Add convergence check to AffinityPropagation (#195)

lib/scholar/cluster/affinity_propagation.ex

@@ -15,14 +15,16 @@ defmodule Scholar.Cluster.AffinityPropagation do
              :cluster_centers_indices,
              :affinity_matrix,
              :cluster_centers,
-             :num_clusters
+             :num_clusters,
+             :iterations
            ]}
   defstruct [
     :labels,
     :cluster_centers_indices,
     :affinity_matrix,
     :cluster_centers,
-    :num_clusters
+    :num_clusters,
+    :iterations
   ]
 
   @opts_schema [
@@ -56,6 +58,14 @@ defmodule Scholar.Cluster.AffinityPropagation do
       doc: ~S"""
       If `true`, the learning loop is unrolled.
       """
+    ],
+    converge_after: [
+      type: :pos_integer,
+      default: 15,
+      doc: ~S"""
+      Number of iterations with no change in the number of estimated clusters
+      that stops the convergence.
+      """
     ]
   ]
 
@@ -101,7 +111,8 @@ defmodule Scholar.Cluster.AffinityPropagation do
             [1.0, -2.0, 5.0, 2.0]
           ]
         ),
-        num_clusters: Nx.tensor(2, type: :u64)
+        num_clusters: Nx.tensor(2, type: :u64),
+        iterations: Nx.tensor(18, type: :s64)
       }
   """
   deftransform fit(data, opts \\ []) do
@@ -116,7 +127,8 @@ defmodule Scholar.Cluster.AffinityPropagation do
     iterations = opts[:iterations]
     damping_factor = opts[:damping_factor]
     self_preference = opts[:self_preference]
-    data = to_float(data)
+    converge_after = opts[:converge_after]
+    num_samples = Nx.axis_size(data, 0)
 
     {initial_a, initial_r, s, affinity_matrix} =
       initialize_matrices(data, self_preference: self_preference)
@@ -132,9 +144,12 @@ defmodule Scholar.Cluster.AffinityPropagation do
 
     range = Nx.iota({n})
 
-    {{a, r}, _} =
-      while {{a = initial_a, r = initial_r}, {s, range, i = 0}},
-            i < iterations do
+    e = Nx.broadcast(Nx.s64(0), {num_samples, converge_after})
+    stop = Nx.u8(0)
+
+    {{a, r, it}, _} =
+      while {{a = initial_a, r = initial_r, i = 0}, {s, range, stop, e}},
+            i < iterations and not stop do
         temp = a + s
         indices = Nx.argmax(temp, axis: 1)
         y = Nx.reduce_max(temp, axes: [1])
@@ -160,7 +175,24 @@ defmodule Scholar.Cluster.AffinityPropagation do
         temp = temp * (1 - damping_factor)
         a = a * damping_factor - temp
 
-        {{a, r}, {s, range, i + 1}}
+        curr_e = Nx.take_diagonal(a) + Nx.take_diagonal(r) > 0
+        curr_e_slice = Nx.reshape(curr_e, {:auto, 1})
+        e = Nx.put_slice(e, [0, Nx.remainder(i, converge_after)], curr_e_slice)
+        k = Nx.sum(curr_e, axes: [0])
+
+        stop =
+          if i >= converge_after do
+            se = Nx.sum(e, axes: [1])
+            unconverged = Nx.sum((se == 0) + (se == converge_after)) != num_samples
+
+            if (not unconverged and k > 0) or i == iterations do
+              Nx.u8(1)
+            else
+              stop
+            end
+          end
+
+        {{a, r, i + 1}, {s, range, stop, e}}
       end
 
     diagonals = Nx.take_diagonal(a) + Nx.take_diagonal(r) > 0
@@ -202,7 +234,8 @@ defmodule Scholar.Cluster.AffinityPropagation do
       cluster_centers_indices: cluster_centers_indices,
       cluster_centers: cluster_centers,
       labels: labels,
-      num_clusters: k
+      num_clusters: k,
+      iterations: it
     }
   end
 
@@ -233,7 +266,8 @@ defmodule Scholar.Cluster.AffinityPropagation do
             [1.0, -2.0, 5.0, 2.0]
           ]
         ),
-        num_clusters: Nx.tensor(2, type: :u64)
+        num_clusters: Nx.tensor(2, type: :u64),
+        iterations: Nx.tensor(18, type: :s64)
       }
   """
   def prune(
@@ -281,16 +315,6 @@ defmodule Scholar.Cluster.AffinityPropagation do
       >
   """
   defn predict(%__MODULE__{cluster_centers: cluster_centers} = _model, x) do
-    {num_clusters, num_features} = Nx.shape(cluster_centers)
-    {num_samples, _} = Nx.shape(x)
-    broadcast_shape = {num_samples, num_clusters, num_features}
-
-    Scholar.Metrics.Distance.euclidean(
-      Nx.new_axis(x, 1) |> Nx.broadcast(broadcast_shape),
-      Nx.new_axis(cluster_centers, 0) |> Nx.broadcast(broadcast_shape),
-      axes: [-1]
-    )
-
     dist = Scholar.Metrics.Distance.pairwise_euclidean(x, cluster_centers)
 
     Nx.select(Nx.is_nan(dist), Nx.Constants.infinity(Nx.type(dist)), dist)