Merge pull request #1360 from NiklasGustafsson/bugs

Address issue #1359

Author: NiklasGustafsson

RELEASENOTES.md

@@ -4,6 +4,12 @@ Releases, starting with 9/2/2021, are listed with the most recent release at the
 
 # NuGet Version 0.102.6
 
+__Bug Fixes__:
+
+#1359 torch.nn.functional.l1_loss computes a criterion with the MSE, not the MAE.<br/>
+
+# NuGet Version 0.102.6
+
 __Breaking Changes__:
 
 When creating a tensor from a 1-D array, and passing in a shape, there is now an ambiguity between the IList and Memory overloads of `torch.tensor()`. The ambiguity is resolved by removing the `dimensions` argument if it is redundant, or by an explicit cast to IList if it is not.

build/BranchInfo.props

@@ -2,8 +2,8 @@
   <PropertyGroup>
     <MajorVersion>0</MajorVersion>
     <MinorVersion>102</MinorVersion>
-    <PatchVersion>7</PatchVersion>
-    <PreviousPackageVersion>0.102.6</PreviousPackageVersion>
+    <PatchVersion>8</PatchVersion>
+    <PreviousPackageVersion>0.102.7</PreviousPackageVersion>
   </PropertyGroup>
 
 </Project>

src/Native/LibTorchSharp/THSLoss.cpp

@@ -94,10 +94,10 @@ Tensor THSNN_kl_div_loss(const Tensor input, const Tensor target, const int64_t
 Tensor THSNN_l1_loss(const Tensor input, const Tensor target, const int64_t reduction)
 {
     CATCH_RETURN_Tensor(
-        auto opts = torch::nn::functional::MSELossFuncOptions();
+        auto opts = torch::nn::functional::L1LossFuncOptions();
         ApplyReduction(opts, reduction);
 
-        res = ResultTensor(torch::nn::functional::mse_loss(*input, *target, opts));
+        res = ResultTensor(torch::nn::functional::l1_loss(*input, *target, opts));
     )
 }
 

test/TorchSharpTest/NN.cs

@@ -1409,15 +1409,16 @@ public void TestCrossEntropyLossF()
         public void TestL1Loss()
         {
             foreach (var device in TestUtils.AvailableDevices()) {
-                using (Tensor input = torch.rand(new long[] { 5, 2 }, device: device))
-                using (Tensor target = torch.rand(new long[] { 5, 2 }, device: device)) {
+                using (Tensor input = torch.arange(10, dtype:float32, device: device))
+                using (Tensor target = torch.zeros(10, dtype:float32, device: device)) {
                     var outTensor = L1Loss().call(input, target);
                     Assert.Equal(device.type, outTensor.device_type);
                     var values = outTensor.data<float>().ToArray();
                     Assert.Multiple(
                     () => Assert.Empty(outTensor.shape),
                     () => Assert.Single(values),
-                    () => Assert.False(float.IsNaN(values[0]))
+                    () => Assert.False(float.IsNaN(values[0])),
+                    () => Assert.Equal(4.5, values[0])
                     );
                 }
             }
@@ -1427,15 +1428,16 @@ public void TestL1Loss()
         public void TestL1LossF()
         {
             foreach (var device in TestUtils.AvailableDevices()) {
-                using (Tensor input = torch.rand(new long[] { 5, 2 }, device: device))
-                using (Tensor target = torch.rand(new long[] { 5, 2 }, device: device)) {
+                using (Tensor input = torch.arange(10, dtype:float32, device: device))
+                using (Tensor target = torch.zeros(10, dtype:float32, device: device)) {
                     var outTensor = l1_loss(input, target);
                     Assert.Equal(device.type, outTensor.device_type);
                     var values = outTensor.data<float>().ToArray();
                     Assert.Multiple(
                     () => Assert.Empty(outTensor.shape),
                     () => Assert.Single(values),
-                    () => Assert.False(float.IsNaN(values[0]))
+                    () => Assert.False(float.IsNaN(values[0])),
+                    () => Assert.Equal(4.5, values[0])
                     );
                 }
             }