Merge pull request #1137 from Beacontownfc/mybranch2

fix: Fix the problem where the  model.save_weights() can't save multi-layer weight

src/TensorFlowNET.Keras/Saving/hdf5_format.cs

@@ -7,6 +7,8 @@ using HDF5CSharp;
 using static Tensorflow.Binding;
 using static Tensorflow.KerasApi;
 using System.Linq;
+using System.Text.RegularExpressions;
+
 namespace Tensorflow.Keras.Saving
 {
     public class hdf5_format
@@ -132,7 +134,9 @@ namespace Tensorflow.Keras.Saving
                 var weight_names = load_attributes_from_hdf5_group(g, "weight_names");
                 foreach (var i_ in weight_names)
                 {
-                    (success, Array result) = Hdf5.ReadDataset<float>(g, i_);
+                    var vm = Regex.Replace(i_, "/", "$");
+                    vm = i_.Split('/')[0] + "/$" + vm.Substring(i_.Split('/')[0].Length + 1, i_.Length - i_.Split('/')[0].Length - 1); 
+                    (success, Array result) = Hdf5.ReadDataset<float>(g, vm);
                     if (success)
                         weight_values.Add(np.array(result));
                 }
@@ -193,7 +197,8 @@ namespace Tensorflow.Keras.Saving
                     if (name.IndexOf("/") > 1)
                     {
                         var crDataGroup = Hdf5.CreateOrOpenGroup(g, Hdf5Utils.NormalizedName(name.Split('/')[0]));
-                        WriteDataset(crDataGroup, name.Split('/')[1], tensor);
+                        var _name = Regex.Replace(name.Substring(name.Split('/')[0].Length, name.Length - name.Split('/')[0].Length), "/", "$");
+                        WriteDataset(crDataGroup, _name, tensor);
                         Hdf5.CloseGroup(crDataGroup);
                     }
                     else