Making TensorFlowTransform trainable.

src/Microsoft.ML.Legacy/CSharpApi.cs

@@ -15801,6 +15801,66 @@ public sealed partial class TensorFlowScorer : Microsoft.ML.Runtime.EntryPoints.
             /// </summary>
             public string[] OutputColumns { get; set; }
 
+            /// <summary>
+            /// Training labels.
+            /// </summary>
+            public string LabeLColumn { get; set; } = "Label";
+
+            /// <summary>
+            /// The name of the optimization operation in the TensorFlow graph.
+            /// </summary>
+            public string OptimizationOperation { get; set; }
+
+            /// <summary>
+            /// The name of the operation in the TensorFlow graph to compute training loss (Optional)
+            /// </summary>
+            public string LossOperation { get; set; }
+
+            /// <summary>
+            /// The name of the operation in the TensorFlow graph to compute performance metric during training (Optional)
+            /// </summary>
+            public string MetricOperation { get; set; }
+
+            /// <summary>
+            /// Number of samples to use for mini-batch training.
+            /// </summary>
+            public int BatchSize { get; set; } = 64;
+
+            /// <summary>
+            /// Number of training iterations.
+            /// </summary>
+            public int Epoch { get; set; } = 5;
+
+            /// <summary>
+            /// The name of the operation in the TensorFlow graph which sets optimizer learning rate (Optional).
+            /// </summary>
+            public string LearningRateOperation { get; set; }
+
+            /// <summary>
+            /// Learning rate to use during optimization.
+            /// </summary>
+            public float LearningRate { get; set; } = 0.01f;
+
+            /// <summary>
+            /// Shuffle data before each iteration.
+            /// </summary>
+            public bool Shuffle { get; set; } = true;
+
+            /// <summary>
+            /// Name of the input in TensorFlow graph that specifiy the location for saving/restoring models from disk.
+            /// </summary>
+            public string SaveLocationOperation { get; set; } = "save/Const";
+
+            /// <summary>
+            /// Name of the input in TensorFlow graph that specifiy the location for saving/restoring models from disk.
+            /// </summary>
+            public string SaveOperation { get; set; } = "save/control_dependency";
+
+            /// <summary>
+            /// Retrain TensorFlow model.
+            /// </summary>
+            public bool ReTrain { get; set; } = false;
+
             /// <summary>
             /// Input dataset
             /// </summary>

src/Microsoft.ML.TensorFlow/TensorFlow/Tensor.cs

@@ -11,6 +11,7 @@
 using System.Text;
 using size_t = System.UIntPtr;
 using TF_Tensor = System.IntPtr;
+using TF_Status = System.IntPtr;
 
 #pragma warning disable MSML_ParameterLocalVarName
 
@@ -74,6 +75,18 @@ internal partial class TFTensor : TFDisposableThreadSafe
         [DllImport(NativeBinding.TensorFlowLibrary)]
         private static extern unsafe TF_Tensor TF_NewTensor(TFDataType dataType, IntPtr zeroDims, int num_dims, IntPtr data, size_t len, Deallocator deallocator, IntPtr deallocator_arg);
 
+        // extern size_t TF_StringEncode (const char *src, size_t src_len, char *dst, size_t dst_len, TF_Status *status);
+        [DllImport(NativeBinding.TensorFlowLibrary)]
+        private static extern unsafe size_t TF_StringEncode(byte* src, size_t src_len, sbyte* dst, size_t dst_len, TF_Status status);
+
+        // extern size_t TF_StringDecode (const char *src, size_t src_len, const char **dst, size_t *dst_len, TF_Status *status);
+        [DllImport(NativeBinding.TensorFlowLibrary)]
+        private static extern unsafe size_t TF_StringDecode(sbyte* src, size_t src_len, sbyte** dst, size_t* dst_len, TF_Status status);
+
+        // extern size_t TF_StringEncodedSize (size_t len);
+        [DllImport(NativeBinding.TensorFlowLibrary)]
+        private static extern size_t TF_StringEncodedSize(size_t len);
+
         internal TFTensor(IntPtr handle) : base(handle) { }
 
         internal static Deallocator FreeTensorDataDelegate = FreeTensorData;
@@ -409,6 +422,31 @@ public TFTensor(long[] data) : base(SetupTensor(TFDataType.Int64, data, size: 8)
         /// <param name="data">Data.</param>
         public TFTensor(Complex[] data) : base(SetupTensor(TFDataType.Complex128, data, size: 16)) { }
 
+        internal static unsafe TFTensor CreateString(byte[] buffer)
+        {
+            if (buffer == null)
+                throw new ArgumentNullException(nameof(buffer));
+            //
+            // TF_STRING tensors are encoded with a table of 8-byte offsets followed by
+            // TF_StringEncode-encoded bytes.
+            //
+            var size = TF_StringEncodedSize((UIntPtr)buffer.Length);
+            IntPtr handle = TF_AllocateTensor(TFDataType.String, IntPtr.Zero, 0, (UIntPtr)((ulong)size + 8));
+
+            // Clear offset table
+            IntPtr dst = TF_TensorData(handle);
+            Marshal.WriteInt64(dst, 0);
+            var status = new TFStatus();
+            fixed (byte* src = &buffer[0])
+            {
+                TF_StringEncode(src, (UIntPtr)buffer.Length, (sbyte*)(dst + 8), size, status.handle);
+                var ok = status.StatusCode == TFCode.Ok;
+                if (!ok)
+                    return null;
+            }
+            return new TFTensor(handle);
+        }
+
         // Convenience function to factor out the setup of a new tensor from an array
         internal static IntPtr SetupTensor(TFDataType dt, long[] dims, Array data, int count, int size)
         {
@@ -591,6 +629,8 @@ public static Type TypeFromTensorType(TFDataType type)
             {
                 case TFDataType.Float:
                     return typeof(float);
+                case TFDataType.Float_ref:
+                    return typeof(float);
                 case TFDataType.Double:
                     return typeof(double);
                 case TFDataType.Int32:

src/Microsoft.ML.TensorFlow/TensorFlow/Tensorflow.cs

@@ -290,7 +290,7 @@ public TFStatus() : base(TF_NewStatus())
 
         // extern void TF_DeleteStatus (TF_Status *);
         [DllImport(NativeBinding.TensorFlowLibrary)]
-        private static extern unsafe void TF_DeleteStatus(TF_Status status);
+        internal static extern unsafe void TF_DeleteStatus(TF_Status status);
 
         internal override void NativeDispose(IntPtr handle)
         {
@@ -313,7 +313,7 @@ public void SetStatusCode(TFCode code, string msg)
 
         // extern TF_Code TF_GetCode (const TF_Status *s);
         [DllImport(NativeBinding.TensorFlowLibrary)]
-        private static extern unsafe TFCode TF_GetCode(TF_Status s);
+        internal static extern unsafe TFCode TF_GetCode(TF_Status s);
 
         /// <summary>
         /// Gets the status code for the status code.
@@ -1666,7 +1666,13 @@ internal enum TFDataType : uint
         /// <summary>
         /// 64-bit unsigned integers
         /// </summary>
-        UInt64 = 23
+        UInt64 = 23,
+
+        /// <summary>
+        /// Float reference type. It used for defining types of Variables.
+        /// Please https://www.tensorflow.org/api_docs/python/tf/DType for more details.
+        /// </summary>
+        Float_ref = 101
     }
 
     /// <summary>

src/Microsoft.ML.TensorFlow/TensorFlow/TensorflowUtils.cs

@@ -134,6 +134,8 @@ private static PrimitiveType Tf2MlNetTypeOrNull(TFDataType type)
             {
                 case TFDataType.Float:
                     return NumberType.R4;
+                case TFDataType.Float_ref:
+                    return NumberType.R4;
                 case TFDataType.Double:
                     return NumberType.R8;
                 case TFDataType.UInt16:
@@ -144,6 +146,12 @@ private static PrimitiveType Tf2MlNetTypeOrNull(TFDataType type)
                     return NumberType.U4;
                 case TFDataType.UInt64:
                     return NumberType.U8;
+                case TFDataType.Int16:
+                    return NumberType.I2;
+                case TFDataType.Int32:
+                    return NumberType.I4;
+                case TFDataType.Int64:
+                    return NumberType.I8;
                 default:
                     return null;
             }
@@ -336,6 +344,9 @@ internal static bool IsTypeSupported(TFDataType tfoutput)
                 case TFDataType.UInt16:
                 case TFDataType.UInt32:
                 case TFDataType.UInt64:
+                case TFDataType.Int16:
+                case TFDataType.Int32:
+                case TFDataType.Int64:
                     return true;
                 default:
                     return false;