     fn xor_reduce(&self) -> PolarsResult<Scalar> {
-        polars_bail!(opq = sum, self._dtype());
+        polars_bail!(opq = xor_reduce, self._dtype());
+    }
+
+    /// Get the first element of the [`Series`] as a [`Scalar`]
+    ///
+    /// If the [`Series`] is empty, a [`Scalar`] with a [`AnyValue::Null`] is returned.
+    fn first(&self) -> Scalar {
+        let dt = self.dtype();
+        let av = self.get(0).map_or(AnyValue::Null, AnyValue::into_static);
+
+        Scalar::new(dt.clone(), av)
+    }
+
+    /// Get the last element of the [`Series`] as a [`Scalar`]
+    ///
+    /// If the [`Series`] is empty, a [`Scalar`] with a [`AnyValue::Null`] is returned.
+    fn last(&self) -> Scalar {
+        let dt = self.dtype();
+        let av = if self.len() == 0 {
+            AnyValue::Null
+        } else {
+            // SAFETY: len-1 < len if len != 0
+            unsafe { self.get_unchecked(self.len() - 1) }.into_static()
+        };
+
+        Scalar::new(dt.clone(), av)
+    }
+
+    #[cfg(feature = "approx_unique")]
+    fn approx_n_unique(&self) -> PolarsResult<IdxSize> {
+        polars_bail!(opq = approx_n_unique, self._dtype());
     }
 
     /// Clone inner ChunkedArray and wrap in a new Arc

crates/polars-ops/src/series/ops/mod.rs

@@ -1,9 +1,5 @@
 #[cfg(feature = "abs")]
 mod abs;
-#[cfg(feature = "approx_unique")]
-mod approx_algo;
-#[cfg(feature = "approx_unique")]
-mod approx_unique;
 mod arg_min_max;
 mod bitwise;
 #[cfg(feature = "business")]
@@ -67,10 +63,6 @@ mod various;
 
 #[cfg(feature = "abs")]
 pub use abs::*;
-#[cfg(feature = "approx_unique")]
-pub use approx_algo::*;
-#[cfg(feature = "approx_unique")]
-pub use approx_unique::*;
 pub use arg_min_max::ArgAgg;
 pub use bitwise::*;
 #[cfg(feature = "business")]

crates/polars-plan/Cargo.toml

@@ -107,7 +107,7 @@ extract_jsonpath = ["polars-ops/extract_jsonpath"]
 
 # operations
 bitwise = ["polars-core/bitwise", "polars-ops/bitwise"]
-approx_unique = ["polars-ops/approx_unique"]
+approx_unique = ["polars-ops/approx_unique", "polars-core/approx_unique"]
 is_in = ["polars-ops/is_in"]
 repeat_by = ["polars-ops/repeat_by"]
 round_series = ["polars-ops/round_series"]

crates/polars-plan/src/dsl/function_expr/dispatch.rs

@@ -6,7 +6,8 @@ pub(super) fn reverse(s: &Column) -> PolarsResult<Column> {
 
 #[cfg(feature = "approx_unique")]
 pub(super) fn approx_n_unique(s: &Column) -> PolarsResult<Column> {
-    polars_ops::prelude::approx_n_unique(s.as_materialized_series()).map(Column::from)
+    s.approx_n_unique()
+        .map(|v| Column::new_scalar(s.name().clone(), Scalar::new(IDX_DTYPE, v.into()), 1))
 }
 
 #[cfg(feature = "diff")]

crates/polars-python/Cargo.toml

@@ -163,6 +163,7 @@ hist = ["polars/hist"]
 find_many = ["polars/find_many"]
 new_streaming = ["polars-lazy/new_streaming"]
 bitwise = ["polars/bitwise"]
+approx_unique = ["polars/approx_unique"]
 
 dtype-i8 = []
 dtype-i16 = []
@@ -181,6 +182,7 @@ dtypes = [
 ]
 
 operations = [
+  "approx_unique",
   "array_any_all",
   "array_count",
   "bitwise",

crates/polars-python/src/expr/general.rs

@@ -418,6 +418,7 @@ impl PyExpr {
             .into()
     }
 
+    #[cfg(feature = "approx_unique")]
     fn approx_n_unique(&self) -> Self {
         self.inner.clone().approx_n_unique().into()
     }

crates/polars-python/src/series/aggregation.rs

@@ -146,6 +146,23 @@ impl PySeries {
         .into_py(py))
     }
 
+    fn first(&self, py: Python) -> PyObject {
+        Wrap(self.series.first().as_any_value()).into_py(py)
+    }
+
+    fn last(&self, py: Python) -> PyObject {
+        Wrap(self.series.last().as_any_value()).into_py(py)
+    }
+
+    #[cfg(feature = "approx_unique")]
+    fn approx_n_unique(&self, py: Python) -> PyResult<PyObject> {
+        Ok(self
+            .series
+            .approx_n_unique()
+            .map_err(PyPolarsErr::from)?
+            .into_py(py))
+    }
+
     #[cfg(feature = "bitwise")]
     fn bitwise_and(&self, py: Python) -> PyResult<PyObject> {
         Ok(Wrap(