Merge coefficient types on linear interpolation, closes #318

Author: josevalim

lib/scholar/interpolation/linear.ex

@@ -131,16 +131,15 @@ defmodule Scholar.Interpolation.Linear do
 
   defnp predict_n(%__MODULE__{x: x, coefficients: coefficients} = _model, target_x, opts) do
     shape = Nx.shape(target_x)
-
     target_x = Nx.flatten(target_x)
-
     indices = Nx.argsort(target_x)
 
     left_bound = x[0]
     right_bound = x[-1]
 
     target_x = Nx.sort(target_x)
-    res = Nx.broadcast(Nx.tensor(0, type: to_float_type(target_x)), {Nx.axis_size(target_x, 0)})
+    type = Nx.Type.merge(to_float_type(target_x), coefficients.type)
+    res = Nx.broadcast(Nx.tensor(0, type: type), {Nx.axis_size(target_x, 0)})
 
     # while with smaller than left_bound
     {{res, i}, _} =

test/scholar/interpolation/linear_test.exs

@@ -51,5 +51,16 @@ defmodule Scholar.Interpolation.LinearTest do
       assert Linear.predict(model, Nx.tensor([[[-0.5], [0.5], [1.5], [2.5], [3.5]]])) ==
                Nx.tensor([[[0.0], [0.0], [0.5], [3], [7]]])
     end
+
+    test "with different types" do
+      x_s = Nx.tensor([1, 2, 3], type: :u64)
+      y_s = Nx.tensor([1.0, 2.0, 3.0], type: :f64)
+      target = Nx.tensor([1, 2], type: :u64)
+
+      assert x_s
+             |> Scholar.Interpolation.Linear.fit(y_s)
+             |> Scholar.Interpolation.Linear.predict(target) ==
+               Nx.tensor([1.0, 2.0], type: :f64)
+    end
   end
 end