public override object Load(XmlNode node)
{
var result = (NeuralNetwork)Activator.CreateInstance(Type);
var loss = XmlStorage.GetObjectValue <LossFunction>(node, "LossFunction");
result.UseLossFunction(loss);
var optimizer = XmlStorage.GetObjectValue <IOptimizer>(node, "Optimizer");
result.UseOptimizer(optimizer);
var regularizer = XmlStorage.GetObjectValue <IRegularizer>(node, "Regularizer");
result.UseRegularizer(regularizer);
var nodeList = node.SelectSingleNode("Layers").ChildNodes;
foreach (XmlNode n in nodeList)
{
var layer = XmlStorage.LoadFromNode <ILayer>(n);
result.AddLayer(layer);
}
return(result);
}
public override object Load(XmlNode node)
{
if (node == null)
{
return(null);
}
if (!node.HasChildNodes)
{
return(null);
}
var unitCount = XmlStorage.GetIntValue(node, "UnitCount");
var result = (FullLayer)Activator.CreateInstance(typeof(FullLayer), unitCount);
var weights = XmlStorage.GetObjectValue <TensorOld>(node, "Weights");
if (weights != null)
{
result.SetWeights(weights);
}
var bias = XmlStorage.GetObjectValue <TensorOld>(node, "Bias");
if (bias != null)
{
result.SetBias(bias);
}
return(result);
}
public override object Load(XmlNode node)
{
var count = XmlStorage.GetIntValue(node, "FilterCount");
var rows = XmlStorage.GetIntValue(node, "FilterRows");
var columns = XmlStorage.GetIntValue(node, "FilterColumns");
var rowStride = XmlStorage.GetIntValue(node, "RowStride");
var columnStride = XmlStorage.GetIntValue(node, "ColumnStride");
var rowPadding = XmlStorage.GetIntValue(node, "RowPadding");
var columnPadding = XmlStorage.GetIntValue(node, "ColumnPadding");
var paddingValue = XmlStorage.GetDoubleValue(node, "PaddingValue");
var result = (ConvLayer)Activator.CreateInstance(typeof(ConvLayer), count, rows, columns, rowStride, columnStride, rowPadding, columnPadding);
result.PaddingValue = paddingValue;
var filters = XmlStorage.GetObjectValue <TensorOld>(node, "Filters");
if (filters != null)
{
result.SetFilters(filters);
}
var bias = XmlStorage.GetObjectValue <TensorOld>(node, "Bias");
if (bias != null)
{
result.SetBias(bias);
}
return(result);
}
public override object Load(XmlNode node)
{
var rows = XmlStorage.GetIntValue(node, "Rows");
var columns = XmlStorage.GetIntValue(node, "Columns");
var rowStride = XmlStorage.GetIntValue(node, "RowStride");
var columnStride = XmlStorage.GetIntValue(node, "ColumnStride");
return(Activator.CreateInstance(typeof(MaxPooling), rows, columns, rowStride, columnStride));
}
internal static XmlElement SavePoolingLayer(string type, XmlDocument doc, PoolingLayer layer)
{
var el = doc.CreateElement(type);
XmlStorage.AddChild(el, "Rows", layer.Rows.ToString());
XmlStorage.AddChild(el, "Columns", layer.Columns.ToString());
XmlStorage.AddChild(el, "RowStride", layer.RowStride.ToString());
XmlStorage.AddChild(el, "ColumnStride", layer.ColumnStride.ToString());
return(el);
}
public override XmlElement Save(XmlDocument doc, object o)
{
if (!(o is FullLayer layer))
{
throw new Exception("layer must be FullLayer!");
}
var el = GetRootElement(doc);
XmlStorage.AddChild(el, "UnitCount", layer.UnitCount);
XmlStorage.AddObjectChild(el, "Weights", layer.Weights);
XmlStorage.AddObjectChild(el, "Bias", layer.Bias);
return(el);
}
public override object Load(XmlNode node)
{
if (node == null)
{
return(null);
}
if (!node.HasChildNodes)
{
return(null);
}
var shape = XmlStorage.GetIntArray(node, "Shape");
var values = XmlStorage.GetDoubleArray(node, "Values");
return(Activator.CreateInstance(typeof(TensorOld), values, shape));
}
public override XmlElement Save(XmlDocument doc, object o)
{
var el = doc.CreateElement("Tensor");
if (o == null)
{
return(el);
}
if (!(o is TensorOld tensor))
{
throw new Exception("layer must be Tensor!");
}
XmlStorage.AddArrayChild(el, "Shape", tensor.shape);
XmlStorage.AddArrayChild(el, "Values", tensor.GetRawValues());
return(el);
}
public override XmlElement Save(XmlDocument doc, object o)
{
if (!(o is NeuralNetwork nn))
{
throw new Exception("object must be NeuralNetwork");
}
var el = GetRootElement(doc);
XmlStorage.AddObjectChild(el, "LossFunction", nn.LossFunction);
XmlStorage.AddObjectChild(el, "Optimizer", nn.Optimizer);
XmlStorage.AddObjectChild(el, "Regularizer", nn.Regularizer);
var layers = doc.CreateElement("Layers");
foreach (var item in nn.TrainingLayers)
{
XmlStorage.AddObjectChild(layers, item);
}
XmlStorage.AddChild(el, layers);
return(el);
}
public override XmlElement Save(XmlDocument doc, object o)
{
if (!(o is ConvLayer layer))
{
throw new Exception("layer must be ConvLayer!");
}
var el = GetRootElement(doc);
XmlStorage.AddChild(el, "FilterCount", layer.FilterCount.ToString());
XmlStorage.AddChild(el, "FilterRows", layer.FilterRows.ToString());
XmlStorage.AddChild(el, "FilterColumns", layer.FilterColumns.ToString());
XmlStorage.AddChild(el, "RowStride", layer.RowStride.ToString());
XmlStorage.AddChild(el, "ColumnStride", layer.ColumnStride.ToString());
XmlStorage.AddChild(el, "RowPadding", layer.RowPadding.ToString());
XmlStorage.AddChild(el, "ColumnPadding", layer.ColumnPadding.ToString());
XmlStorage.AddChild(el, "PaddingValue", layer.PaddingValue.ToString());
XmlStorage.AddObjectChild(el, "Filters", layer.Filters);
XmlStorage.AddObjectChild(el, "Bias", layer.Bias);
return(el);
}