public Linear(
int inputCount,
int outputCount,
bool noBias = false,
Array initialW = null,
Array initialb = null,
CompressibleActivation activation = null,
string name = FUNCTION_NAME,
string[] inputNames = null,
string[] outputNames = null,
bool gpuEnable = false
) : base(
FUNCTION_NAME,
activation,
new[] { new KeyValuePair <string, string>(PARAM_NAME, PARAM_VALUE) },
name,
inputNames,
outputNames,
gpuEnable)
{
OutputCount = outputCount;
InputCount = inputCount;
Weight = new NdArray(outputCount, inputCount);
Weight.Name = Name + ".Weight";
NoBias = noBias;
Parameters = new NdArray[noBias ? 1 : 2];
if (initialW == null)
{
Initializer.InitWeight(Weight);
}
else
{
Weight.Data = Real.GetArray(initialW);
}
Parameters[0] = Weight;
if (!noBias)
{
Bias = new NdArray(outputCount);
Bias.Name = Name + ".Bias";
if (initialb != null)
{
Bias.Data = Real.GetArray(initialb);
}
Parameters[1] = Bias;
}
SetArray(Weight.Name, Weight);
SetArray(Bias.Name, Bias);
}
private static void HandleGPU(bool verbose, Stopwatch sw, CompressibleActivation ca, NdArray nArray)
{
sw.Restart();
NdArray[] gradArrayGpu = ca.Forward(verbose, nArray);
sw.Stop();
if (verbose)
{
RILogManager.Default?.SendDebug("Forward [Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
}
gradArrayGpu[0].Grad = gradArrayGpu[0].Data;
sw.Restart();
ca.Backward(verbose, gradArrayGpu);
sw.Stop();
if (verbose)
{
RILogManager.Default?.SendDebug("Backward[Gpu] : " + (sw.ElapsedTicks / (Stopwatch.Frequency / (1000L * 1000L))).ToString("n0") + "μｓ");
}
}
////////////////////////////////////////////////////////////////////////////////////////////////////
/// <summary>
/// Initializes a new instance of the KelpNet.Functions.Connections.Deconvolution2D class.
/// </summary>
///
/// <param name="verbose"> True to verbose. </param>
/// <param name="inputChannels"> The input channels. </param>
/// <param name="outputChannels"> The output channels. </param>
/// <param name="kSize"> The size. </param>
/// <param name="subSample"> (Optional) The sub sample. </param>
/// <param name="trim"> (Optional) The trim. </param>
/// <param name="noBias"> (Optional) True to no bias. </param>
/// <param name="initialW"> (Optional) The initial w. </param>
/// <param name="initialb"> (Optional) The initialb. </param>
/// <param name="activation"> (Optional) The activation. </param>
/// <param name="name"> (Optional) The name. </param>
/// <param name="inputNames"> (Optional) List of names of the inputs. </param>
/// <param name="outputNames"> (Optional) List of names of the outputs. </param>
/// <param name="gpuEnable"> (Optional) True if GPU enable. </param>
////////////////////////////////////////////////////////////////////////////////////////////////////
public Deconvolution2D(bool verbose, int inputChannels, int outputChannels, int kSize, int subSample = 1, int trim = 0, bool noBias = false, [CanBeNull] Array initialW = null, [CanBeNull] Array initialb = null, [CanBeNull] CompressibleActivation activation = null, [CanBeNull] string name = FUNCTION_NAME, [CanBeNull] string[] inputNames = null, [CanBeNull] string[] outputNames = null, bool gpuEnable = false)
: base(verbose, FUNCTION_NAME, activation, new [] { new KeyValuePair <string, string>(PARAM_NAME, PARAM_VALUE) }, name, inputNames, outputNames, gpuEnable)
{
_kWidth = kSize;
_kHeight = kSize;
_trimX = trim;
_trimY = trim;
_subSampleX = subSample;
_subSampleY = subSample;
NoBias = noBias;
Parameters = new NdArray[noBias ? 1 : 2];
OutputCount = outputChannels;
InputCount = inputChannels;
Initialize(initialW, initialb);
}
////////////////////////////////////////////////////////////////////////////////////////////////////
/// <summary>
/// Initializes a new instance of the KelpNet.Functions.Connections.Deconvolution2D class.
/// </summary>
///
/// <param name="verbose"> True to verbose. </param>
/// <param name="inputChannels"> The input channels. </param>
/// <param name="outputChannels"> The output channels. </param>
/// <param name="kSize"> The size. </param>
/// <param name="subSample"> (Optional) The sub sample. </param>
/// <param name="trim"> (Optional) The trim. </param>
/// <param name="noBias"> (Optional) True to no bias. </param>
/// <param name="initialW"> (Optional) The initial w. </param>
/// <param name="initialb"> (Optional) The initialb. </param>
/// <param name="activation"> (Optional) The activation. </param>
/// <param name="name"> (Optional) The name. </param>
/// <param name="inputNames"> (Optional) List of names of the inputs. </param>
/// <param name="outputNames"> (Optional) List of names of the outputs. </param>
/// <param name="gpuEnable"> (Optional) True if GPU enable. </param>
////////////////////////////////////////////////////////////////////////////////////////////////////
public Deconvolution2D(bool verbose, int inputChannels, int outputChannels, Size kSize, Size subSample = new Size(), Size trim = new Size(), bool noBias = false, [CanBeNull] Array initialW = null, [CanBeNull] Array initialb = null, [CanBeNull] CompressibleActivation activation = null, [CanBeNull] string name = FUNCTION_NAME, [CanBeNull] string[] inputNames = null, [CanBeNull] string[] outputNames = null, bool gpuEnable = false)
: base(verbose, FUNCTION_NAME, activation, new [] { new KeyValuePair <string, string>(PARAM_NAME, PARAM_VALUE) }, name, inputNames, outputNames, gpuEnable)
{
if (subSample == Size.Empty)
{
subSample = new Size(1, 1);
}
if (trim == Size.Empty)
{
trim = new Size(0, 0);
}
_kWidth = kSize.Width;
_kHeight = kSize.Height;
_trimX = trim.Width;
_trimY = trim.Height;
NoBias = noBias;
_subSampleX = subSample.Width;
_subSampleY = subSample.Height;
Parameters = new NdArray[noBias ? 1 : 2];
OutputCount = outputChannels;
InputCount = inputChannels;
Initialize(initialW, initialb);
}
public Convolution2D(int inputChannels, int outputChannels, Size kSize, Size stride = new Size(), Size pad = new Size(), bool noBias = false, Array initialW = null, Array initialb = null, CompressibleActivation activation = null, string name = FUNCTION_NAME, string[] inputNames = null, string[] outputNames = null, bool gpuEnable = false) : base(FUNCTION_NAME, activation, new[] { new KeyValuePair <string, string>(PARAM_NAME, PARAM_VALUE) }, name, inputNames, outputNames, gpuEnable)
{
if (pad == Size.Empty)
{
pad = new Size(0, 0);
}
if (stride == Size.Empty)
{
stride = new Size(1, 1);
}
_kWidth = kSize.Width;
_kHeight = kSize.Height;
_strideX = stride.Width;
_strideY = stride.Height;
_padX = pad.Width;
_padY = pad.Height;
NoBias = noBias;
Parameters = new NdArray[noBias ? 1 : 2];
OutputCount = outputChannels;
InputCount = inputChannels;
Initialize(initialW, initialb);
}
public Convolution2D(int inputChannels, int outputChannels, int kSize, int stride = 1, int pad = 0, bool noBias = false, Array initialW = null, Array initialb = null, CompressibleActivation activation = null, string name = FUNCTION_NAME, string[] inputNames = null, string[] outputNames = null, bool gpuEnable = false) : base(FUNCTION_NAME, activation, new[] { new KeyValuePair <string, string>(PARAM_NAME, PARAM_VALUE) }, name, inputNames, outputNames, gpuEnable)
{
_kWidth = kSize;
_kHeight = kSize;
_strideX = stride;
_strideY = stride;
_padX = pad;
_padY = pad;
NoBias = noBias;
Parameters = new NdArray[noBias ? 1 : 2];
OutputCount = outputChannels;
InputCount = inputChannels;
Initialize(initialW, initialb);
}
////////////////////////////////////////////////////////////////////////////////////////////////////
/// <summary>
/// Initializes a new instance of the KelpNet.Functions.Connections.Linear class.
/// </summary>
///
/// <param name="verbose"> True to verbose. </param>
/// <param name="inputCount"> Number of inputs. </param>
/// <param name="outputCount"> Number of outputs. </param>
/// <param name="noBias"> (Optional) True to no bias. </param>
/// <param name="initialW"> (Optional) The initial w. </param>
/// <param name="initialb"> (Optional) The initialb. </param>
/// <param name="activation"> (Optional) The activation. </param>
/// <param name="name"> (Optional) The name. </param>
/// <param name="inputNames"> (Optional) List of names of the inputs. </param>
/// <param name="outputNames"> (Optional) List of names of the outputs. </param>
/// <param name="gpuEnable"> (Optional) True if GPU enable. </param>
////////////////////////////////////////////////////////////////////////////////////////////////////
public Linear(bool verbose, int inputCount, int outputCount, bool noBias = false, [CanBeNull] Array initialW = null, [CanBeNull] Array initialb = null, [CanBeNull] CompressibleActivation activation = null, [CanBeNull] string name = FUNCTION_NAME, [CanBeNull] string[] inputNames = null, [CanBeNull] string[] outputNames = null, bool gpuEnable = false)
: base(verbose, FUNCTION_NAME, activation, new[] { new KeyValuePair <string, string>(PARAM_NAME, PARAM_VALUE) }, name, inputNames, outputNames, gpuEnable)
{
Verbose = verbose;
OutputCount = outputCount;
InputCount = inputCount;
Weight = new NdArray(outputCount, inputCount)
{
Name = Name + " Weight"
};
NoBias = noBias;
Parameters = new NdArray[noBias ? 1 : 2];
if (initialW == null)
{
Initializer.InitWeight(Weight);
}
else
{
Weight.Data = Real.GetArray(initialW);
}
Parameters[0] = Weight;
if (Verbose)
{
RILogManager.Default?.ViewerSendWatch(Name + " Weight", Weight.Length <= 30 ? Weight : Weight.Shape.Length);
}
if (!noBias)
{
Bias = new NdArray(outputCount)
{
Name = Name + " Bias"
};
if (initialb != null)
{
Bias.Data = Real.GetArray(initialb);
}
Parameters[1] = Bias;
if (Verbose)
{
RILogManager.Default?.ViewerSendWatch(Name + " Bias", Bias.Length <= 30 ? Bias : Bias.Shape.Length);
}
}
}
