fixed assumed_centered? true tests

lib/scholar/covariance/ledoit_wolf.ex

@@ -93,14 +93,15 @@ defmodule Scholar.Covariance.LedoitWolf do
 
       iex> key = Nx.Random.key(0)
       iex> {x, _new_key} = Nx.Random.multivariate_normal(key, Nx.tensor([0.0, 0.0, 0.0]), Nx.tensor([[3.0, 2.0, 1.0], [1.0, 2.0, 3.0], [1.3, 1.0, 2.2]]), shape: {10}, type: :f32)
+      iex> {x, _} = Scholar.Covariance.Utils.center(x)
       iex> cov = Scholar.Covariance.LedoitWolf.fit(x, assume_centered?: true)
       iex> cov.covariance
       #Nx.Tensor<
         f32[3][3]
         [
-          [3.8574986457824707, 2.2048025131225586, 2.1504499912261963],
-          [2.2048025131225586, 2.4572863578796387, 1.7215262651443481],
-          [2.1504499912261963, 1.7215262651443481, 2.154898166656494]
+          [2.5945029258728027, 1.5078359842300415, 1.1623677015304565],
+          [1.5078359842300415, 2.106797456741333, 1.1812156438827515],
+          [1.1623677015304565, 1.1812156438827515, 1.4606266021728516]
         ]
       >
   """
@@ -148,7 +149,7 @@ defmodule Scholar.Covariance.LedoitWolf do
 
   defnp ledoit_wolf_shrinkage_complex(x) do
     {num_samples, num_features} = Nx.shape(x)
-    emp_cov = Scholar.Covariance.Utils.empirical_covariance(x)
+    emp_cov = Nx.covariance(x)
 
     emp_cov_trace = Scholar.Covariance.Utils.trace(emp_cov)
     mu = Nx.sum(emp_cov_trace) / num_features

lib/scholar/covariance/shrunk_covariance.ex

@@ -97,7 +97,7 @@ defmodule Scholar.Covariance.ShrunkCovariance do
     {x, location} = Scholar.Covariance.Utils.center(x, opts[:assume_centered?])
 
     covariance =
-      Scholar.Covariance.Utils.empirical_covariance(x)
+      Nx.covariance(x)
       |> shrunk_covariance(shrinkage)
 
     %__MODULE__{

lib/scholar/covariance/utils.ex

@@ -3,34 +3,18 @@ defmodule Scholar.Covariance.Utils do
   import Nx.Defn
   require Nx
 
-  defn center(x, assume_centered \\ false) do
+  defn center(x, assume_centered \\ Nx.tensor(0)) do
     x =
       case Nx.shape(x) do
         {_} -> Nx.new_axis(x, 1)
         _ -> x
       end
 
-    location =
-      if assume_centered do
-        0
-      else
-        Nx.mean(x, axes: [0])
-      end
+    location = Nx.select(assume_centered == Nx.tensor(1), 0, Nx.mean(x, axes: [0]))
 
     {x - location, location}
   end
 
-  defn empirical_covariance(x) do
-    n = Nx.axis_size(x, 0)
-
-    covariance = Nx.dot(x, [0], x, [0]) / n
-
-    case Nx.shape(covariance) do
-      {} -> Nx.reshape(covariance, {1, 1})
-      _ -> covariance
-    end
-  end
-
   defn trace(x) do
     x
     |> Nx.take_diagonal()

test/scholar/covariance/ledoit_wolf_test.exs

@@ -52,14 +52,16 @@ defmodule Scholar.Covariance.LedoitWolfTest do
         type: :f32
       )
 
+    {x, _} = Scholar.Covariance.Utils.center(x)
+
     model = LedoitWolf.fit(x, assume_centered?: true)
 
     assert_all_close(
       model.covariance,
       Nx.tensor([
-        [1.852303147315979, 0.0, 0.0],
-        [0.0, 1.852303147315979, 0.0],
-        [0.0, 0.0, 1.852303147315979]
+        [1.439786434173584, -0.0, 0.0],
+        [-0.0, 1.439786434173584, 0.0],
+        [0.0, 0.0, 1.439786434173584]
       ]),
       atol: 1.0e-3
     )

test/scholar/covariance/shrunk_covariance_test.exs

@@ -50,14 +50,16 @@ defmodule Scholar.Covariance.ShrunkCovarianceTest do
         type: :f32
       )
 
+    {x, _} = Scholar.Covariance.Utils.center(x)
+
     model = ShrunkCovariance.fit(x, assume_centered?: true)
 
     assert_all_close(
       model.covariance,
       Nx.tensor([
-        [3.0643274784088135, 0.27685147523880005, 0.4822050631046295],
-        [0.27685147523880005, 1.5171942710876465, 0.03596973791718483],
-        [0.4822050631046295, 0.03596973791718483, 0.975387692451477]
+        [2.0949244499206543, -0.13400490581989288, 0.5413897037506104],
+        [-0.13400490581989288, 1.2940725088119507, 0.0621684193611145],
+        [0.5413897037506104, 0.0621684193611145, 0.9303621053695679]
       ]),
       atol: 1.0e-3
     )