public override JObject ToJObject()
{
var jobj = new JObject();
jobj["units"] = _units;
if (_inputShape != null)
{
jobj["input_shape"] = new JArray(_inputShape);
}
KerasUtils.AddActivation(jobj, _activation);
if (_kernelInitializer.GetType() == typeof(string))
{
jobj["kernel_initializer"] = (string)_kernelInitializer;
}
else
{
jobj["kernel_initializer"] = (_kernelInitializer as GraphOp).ToJObject();
}
jobj["op"] = "Dense";
return(jobj);
}
public override JObject ToJObject()
{
var jobj = new JObject();
jobj["units"] = _units;
if (_inputShape != null)
{
jobj["input_shape"] = new JArray(_inputShape);
}
jobj["use_bias"] = _useBias;
KerasUtils.AddActivation(jobj, _activation);
KerasUtils.AddActivation(jobj, "recurrent_activation", _recurrentActivation);
KerasUtils.AddStringOrObject(jobj, "kernel_initializer", _kernelInitializer);
KerasUtils.AddStringOrObject(jobj, "recurrent_initializer", _recurrentInitializer);
KerasUtils.AddStringOrObject(jobj, "bias_initializer", _biasInitializer);
jobj["unit_forget_bias"] = _unitForgetBias;
KerasUtils.AddStringOrObject(jobj, "kernel_regularizer", _kernelRegularizer);
KerasUtils.AddStringOrObject(jobj, "recurrent_regularizer", _recurrentRegularizer);
KerasUtils.AddStringOrObject(jobj, "bias_regularizer", _biasRegularizer);
KerasUtils.AddStringOrObject(jobj, "activity_regularizer", _activityRegularizer);
jobj["dropout"] = _dropout;
jobj["recurrent_dropout"] = _recurrentDropout;
jobj["op"] = "LSTM";
return(jobj);
}
public override JObject ToJObject()
{
var jobj = new JObject()
{
["input_dim"] = _inputDim,
["output_dim"] = _outputDim,
["mask_zero"] = _maskZero
};
KerasUtils.AddStringOrObject(jobj, "embedding_initializer", _embeddingInitializer);
KerasUtils.AddStringOrObject(jobj, "embedding_regularizer", _embeddingRegularizer);
if (_inputLength.HasValue)
{
jobj.Add("input_length", _inputLength.Value);
}
if (_inputShape != null)
{
jobj["input_shape"] = new JArray(_inputShape);
}
jobj["op"] = "Embedding";
return(jobj);
}
public AveragePooling2D(object poolSize = null, object strides = null, string padding = "valid", string dataFormat = null, int[] inputShape = null)
{
_poolSize = poolSize == null ? new int[] { 2, 2 } : KerasUtils.GetArray(poolSize, 2);
_strides = strides == null ? _poolSize : KerasUtils.GetArray(strides, 2);
_padding = padding;
_dataFormat = dataFormat == null ? Globals.DataFormat : dataFormat;
_inputShape = inputShape;
}
public override JObject ToJObject()
{
var jobj = new JObject();
jobj["filters"] = _filters;
jobj["kernel_size"] = new JArray(_kernelSize);
jobj["strides"] = new JArray(_strides);
jobj["use_bias"] = _useBias;
if (_inputShape != null)
{
jobj["input_shape"] = new JArray(_inputShape);
}
KerasUtils.AddActivation(jobj, _activation);
jobj["bias_initializer"] = "zeros";
jobj.Add("op", "Conv2D");
return(jobj);
}
public Conv2D(int filters, object kernelSize, object strides = null, object activation = null, bool useBias = true, int [] inputShape = null)
{
_inputShape = inputShape;
_kernelSize = KerasUtils.GetArray(kernelSize, 2);
if (_kernelSize == null)
{
throw new ArgumentException("The kernelSize parameter type is not supported.");
}
_filters = filters;
_activation = activation;
_useBias = useBias;
if (strides == null)
{
_strides = new int[] { 1, 1 }
}
;
else
{
_strides = KerasUtils.GetArray(strides, 2);
}
_op = "Conv2D";
}
