From 7d0d2715d735be335a7ed10ab9614d44b3a68f80 Mon Sep 17 00:00:00 2001
From: haiping008 <haiping008@gmail.com>
Date: Wed, 20 Mar 2019 15:31:32 -0500
Subject: [PATCH] fix reduce_sum when axis is not null

---
 src/TensorFlowNET.Core/APIs/tf.math.cs            | 8 ++++++--
 src/TensorFlowNET.Core/Operations/gen_math_ops.cs | 7 +++++++
 src/TensorFlowNET.Core/Operations/math_ops.py.cs  | 8 +++++++-
 test/TensorFlowNET.Examples/LogisticRegression.cs | 9 ++++++---
 4 files changed, 26 insertions(+), 6 deletions(-)

diff --git a/src/TensorFlowNET.Core/APIs/tf.math.cs b/src/TensorFlowNET.Core/APIs/tf.math.cs
index e01da00c..0039e0d2 100644
--- a/src/TensorFlowNET.Core/APIs/tf.math.cs
+++ b/src/TensorFlowNET.Core/APIs/tf.math.cs
@@ -36,8 +36,12 @@ namespace Tensorflow
         /// <param name="input"></param>
         /// <param name="axis"></param>
         /// <returns></returns>
-        public static Tensor reduce_sum(Tensor input, int[] axis = null, int? reduction_indices = null) 
-            => math_ops.reduce_sum(input);
+        public static Tensor reduce_sum(Tensor input, int? axis = null, int? reduction_indices = null)
+        {
+            if(!axis.HasValue && reduction_indices.HasValue)
+                return math_ops.reduce_sum(input, reduction_indices.Value);
+            return math_ops.reduce_sum(input);
+        }
 
         public static Tensor reduce_mean(Tensor input_tensor, int[] axis = null, bool keepdims = false, string name = null, int? reduction_indices = null)
             => math_ops.reduce_mean(input_tensor, axis: axis, keepdims: keepdims, name: name, reduction_indices: reduction_indices);
diff --git a/src/TensorFlowNET.Core/Operations/gen_math_ops.cs b/src/TensorFlowNET.Core/Operations/gen_math_ops.cs
index 26513051..1ccb0e96 100644
--- a/src/TensorFlowNET.Core/Operations/gen_math_ops.cs
+++ b/src/TensorFlowNET.Core/Operations/gen_math_ops.cs
@@ -207,6 +207,13 @@ namespace Tensorflow
             return _op.outputs[0];
         }
 
+        public static Tensor sum(Tensor input, int axis, bool keep_dims = false, string name = null)
+        {
+            var _op = _op_def_lib._apply_op_helper("Sum", name, args: new { input, reduction_indices = axis, keep_dims });
+
+            return _op.outputs[0];
+        }
+
         /// <summary>
         /// Creates a sequence of numbers.
         /// </summary>
diff --git a/src/TensorFlowNET.Core/Operations/math_ops.py.cs b/src/TensorFlowNET.Core/Operations/math_ops.py.cs
index fd2a0644..3c6ea594 100644
--- a/src/TensorFlowNET.Core/Operations/math_ops.py.cs
+++ b/src/TensorFlowNET.Core/Operations/math_ops.py.cs
@@ -209,6 +209,12 @@ namespace Tensorflow
             return _may_reduce_to_scalar(keepdims, m);
         }
 
+        public static Tensor reduce_sum(Tensor input_tensor, int axis, bool keepdims = false)
+        {
+            var m = gen_math_ops.sum(input_tensor, axis);
+            return _may_reduce_to_scalar(keepdims, m);
+        }
+
         private static Tensor _may_reduce_to_scalar(bool keepdims, Tensor output)
         {
             output.shape = new long[0];
@@ -233,7 +239,7 @@ namespace Tensorflow
                 return range(0, rank, 1);
             }
         }
-        
+
         private static Tensor _ReductionDims(Tensor x, int[] axis)
         {
             if (axis != null)
diff --git a/test/TensorFlowNET.Examples/LogisticRegression.cs b/test/TensorFlowNET.Examples/LogisticRegression.cs
index 71760404..6cee6d8b 100644
--- a/test/TensorFlowNET.Examples/LogisticRegression.cs
+++ b/test/TensorFlowNET.Examples/LogisticRegression.cs
@@ -26,7 +26,7 @@ namespace TensorFlowNET.Examples
 
         private void PrepareData()
         {
-           var mnist = MnistDataSet.read_data_sets("logistic_regression", one_hot: true);
+           //var mnist = MnistDataSet.read_data_sets("logistic_regression", one_hot: true);
 
             // tf Graph Input
             var x = tf.placeholder(tf.float32, new TensorShape(-1, 784)); // mnist data image of shape 28*28=784
@@ -40,8 +40,11 @@ namespace TensorFlowNET.Examples
             var pred = tf.nn.softmax(tf.matmul(x, W) + b); // Softmax
 
             // Minimize error using cross entropy
-            var sum = -tf.reduce_sum(y * tf.log(pred), reduction_indices: 1);
-            var cost = tf.reduce_mean(sum);
+            var log = tf.log(pred);
+            var mul = y * log;
+            var sum = tf.reduce_sum(mul, reduction_indices: 1);
+            var neg = -sum;
+            var cost = tf.reduce_mean(neg);
 
             // Gradient Descent
             var optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(cost);