Add metric of HammingLoss.

src/TensorFlowNET.Core/APIs/tf.math.cs

@@ -24,6 +24,8 @@ namespace Tensorflow
             public Tensor argmax(Tensor input, Axis axis = null, string name = null, int? dimension = null, TF_DataType output_type = TF_DataType.TF_INT64)
                 => gen_math_ops.arg_max(input, axis, name: name, output_type: output_type);
 
+            public Tensor count_nonzero(Tensor input, Axis? axis = null, bool? keepdims = null, TF_DataType dtype = TF_DataType.TF_INT64, string name = null)
+                => math_ops.count_nonzero_v2(input, axis: axis, keepdims: keepdims ?? false, dtype: dtype);
             public Tensor log(Tensor x, string name = null)
                 => gen_math_ops.log(x, name);
 

src/TensorFlowNET.Core/Keras/Metrics/IMetricsApi.cs

@@ -91,7 +91,20 @@ public interface IMetricsApi
         float? threshold = null,
         string name = "fbeta_score",
         TF_DataType dtype = TF_DataType.TF_FLOAT);
-    
+
+    /// <summary>
+    /// Computes hamming loss.
+    /// </summary>
+    /// <param name="mode">multiclass or multilabel</param>
+    /// <param name="threshold"></param>
+    /// <param name="name"></param>
+    /// <param name="dtype"></param>
+    /// <returns></returns>
+    IMetricFunc HammingLoss(string mode,
+        float? threshold = null,
+        string name = "hamming_loss",
+        TF_DataType dtype = TF_DataType.TF_FLOAT);
+
     /// <summary>
     /// Computes how often targets are in the top K predictions.
     /// </summary>

src/TensorFlowNET.Core/Operations/math_ops.cs

@@ -821,6 +821,18 @@ namespace Tensorflow
                     .SetAttributes(new { adj_x, adj_y }));
             });
 
+        public static Tensor count_nonzero_v2(Tensor input,
+                Axis? axis,
+                bool keepdims = false,
+                string name = null,
+                TF_DataType dtype = TF_DataType.TF_INT64)
+            => tf_with(ops.name_scope(name, "count_nonzero", input), scope =>
+               {
+                   name = scope;
+                   var zero = array_ops.zeros(Shape.Scalar, dtype: input.dtype);
+                   return reduce_sum(cast(gen_math_ops.not_equal(input, zero), dtype), axis: axis, keepdims: keepdims);
+               });
+
         public static Tensor bincount(Tensor arr, Tensor weights = null,
              Tensor minlength = null,
              Tensor maxlength = null,

src/TensorFlowNET.Core/Tensorflow.Binding.csproj

@@ -109,7 +109,7 @@ https://tensorflownet.readthedocs.io</Description>
   <ItemGroup>
     <PackageReference Include="MethodBoundaryAspect.Fody" Version="2.0.148" />
     <PackageReference Include="Newtonsoft.Json" Version="13.0.2" />
-    <PackageReference Include="Protobuf.Text" Version="0.6.0" />
+    <PackageReference Include="Protobuf.Text" Version="0.6.1" />
     <PackageReference Include="Serilog.Sinks.Console" Version="4.1.0" />
   </ItemGroup>
 </Project>

src/TensorFlowNET.Keras/Metrics/HammingLoss.cs

@@ -0,0 +1,15 @@
+namespace Tensorflow.Keras.Metrics;
+
+public class HammingLoss : MeanMetricWrapper
+{
+    public HammingLoss(string mode, 
+        NDArray threshold = null,
+        string name = "hamming_loss",
+        TF_DataType dtype = TF_DataType.TF_FLOAT)
+        : base((yt, yp) => metrics_utils.hamming_loss_fn(yt, yp, threshold, mode), 
+            name: name, 
+            dtype: dtype)
+    {
+        _dtype = dtype;
+    }
+}

src/TensorFlowNET.Keras/Metrics/MetricsApi.cs

@@ -92,6 +92,9 @@
         public IMetricFunc FBetaScore(int num_classes, string? average = null, float beta = 0.1F, float? threshold = null, string name = "fbeta_score", TF_DataType dtype = TF_DataType.TF_FLOAT)
             => new FBetaScore(num_classes, average: average,beta: beta, threshold: threshold, name: name, dtype: dtype);
 
+        public IMetricFunc HammingLoss(string mode, float? threshold = null, string name = "hamming_loss", TF_DataType dtype = TF_DataType.TF_FLOAT)
+            => new HammingLoss(mode, threshold: threshold, name: name, dtype: dtype);
+
         public IMetricFunc TopKCategoricalAccuracy(int k = 5, string name = "top_k_categorical_accuracy", TF_DataType dtype = TF_DataType.TF_FLOAT)
             => new TopKCategoricalAccuracy(k: k, name: name, dtype: dtype);
 

src/TensorFlowNET.Keras/Metrics/metrics_utils.cs

@@ -24,6 +24,36 @@ public class metrics_utils
         return tf.reduce_sum(y_true * y_pred, axis: axis ?? -1);
     }
 
+    public static Tensor hamming_loss_fn(Tensor y_true, Tensor y_pred, Tensor threshold, string mode)
+    {
+        if (threshold == null)
+        {
+            threshold = tf.reduce_max(y_pred, axis: -1, keepdims: true);
+            // make sure [0, 0, 0] doesn't become [1, 1, 1]
+            // Use abs(x) > eps, instead of x != 0 to check for zero
+            y_pred = tf.logical_and(y_pred >= threshold, tf.abs(y_pred) > 1e-12);
+        }
+        else
+        {
+            y_pred = y_pred > threshold;
+        }
+
+
+        y_true = tf.cast(y_true, tf.int32);
+        y_pred = tf.cast(y_pred, tf.int32);
+
+        if (mode == "multiclass")
+        {
+            var nonzero = tf.cast(tf.math.count_nonzero(y_true * y_pred, axis: -1), tf.float32);
+            return 1.0 - nonzero;
+        }
+        else
+        {
+            var nonzero = tf.cast(tf.math.count_nonzero(y_true - y_pred, axis: -1), tf.float32);
+            return nonzero / y_true.shape[-1];
+        }
+    }
+    
     /// <summary>
     /// Creates float Tensor, 1.0 for label-prediction match, 0.0 for mismatch.
     /// </summary>

test/TensorFlowNET.Keras.UnitTest/Metrics/MetricsTest.cs

@@ -142,6 +142,51 @@ public class MetricsTest : EagerModeTestBase
         Assert.AreEqual(r, new[] { 0.3846154f, 0.90909094f, 0.8333334f });
     }
 
+    /// <summary>
+    /// https://www.tensorflow.org/addons/api_docs/python/tfa/metrics/HammingLoss
+    /// </summary>
+    [TestMethod]
+    public void HammingLoss()
+    {
+        // multi-class hamming loss
+        var y_true = np.array(new[,] 
+        { 
+            { 1, 0, 0, 0 }, 
+            { 0, 0, 1, 0 }, 
+            { 0, 0, 0, 1 }, 
+            { 0, 1, 0, 0 } 
+        });
+        var y_pred = np.array(new[,] 
+        { 
+            { 0.8f, 0.1f, 0.1f, 0.0f }, 
+            { 0.2f, 0.0f, 0.8f, 0.0f }, 
+            { 0.05f, 0.05f, 0.1f, 0.8f },
+            { 1.0f, 0.0f, 0.0f, 0.0f }
+        });
+        var m = tf.keras.metrics.HammingLoss(mode: "multiclass", threshold: 0.6f);
+        m.update_state(y_true, y_pred);
+        var r = m.result().numpy();
+        Assert.AreEqual(r, 0.25f);
+
+        // multi-label hamming loss
+        y_true = np.array(new[,]
+        {
+            { 1, 0, 1, 0 },
+            { 0, 1, 0, 1 },
+            { 0, 0, 0, 1 }
+        });
+        y_pred = np.array(new[,]
+        {
+            { 0.82f, 0.5f, 0.9f, 0.0f },
+            { 0f, 1f, 0.4f, 0.98f },
+            { 0.89f, 0.79f, 0f, 0.3f }
+        });
+        m = tf.keras.metrics.HammingLoss(mode: "multilabel", threshold: 0.8f);
+        m.update_state(y_true, y_pred);
+        r = m.result().numpy();
+        Assert.AreEqual(r, 0.16666667f);
+    }
+
     /// <summary>
     /// https://www.tensorflow.org/api_docs/python/tf/keras/metrics/TopKCategoricalAccuracy
     /// </summary>