Support {:list, numeric_type} for Series.frequencies/1 (#788)

It is related to #787,
but does not fix the issue because Polars can only get frequencies of
`{:list, numeric_dtype}`.

lib/explorer/polars_backend/series.ex

@@ -14,16 +14,8 @@ defmodule Explorer.PolarsBackend.Series do
   defguardp is_non_finite(n) when n in [:nan, :infinity, :neg_infinity]
   defguardp is_numeric(n) when is_number(n) or is_non_finite(n)
 
-  @integer_types [
-    {:s, 8},
-    {:s, 16},
-    {:s, 32},
-    {:s, 64},
-    {:u, 8},
-    {:u, 16},
-    {:u, 32},
-    {:u, 64}
-  ]
+  @integer_types Explorer.Shared.integer_types()
+  @numeric_types Explorer.Shared.numeric_types()
 
   # Conversion
 
@@ -456,7 +448,13 @@ defmodule Explorer.PolarsBackend.Series do
   def n_distinct(series), do: Shared.apply_series(series, :s_n_distinct)
 
   @impl true
-  def frequencies(series) do
+  def frequencies(%Series{dtype: {:list, inner_dtype} = dtype})
+      when inner_dtype not in @numeric_types do
+    raise ArgumentError,
+          "frequencies/1 only works with series of lists of numeric types, but #{Explorer.Shared.dtype_to_string(dtype)} was given"
+  end
+
+  def frequencies(%Series{} = series) do
     Shared.apply(:s_frequencies, [series.data])
     |> Shared.create_dataframe()
     |> DataFrame.rename(["values", "counts"])

native/explorer/Cargo.toml

@@ -51,6 +51,7 @@ features = [
   "decompress-fast",
   "describe",
   "dtype-full",
+  "group_by_list",
   "ipc",
   "ipc_streaming",
   "lazy",

test/explorer/series_test.exs

@@ -5168,4 +5168,58 @@ defmodule Explorer.SeriesTest do
              ]
     end
   end
+
+  describe "frequencies/1" do
+    test "integer" do
+      s = Series.from_list([1, 2, 3, 1, 3, 4, 1, 5, 6, 1, 2])
+
+      df = Series.frequencies(s)
+
+      assert Series.dtype(df[:values]) == {:s, 64}
+      assert Series.dtype(df[:counts]) == {:s, 64}
+
+      assert Explorer.DataFrame.to_columns(df, atom_keys: true) == %{
+               values: [1, 2, 3, 4, 5, 6],
+               counts: [4, 2, 2, 1, 1, 1]
+             }
+    end
+
+    test "string" do
+      s = Series.from_list(["a", "a", "b", "c", "c", "c"])
+
+      df = Series.frequencies(s)
+
+      assert Series.dtype(df[:values]) == :string
+      assert Series.dtype(df[:counts]) == {:s, 64}
+
+      assert Explorer.DataFrame.to_columns(df, atom_keys: true) == %{
+               values: ["c", "a", "b"],
+               counts: [3, 2, 1]
+             }
+    end
+
+    test "list of integer" do
+      s = Series.from_list([[1, 2], [3, 1, 3], [4, 1], [5, 6], [1, 2], [4, 1]])
+
+      df = Series.frequencies(s)
+
+      assert Series.dtype(df[:values]) == {:list, {:s, 64}}
+      assert Series.dtype(df[:counts]) == {:s, 64}
+
+      assert Explorer.DataFrame.to_columns(df, atom_keys: true) == %{
+               values: [[1, 2], [4, 1], [3, 1, 3], [5, 6]],
+               counts: [2, 2, 1, 1]
+             }
+    end
+
+    test "list of list of string" do
+      s = Series.from_list([["a"], ["a", "b"], ["c"], ["c"], ["c"]])
+
+      assert_raise ArgumentError,
+                   "frequencies/1 only works with series of lists of numeric types, but list[string] was given",
+                   fn ->
+                     Series.frequencies(s)
+                   end
+    end
+  end
 end