internal static Layer FromXML(XMLElement root, Layer previousLayer) {
float[] StringToArray(string s) {
string[] valStrs = s.Split(',');
float[] array = new float[valStrs.Length];
for (int i = 0; i < valStrs.Length; i++) {
array[i] = float.Parse(valStrs[i]);
}
return array;
}
string type = root.Tag;
if (type == typeof(InputLayer).Name) {
int width = int.Parse(root.GetAttribute("Width").Value);
int height = int.Parse(root.GetAttribute("Height").Value);
int depth = int.Parse(root.GetAttribute("Depth").Value);
return new InputLayer(width, height, depth);
}
if (type == typeof(MaxPoolingLayer).Name) {
int filterSize = int.Parse(root.GetAttribute("FilterSize").Value);
int stride = int.Parse(root.GetAttribute("Stride").Value);
return new MaxPoolingLayer(filterSize, stride, previousLayer);
}
if (type == typeof(FullyConnectedLayer).Name) {
int neurons = int.Parse(root.GetAttribute("Neurons").Value);
ActivationFunctions.ActivationFunction activation = ActivationFunctions.FromXML(root.ChildWithTag("ActivationFunction"));
FullyConnectedLayer layer = new FullyConnectedLayer(neurons, previousLayer, activation);
float[] weights = StringToArray(root.ChildWithTag("Weights").Value);
for (int i = 0; i < layer.Weights; i++) {
layer.SetWeight(i, weights[i]);
}
float[] biases = StringToArray(root.ChildWithTag("Biases").Value);
for (int i = 0; i < layer.Biases; i++) {
layer.SetBias(i, biases[i]);
}
return layer;
}
if (type == typeof(ConvolutionalLayer).Name) {
int filterCount = int.Parse(root.GetAttribute("FilterCount").Value);
int filterSize = int.Parse(root.GetAttribute("FilterSize").Value);
int stride = int.Parse(root.GetAttribute("Stride").Value);
int zeroPadding = int.Parse(root.GetAttribute("ZeroPadding").Value);
ActivationFunctions.ActivationFunction activation = ActivationFunctions.FromXML(root.ChildWithTag("ActivationFunction"));
ConvolutionalLayer layer = new ConvolutionalLayer(filterCount, filterSize, stride, zeroPadding, previousLayer, activation);
float[] weights = StringToArray(root.ChildWithTag("Weights").Value);
for (int i = 0; i < layer.Weights; i++) {
layer.SetWeight(i, weights[i]);
}
float[] biases = StringToArray(root.ChildWithTag("Biases").Value);
for (int i = 0; i < layer.Biases; i++) {
layer.SetBias(i, biases[i]);
}
return layer;
}
throw new NotImplementedException(); // Never reached
}