Update to latest Nx

lib/scholar/cluster/affinity_propagation.ex

@@ -103,18 +103,18 @@ defmodule Scholar.Cluster.AffinityPropagation do
       iex> x = Nx.tensor([[12,5,78,2], [9,3,81,-2], [-1,3,6,1], [1,-2,5,2]])
       iex> Scholar.Cluster.AffinityPropagation.fit(x, key: key)
       %Scholar.Cluster.AffinityPropagation{
-        labels: Nx.tensor([0, 0, 2, 2]),
-        cluster_centers_indices: Nx.tensor([0, -1, 2, -1]),
-        cluster_centers: Nx.tensor(
+        labels: Nx.s32([0, 0, 2, 2]),
+        cluster_centers_indices: Nx.s32([0, -1, 2, -1]),
+        cluster_centers: Nx.f32(
           [
             [12.0, 5.0, 78.0, 2.0],
             [:infinity, :infinity, :infinity, :infinity],
             [-1.0, 3.0, 6.0, 1.0],
             [:infinity, :infinity, :infinity, :infinity]
           ]
         ),
-        num_clusters: Nx.tensor(2, type: :u64),
-        iterations: Nx.tensor(22, type: :s64)
+        num_clusters: Nx.u32(2),
+        iterations: Nx.u32(22)
       }
   """
   deftransform fit(data, opts \\ []) do
@@ -125,7 +125,7 @@ defmodule Scholar.Cluster.AffinityPropagation do
 
   defnp fit_n(data, key, opts) do
     data = to_float(data)
-    iterations = opts[:iterations]
+    iterations = opts[:iterations] |> Nx.as_type(:u32)
     damping_factor = opts[:damping_factor]
     converge_after = opts[:converge_after]
     n = Nx.axis_size(data, 0)
@@ -146,7 +146,7 @@ defmodule Scholar.Cluster.AffinityPropagation do
     stop = Nx.u8(0)
 
     {{a, r, it}, _} =
-      while {{a = zero_n, r = zero_n, i = 0}, {s, range, stop, e}},
+      while {{a = zero_n, r = zero_n, i = Nx.u32(0)}, {s, range, stop, e}},
             i < iterations and not stop do
         temp = a + s
         indices = Nx.argmax(temp, axis: 1)
@@ -204,7 +204,7 @@ defmodule Scholar.Cluster.AffinityPropagation do
 
         indices =
           Nx.select(mask, Nx.iota(Nx.shape(diagonals)), -1)
-          |> Nx.as_type({:s, 64})
+          |> Nx.as_type(:s32)
 
         cluster_centers =
           Nx.select(
@@ -216,15 +216,14 @@ defmodule Scholar.Cluster.AffinityPropagation do
         labels =
           Nx.broadcast(mask, Nx.shape(s))
           |> Nx.select(s, Nx.Constants.neg_infinity(Nx.type(s)))
-          |> Nx.argmax(axis: 1)
-          |> Nx.as_type({:s, 64})
+          |> Nx.argmax(axis: 1, type: :s32)
 
         labels = Nx.select(mask, Nx.iota(Nx.shape(labels)), labels)
 
         {cluster_centers, indices, labels}
       else
-        {Nx.tensor(-1, type: Nx.type(data)), Nx.broadcast(Nx.tensor(-1, type: :s64), {n}),
-         Nx.broadcast(Nx.tensor(-1, type: :s64), {n})}
+        {Nx.tensor(-1, type: Nx.type(data)), Nx.broadcast(Nx.tensor(-1, type: :s32), {n}),
+         Nx.broadcast(Nx.tensor(-1, type: :s32), {n})}
       end
 
     %__MODULE__{
@@ -262,16 +261,16 @@ defmodule Scholar.Cluster.AffinityPropagation do
       iex> model = Scholar.Cluster.AffinityPropagation.fit(x, key: key)
       iex> Scholar.Cluster.AffinityPropagation.prune(model)
       %Scholar.Cluster.AffinityPropagation{
-        labels: Nx.tensor([0, 0, 1, 1]),
-        cluster_centers_indices: Nx.tensor([0, 2]),
+        labels: Nx.s32([0, 0, 1, 1]),
+        cluster_centers_indices: Nx.s32([0, 2]),
         cluster_centers: Nx.tensor(
           [
             [12.0, 5.0, 78.0, 2.0],
             [-1.0, 3.0, 6.0, 1.0]
           ]
         ),
-        num_clusters: Nx.tensor(2, type: :u64),
-        iterations: Nx.tensor(22, type: :s64)
+        num_clusters: Nx.u32(2),
+        iterations: Nx.u32(22)
       }
   """
   def prune(
@@ -293,7 +292,9 @@ defmodule Scholar.Cluster.AffinityPropagation do
       end)
 
     mapping = Map.new(mapping)
-    cluster_centers_indices = Nx.tensor(Enum.reverse(indices))
+
+    cluster_centers_indices =
+      Nx.tensor(Enum.reverse(indices), type: Nx.type(cluster_centers_indices))
 
     %__MODULE__{
       model
@@ -314,7 +315,7 @@ defmodule Scholar.Cluster.AffinityPropagation do
       iex> model = Scholar.Cluster.AffinityPropagation.prune(model)
       iex> Scholar.Cluster.AffinityPropagation.predict(model, Nx.tensor([[10,3,50,6], [8,3,8,2]]))
       #Nx.Tensor<
-        s64[2]
+        s32[2]
         [0, 1]
       >
   """

lib/scholar/cluster/optics.ex

@@ -75,32 +75,32 @@ defmodule Scholar.Cluster.OPTICS do
       iex> x = Nx.tensor([[1, 2], [2, 5], [3, 6], [8, 7], [8, 8], [7, 3]])
       iex> Scholar.Cluster.OPTICS.fit(x, min_samples: 2).labels
       #Nx.Tensor<
-        s64[6]
+        s32[6]
         [-1, -1, -1, -1, -1, -1]
       >
       iex> Scholar.Cluster.OPTICS.fit(x, eps: 4.5, min_samples: 2).labels
       #Nx.Tensor<
-        s64[6]
+        s32[6]
         [0, 0, 0, 1, 1, 1]
       >
       iex> Scholar.Cluster.OPTICS.fit(x, eps: 2, min_samples: 2).labels
       #Nx.Tensor<
-        s64[6]
+        s32[6]
         [-1, 0, 0, 1, 1, -1]
       >
       iex> Scholar.Cluster.OPTICS.fit(x, eps: 2, min_samples: 2, algorithm: :kd_tree, metric: {:minkowski, 1}).labels
       #Nx.Tensor<
-        s64[6]
+        s32[6]
         [-1, 0, 0, 1, 1, -1]
       >
       iex> Scholar.Cluster.OPTICS.fit(x, eps: 1, min_samples: 2).labels
       #Nx.Tensor<
-        s64[6]
+        s32[6]
         [-1, -1, -1, 0, 0, -1]
       >
       iex> Scholar.Cluster.OPTICS.fit(x, eps: 4.5, min_samples: 3).labels
       #Nx.Tensor<
-        s64[6]
+        s32[6]
         [0, 0, 0, 1, 1, -1]
       >
   """

lib/scholar/linear/isotonic_regression.ex

@@ -158,7 +158,6 @@ defmodule Scholar.Linear.IsotonicRegression do
     {sample_weights, opts} = Keyword.pop(opts, :sample_weights, 1.0)
     x_type = to_float_type(x)
     x = to_float(x)
-
     y = to_float(y)
 
     sample_weights =
@@ -202,7 +201,7 @@ defmodule Scholar.Linear.IsotonicRegression do
   @doc """
   Makes predictions with the given `model` on input `x` and interpolating `function`.
 
-  Output predictions have shape `{n_samples}` when train target is shaped either `{n_samples}` or `{n_samples, 1}`.  
+  Output predictions have shape `{n_samples}` when train target is shaped either `{n_samples}` or `{n_samples, 1}`.
   Otherwise, predictions match train target shape.
 
   ## Examples
@@ -443,19 +442,19 @@ defmodule Scholar.Linear.IsotonicRegression do
   end
 
   defnp contiguous_isotonic_regression(y, sample_weights, max_size, increasing) do
-    y_size = if increasing, do: max_size, else: Nx.axis_size(y, 0) - 1
+    y_size = if(increasing, do: max_size, else: Nx.axis_size(y, 0) - 1) |> Nx.as_type(:u32)
     y = if increasing, do: y, else: Nx.reverse(y)
     sample_weights = if increasing, do: sample_weights, else: Nx.reverse(sample_weights)
 
-    target = Nx.iota({Nx.axis_size(y, 0)}, type: :s64)
+    target = Nx.iota({Nx.axis_size(y, 0)}, type: :u32)
     type_wy = Nx.Type.merge(Nx.type(y), Nx.type(sample_weights))
-    i = if increasing, do: 0, else: Nx.axis_size(y, 0) - 1 - max_size
+    i = if(increasing, do: 0, else: Nx.axis_size(y, 0) - 1 - max_size) |> Nx.as_type(:u32)
 
     {{y, target}, _} =
       while {{y, target},
              {i, sample_weights, sum_w = Nx.tensor(0, type: Nx.type(sample_weights)),
-              sum_wy = Nx.tensor(0, type: type_wy), prev_y = Nx.tensor(0, type: type_wy), _k = 0,
-              terminating_flag = 0, y_size}},
+              sum_wy = Nx.tensor(0, type: type_wy), prev_y = Nx.tensor(0, type: type_wy),
+              _k = Nx.u32(0), terminating_flag = Nx.u8(0), y_size}},
             i < y_size + 1 and not terminating_flag do
         k = target[i] + 1
 
@@ -509,12 +508,12 @@ defmodule Scholar.Linear.IsotonicRegression do
         end
       end
 
-    i = if increasing, do: 0, else: Nx.axis_size(y, 0) - 1 - max_size
+    i = if(increasing, do: 0, else: Nx.axis_size(y, 0) - 1 - max_size) |> Nx.as_type(:u32)
 
     {y, _} =
-      while {y, {target, i, _k = 0, max_size}}, i < max_size + 1 do
+      while {y, {target, i, _k = Nx.u32(0), max_size}}, i < max_size + 1 do
         k = target[i] + 1
-        indices = Nx.iota({Nx.axis_size(y, 0)})
+        indices = Nx.iota({Nx.axis_size(y, 0)}, type: :u32)
         in_range? = Nx.logical_and(i + 1 <= indices, indices < k)
         y = Nx.select(in_range?, y[i], y)
         i = k

lib/scholar/linear/logistic_regression.ex

@@ -211,7 +211,7 @@ defmodule Scholar.Linear.LogisticRegression do
   @doc """
   Makes predictions with the given `model` on inputs `x`.
 
-  Output predictions have shape `{n_samples}` when train target is shaped either `{n_samples}` or `{n_samples, 1}`.        
+  Output predictions have shape `{n_samples}` when train target is shaped either `{n_samples}` or `{n_samples, 1}`.
 
   ## Examples
 
@@ -220,7 +220,7 @@ defmodule Scholar.Linear.LogisticRegression do
       iex> model = Scholar.Linear.LogisticRegression.fit(x, y, num_classes: 2)
       iex> Scholar.Linear.LogisticRegression.predict(model, Nx.tensor([[-3.0, 5.0]]))
       #Nx.Tensor<
-        s64[1]
+        s32[1]
         [1]
       >
   """

lib/scholar/linear/polynomial_regression.ex

@@ -110,8 +110,8 @@ defmodule Scholar.Linear.PolynomialRegression do
   @doc """
   Makes predictions with the given `model` on input `x`.
 
-  Output predictions have shape `{n_samples}` when train target is shaped either `{n_samples}` or `{n_samples, 1}`.    
-  Otherwise, predictions match train target shape.  
+  Output predictions have shape `{n_samples}` when train target is shaped either `{n_samples}` or `{n_samples, 1}`.
+  Otherwise, predictions match train target shape.
 
   ## Examples
 
@@ -148,7 +148,7 @@ defmodule Scholar.Linear.PolynomialRegression do
       iex> x = Nx.tensor([[2]])
       iex> Scholar.Linear.PolynomialRegression.transform(x, degree: 5, fit_intercept?: false)
       #Nx.Tensor<
-        s64[1][5]
+        s32[1][5]
         [
           [2, 4, 8, 16, 32]
         ]
@@ -157,7 +157,7 @@ defmodule Scholar.Linear.PolynomialRegression do
       iex> x = Nx.tensor([[2, 3]])
       iex> Scholar.Linear.PolynomialRegression.transform(x)
       #Nx.Tensor<
-        s64[1][6]
+        s32[1][6]
         [
           [1, 2, 3, 4, 6, 9]
         ]
@@ -166,7 +166,7 @@ defmodule Scholar.Linear.PolynomialRegression do
       iex> x = Nx.iota({3, 2})
       iex> Scholar.Linear.PolynomialRegression.transform(x, fit_intercept?: false)
       #Nx.Tensor<
-        s64[3][5]
+        s32[3][5]
         [
           [0, 1, 0, 0, 1],
           [2, 3, 4, 6, 9],

lib/scholar/linear/svm.ex

@@ -259,7 +259,7 @@ defmodule Scholar.Linear.SVM do
       iex> model = Scholar.Linear.SVM.fit(x, y, num_classes: 2)
       iex> Scholar.Linear.SVM.predict(model, Nx.tensor([[-3.0, 5.0]]))
       #Nx.Tensor<
-        s64[1]
+        s32[1]
         [1]
       >
   """

lib/scholar/manifold/trimap.ex

@@ -153,14 +153,14 @@ defmodule Scholar.Manifold.Trimap do
 
     # binsearch which checks if the elements of tensor1 are in tensor2
     {is_in, _} =
-      while {is_in, {tensor1, tensor2, prev = Nx.s64(-1), i = Nx.s64(0)}}, i < Nx.size(tensor1) do
+      while {is_in, {tensor1, tensor2, prev = Nx.s64(-1), i = Nx.u32(0)}}, i < Nx.size(tensor1) do
         if i > 0 and prev == tensor1[i] do
           is_in = Nx.indexed_put(is_in, Nx.new_axis(i, 0), is_in[i - 1])
           {is_in, {tensor1, tensor2, prev, i + 1}}
         else
           {found?, _} =
             while {stop = Nx.u8(0),
-                   {tensor1, tensor2, left = Nx.s64(0), right = Nx.size(tensor2) - 1, i}},
+                   {tensor1, tensor2, left = Nx.s64(0), right = Nx.s64(Nx.size(tensor2) - 1), i}},
                   left <= right and not stop do
               mid = div(left + right, 2)
 
@@ -188,13 +188,19 @@ defmodule Scholar.Manifold.Trimap do
     final_samples = Nx.broadcast(Nx.s64(0), shape)
 
     {final_samples, key, _, _} =
-      while {final_samples, key, rejects, i = Nx.s64(0)}, i < elem(shape, 0) do
-        {samples, key} = Nx.Random.randint(key, 0, opts[:maxval], shape: {elem(shape, 1)})
+      while {final_samples, key, rejects, i = Nx.u32(0)}, i < elem(shape, 0) do
+        # TODO: See if we can relax the samples to u32
+        {samples, key} =
+          Nx.Random.randint(key, 0, opts[:maxval], shape: {elem(shape, 1)}, type: :s64)
+
         discard = in1d(samples, rejects[i])
 
         {samples, key, _, _, _} =
           while {samples, key, discard, rejects, i}, Nx.any(discard) do
-            {new_samples, key} = Nx.Random.randint(key, 0, opts[:maxval], shape: {elem(shape, 1)})
+            # TODO: See if we can relax the samples to u32
+            {new_samples, key} =
+              Nx.Random.randint(key, 0, opts[:maxval], shape: {elem(shape, 1)}, type: :s64)
+
             discard = in1d(new_samples, rejects[i]) or in1d(new_samples, samples)
             samples = Nx.select(discard, samples, new_samples)
             {samples, key, in1d(samples, rejects[i]), rejects, i}
@@ -552,7 +558,7 @@ defmodule Scholar.Manifold.Trimap do
     gain = Nx.broadcast(Nx.tensor(1.0, type: to_float_type(embeddings)), Nx.shape(embeddings))
 
     {embeddings, _} =
-      while {embeddings, {vel, gain, lr, triplets, weights, i = Nx.s64(0)}},
+      while {embeddings, {vel, gain, lr, triplets, weights, i = Nx.u32(0)}},
             i < opts[:num_iters] do
         gamma = if i < @switch_iter, do: @init_momentum, else: @final_momentum