Specify 'None' as Shape dimension #231
Description
I'm trying to do a Conv1D layer with variable features
1st attempt
let model = new Sequential()
model.Add(new Input(Shape(3)))
model.Add(new Conv1D(50, 1 ))
Error: Python.Runtime.PythonException: ValueError : Input 0 of layer "conv1d_9" is incompatible with the layer: expected min_ndim=3, found ndim=2. Full shape received: (None, 3)
2nd attempt setting the batch_shape arg (thinking the batch size will fix it)
let model = new Sequential()
model.Add(new Input(Shape(3), Shape(1)))
model.Add(new Conv1D(50, 1 ))
Error: Python.Runtime.PythonException: ValueError : Only provide the
shapeORbatch_input_shapeargument to Input, not both at the same time.
According to this issue batch_size is being depreciated from tensorflow. So I figure that is the reason for the latter python error.
I've looked at the design of the Shape class and in it's current form and it only takes int types so dead end there.
The first error suggests that internally within Keras.NET a conversion is being done to emit (None,3) to python but in my brief look I cant see how.
Suggestions for a work around ?
Or even a pointer to the section of code in Keras.NET that does the shape conversion, so I can hack in something local ?