[MachineLearning.Inference] Change internal rank limit and default value

This patch changes internal rank limit in MachineLeanring.Inference.
To use higher rank limit, ml_tensors_info_create_extended is needed.
Also, default dimension value is changed to 0 from 1.

Signed-off-by: Yelin Jeong <yelini.jeong@samsung.com>

diff --git a/src/Tizen.MachineLearning.Inference/Interop/Interop.Nnstreamer.cs b/src/Tizen.MachineLearning.Inference/Interop/Interop.Nnstreamer.cs
index 1d6d5df..b8c0c4e 100755 (executable)
--- a/src/Tizen.MachineLearning.Inference/Interop/Interop.Nnstreamer.cs
+++ b/src/Tizen.MachineLearning.Inference/Interop/Interop.Nnstreamer.cs
@@ -235,6 +235,10 @@ internal static partial class Interop
         [DllImport(Libraries.MlCommon, EntryPoint = "ml_tensors_info_create", CallingConvention = CallingConvention.Cdecl)]
         internal static extern NNStreamerError CreateTensorsInfo(out IntPtr info);
 
+        /* int ml_tensors_info_create_extended (ml_tensors_info_h *info) */
+        [DllImport(Libraries.MlCommon, EntryPoint = "ml_tensors_info_create_extended", CallingConvention = CallingConvention.Cdecl)]
+        internal static extern NNStreamerError CreateTensorsInfoExtended(out IntPtr info);
+
         /* int ml_tensors_info_destroy (ml_tensors_info_h info) */
         [DllImport(Libraries.MlCommon, EntryPoint = "ml_tensors_info_destroy", CallingConvention = CallingConvention.Cdecl)]
         internal static extern NNStreamerError DestroyTensorsInfo(IntPtr info);

diff --git a/src/Tizen.MachineLearning.Inference/Tizen.MachineLearning.Inference/Commons.cs b/src/Tizen.MachineLearning.Inference/Tizen.MachineLearning.Inference/Commons.cs
index fb81b3c..951ca33 100755 (executable)
--- a/src/Tizen.MachineLearning.Inference/Tizen.MachineLearning.Inference/Commons.cs
+++ b/src/Tizen.MachineLearning.Inference/Tizen.MachineLearning.Inference/Commons.cs
@@ -267,7 +267,7 @@ namespace Tizen.MachineLearning.Inference
         /// <summary>
         /// The maximum rank that NNStreamer supports with Tizen APIs.
         /// </summary>
-        internal const int RankLimit = 4;
+        internal const int RankLimit = 16;
 
         /// <summary>
         /// The maximum number of other/tensor instances that other/tensors may have.

diff --git a/src/Tizen.MachineLearning.Inference/Tizen.MachineLearning.Inference/TensorsInfo.cs b/src/Tizen.MachineLearning.Inference/Tizen.MachineLearning.Inference/TensorsInfo.cs
index 2df0d49..094c044 100755 (executable)
--- a/src/Tizen.MachineLearning.Inference/Tizen.MachineLearning.Inference/TensorsInfo.cs
+++ b/src/Tizen.MachineLearning.Inference/Tizen.MachineLearning.Inference/TensorsInfo.cs
@@ -95,7 +95,9 @@ namespace Tizen.MachineLearning.Inference
             if (idx >= Tensor.SizeLimit) {
                 throw new IndexOutOfRangeException("Max size of the tensors is " + Tensor.SizeLimit);
             }
-            _infoList.Add(new TensorInfo(name, type, dimension));
+
+            int[] dim = ConvertDimension(dimension);
+            _infoList.Add(new TensorInfo(name, type, dim));
 
             if (_handle != IntPtr.Zero)
             {
@@ -104,7 +106,7 @@ namespace Tizen.MachineLearning.Inference
                 ret = Interop.Util.SetTensorsCount(_handle, _infoList.Count);
                 NNStreamer.CheckException(ret, "Failed to set the number of tensors");
 
-                UpdateInfoHandle(_handle, idx, name, type, dimension);
+                UpdateInfoHandle(_handle, idx, name, type, dim);
             }
         }
 
@@ -204,12 +206,15 @@ namespace Tizen.MachineLearning.Inference
             NNStreamer.CheckNNStreamerSupport();
 
             CheckIndexBoundary(idx);
-            _infoList[idx].SetDimension(dimension);
+
+            int[] dim = ConvertDimension(dimension);
+
+            _infoList[idx].SetDimension(dim);
 
             if (_handle != IntPtr.Zero)
             {
                 NNStreamerError ret = NNStreamerError.None;
-                ret = Interop.Util.SetTensorDimension(_handle, idx, dimension);
+                ret = Interop.Util.SetTensorDimension(_handle, idx, dim);
                 NNStreamer.CheckException(ret, "unable to set the dimension of tensor: " + idx.ToString());
             }
         }
@@ -412,7 +417,7 @@ namespace Tizen.MachineLearning.Inference
             NNStreamer.CheckException(ret, "number of Tensor in TensorsInfo is invalid: " + _infoList.Count);
 
             /* Create TensorsInfo object */
-            ret = Interop.Util.CreateTensorsInfo(out ret_handle);
+            ret = Interop.Util.CreateTensorsInfoExtended(out ret_handle);
             NNStreamer.CheckException(ret, "fail to create TensorsInfo object");
 
             /* Set the number of tensors */
@@ -469,6 +474,29 @@ namespace Tizen.MachineLearning.Inference
             _disposed = true;
         }
 
+        private static int[] ConvertDimension(int[] dimension)
+        {
+            if (dimension == null) {
+                throw NNStreamerExceptionFactory.CreateException(NNStreamerError.InvalidParameter, "The dimension is null, it should be a valid array.");
+            }
+
+            if (dimension.Length > Tensor.RankLimit) {
+                throw new IndexOutOfRangeException("Max rank limit is " + Tensor.RankLimit);
+            }
+
+            int[] dim = new int[Tensor.RankLimit];
+            int i;
+
+            for (i = 0 ; i < dimension.Length ; i++) {
+                dim[i] = dimension[i];
+            }
+            for (; i < Tensor.RankLimit ; i++) {
+                dim[i] = 0;
+            }
+
+            return dim;
+        }
+
         private void UpdateInfoHandle(IntPtr handle, int idx, string name, TensorType type, int[] dimension)
         {
             if (handle != IntPtr.Zero)
@@ -513,11 +541,11 @@ namespace Tizen.MachineLearning.Inference
             public void SetDimension(int[] dimension)
             {
                 if (dimension == null) {
-                    throw new ArgumentException("Max size of the tensor rank is" + Tensor.RankLimit);
+                    throw new ArgumentException("The dimension is null, it should be a valid array.");
                 }
 
-                if (dimension.Length > Tensor.RankLimit) {
-                    throw new ArgumentException("Max size of the tensor rank is" + Tensor.RankLimit);
+                if (dimension.Length != Tensor.RankLimit) {
+                    throw new ArgumentException("The length of the dimension should be " + Tensor.RankLimit);
                 }
                 Dimension = (int[])dimension.Clone();
             }
@@ -555,6 +583,8 @@ namespace Tizen.MachineLearning.Inference
                 }
                 for (int i = 0; i < Tensor.RankLimit; ++i)
                 {
+                    if (Dimension[i] == 0)
+                        break;
                     size *= Dimension[i];
                 }
                 return size;

diff --git a/src/Tizen.MachineLearning.Train/Tizen.MachineLearning.Train/Model.cs b/src/Tizen.MachineLearning.Train/Tizen.MachineLearning.Train/Model.cs
index 08672eb..d92636e 100644 (file)
--- a/src/Tizen.MachineLearning.Train/Tizen.MachineLearning.Train/Model.cs
+++ b/src/Tizen.MachineLearning.Train/Tizen.MachineLearning.Train/Model.cs
@@ -339,7 +339,7 @@ namespace Tizen.MachineLearning.Train
 
         internal static TensorsInfo CreateTensorsInfoFormHandle(IntPtr handle)
         {
-            const int RankLimit = 4;
+            const int RankLimit = 16;
             NNTrainerError ret = NNTrainerError.None;
             int count;