keras-team/autokeras

Custom Search Space for ImageClassification

omkale221 opened this issue · 0 comments

inputnode = ak.ImageInput()
output_node = ak.ConvBlock(separable=True)(inputnode)

output_node = ak.DenseBlock()(output_node)

output_node = ak.ClassificationHead()(output_node)

I am using the above template as a basic search space, but this creates a sequential conv model.

  1. How to enable non-sequential search as a strategy
  2. How to make automodel search for skip connections without explicitely providing them by ak.merge at a location