Add mode (#453)

flake.nix

@@ -29,9 +29,8 @@
           buildInputs = with pkgs; [
             act
             clang
-            beam.packages.erlangR25.elixir_1_14
-            erlangR25
-            gdb
+            elixir
+            erlangR26
             libiconv
             openssl
             pkg-config
@@ -40,6 +39,7 @@
             cmake
           ] ++ lib.optionals stdenv.isDarwin [
             darwin.apple_sdk.frameworks.Security
+            darwin.apple_sdk.frameworks.SystemConfiguration
           ];
           shellHook = ''
             mkdir -p .nix-mix

lib/explorer/backend/lazy_series.ex

@@ -99,6 +99,7 @@ defmodule Explorer.Backend.LazySeries do
     argmax: 1,
     mean: 1,
     median: 1,
+    mode: 1,
     n_distinct: 1,
     variance: 1,
     standard_deviation: 1,
@@ -149,6 +150,7 @@ defmodule Explorer.Backend.LazySeries do
     :argmax,
     :mean,
     :median,
+    :mode,
     :variance,
     :standard_deviation,
     :count,

lib/explorer/backend/series.ex

@@ -82,6 +82,7 @@ defmodule Explorer.Backend.Series do
   @callback argmax(s) :: number() | non_finite() | lazy_s() | nil
   @callback mean(s) :: float() | non_finite() | lazy_s() | nil
   @callback median(s) :: float() | non_finite() | lazy_s() | nil
+  @callback mode(s) :: s | lazy_s()
   @callback variance(s) :: float() | non_finite() | lazy_s() | nil
   @callback standard_deviation(s) :: float() | non_finite() | lazy_s() | nil
   @callback quantile(s, float()) ::

lib/explorer/polars_backend/expression.ex

@@ -51,6 +51,7 @@ defmodule Explorer.PolarsBackend.Expression do
     mean: 1,
     median: 1,
     min: 1,
+    mode: 1,
     n_distinct: 1,
     nil_count: 1,
     not_equal: 2,

lib/explorer/polars_backend/native.ex

@@ -300,6 +300,7 @@ defmodule Explorer.PolarsBackend.Native do
   def s_median(_s), do: err()
   def s_product(_s), do: err()
   def s_min(_s), do: err()
+  def s_mode(_s), do: err()
   def s_multiply(_s, _other), do: err()
   def s_n_chunks(_s), do: err()
   def s_n_distinct(_s), do: err()

lib/explorer/polars_backend/series.ex

@@ -198,6 +198,9 @@ defmodule Explorer.PolarsBackend.Series do
   @impl true
   def median(series), do: Shared.apply_series(series, :s_median)
 
+  @impl true
+  def mode(series), do: Shared.apply_series(series, :s_mode)
+
   @impl true
   def variance(series), do: Shared.apply_series(series, :s_variance)
 

lib/explorer/series.ex

@@ -2164,6 +2164,46 @@ defmodule Explorer.Series do
 
   def mean(%Series{dtype: dtype}), do: dtype_error("mean/1", dtype, [:integer, :float])
 
+  @doc """
+  Gets the most common value(s) of the series.
+
+  This function will return multiple values when there's a tie.
+
+  ## Supported dtypes
+
+  All except `:list`.
+
+  ## Examples
+
+      iex> s = Explorer.Series.from_list([1, 2, 2, nil])
+      iex> Explorer.Series.mode(s)
+      #Explorer.Series<
+        Polars[1]
+        integer [2]
+      >
+
+      iex> s = Explorer.Series.from_list(["a", "b", "b", "c"])
+      iex> Explorer.Series.mode(s)
+      #Explorer.Series<
+        Polars[1]
+        string ["b"]
+      >
+
+      s = Explorer.Series.from_list([1.0, 2.0, 2.0, 3.0, 3.0])
+      Explorer.Series.mode(s)
+      #Explorer.Series<
+        Polars[2]
+        float [2.0, 3.0]
+      >
+  """
+  @doc type: :aggregation
+  @spec mode(series :: Series.t()) :: Series.t() | nil
+  def mode(%Series{dtype: {:list, _} = dtype}),
+    do: dtype_error("mode/1", dtype, Shared.non_list_types())
+
+  def mode(%Series{} = series),
+    do: Shared.apply_impl(series, :mode)
+
   @doc """
   Gets the median value of the series.
 

lib/explorer/shared.ex

@@ -9,7 +9,16 @@ defmodule Explorer.Shared do
   within lists inside.
   """
   def dtypes do
-    non_list_dtypes = [
+    non_list_dtypes = non_list_types()
+    list_dtypes = for dtype <- non_list_dtypes, do: {:list, dtype}
+    non_list_dtypes ++ list_dtypes
+  end
+
+  @doc """
+  Non-list dtypes.
+  """
+  def non_list_types,
+    do: [
       :binary,
       :boolean,
       :category,
@@ -26,11 +35,6 @@ defmodule Explorer.Shared do
       {:duration, :nanosecond}
     ]
 
-    list_dtypes = for dtype <- non_list_dtypes, do: {:list, dtype}
-
-    non_list_dtypes ++ list_dtypes
-  end
-
   @doc """
   Supported datetime dtypes.
   """

native/explorer/Cargo.toml

@@ -21,8 +21,13 @@ thiserror = "1"
 smartstring = "1"
 
 # Deps necessary for cloud features.
-tokio = { version = "1.33", default-features = false, features = ["rt"], optional = true }
-tokio-util = { version = "0.7", default-features = false, features = ["io", "io-util"], optional = true }
+tokio = { version = "1.33", default-features = false, features = [
+  "rt",
+], optional = true }
+tokio-util = { version = "0.7", default-features = false, features = [
+  "io",
+  "io-util",
+], optional = true }
 object_store = { version = "0.7", default-features = false, optional = true }
 
 # MiMalloc won´t compile on Windows with the GCC compiler.
@@ -52,6 +57,7 @@ features = [
   "lazy",
   "lazy_regex",
   "log",
+  "mode",
   "parquet",
   "performant",
   "pivot",
@@ -71,15 +77,12 @@ features = [
   "product",
   "peaks",
   "moment",
-  "rank"
+  "rank",
 ]
 
 [dependencies.polars-ops]
 version = "0.34"
-features = [
-  "ewma",
-  "cum_agg"
-]
+features = ["ewma", "cum_agg"]
 
 [dependencies.polars-algo]
 version = "0.34"

native/explorer/src/expressions.rs

@@ -468,6 +468,13 @@ pub fn expr_median(expr: ExExpr) -> ExExpr {
     ExExpr::new(expr.median())
 }
 
+#[rustler::nif]
+pub fn expr_mode(expr: ExExpr) -> ExExpr {
+    let expr = expr.clone_inner();
+
+    ExExpr::new(expr.mode())
+}
+
 #[rustler::nif]
 pub fn expr_product(expr: ExExpr) -> ExExpr {
     let expr = expr.clone_inner();

native/explorer/src/lib.rs

@@ -229,6 +229,7 @@ rustler::init!(
         expr_mean,
         expr_median,
         expr_min,
+        expr_mode,
         expr_n_distinct,
         expr_nil_count,
         expr_quantile,
@@ -366,6 +367,7 @@ rustler::init!(
         s_correlation,
         s_covariance,
         s_min,
+        s_mode,
         s_multiply,
         s_n_distinct,
         s_name,

native/explorer/src/series.rs

@@ -937,6 +937,14 @@ pub fn s_median(env: Env, s: ExSeries) -> Result<Term, ExplorerError> {
     }
 }
 
+#[rustler::nif(schedule = "DirtyCpu")]
+pub fn s_mode(s: ExSeries) -> Result<ExSeries, ExplorerError> {
+    match mode::mode(&s) {
+        Ok(s) => Ok(ExSeries::new(s)),
+        Err(e) => Err(e.into()),
+    }
+}
+
 #[rustler::nif(schedule = "DirtyCpu")]
 pub fn s_product(s: ExSeries) -> Result<ExSeries, ExplorerError> {
     match s.dtype() {

test/explorer/data_frame_test.exs

@@ -3526,6 +3526,18 @@ defmodule Explorer.DataFrameTest do
              }
     end
 
+    test "mode/1" do
+      df =
+        Datasets.iris()
+        |> DF.group_by(:species)
+        |> DF.summarise(petal_width_mode: mode(petal_width))
+
+      assert DF.to_columns(df) == %{
+               "petal_width_mode" => [[0.2], [1.3], [1.8]],
+               "species" => ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
+             }
+    end
+
     test "argmax/1 and argmin/1" do
       df =
         DF.new(