Step function as neuron activation function

Hi all,

I am trying to analyze a big chunk of data using an MLP in TMVA. Unfortunately the neuron activation function slows things down. I am running for 3 days now and only managed to analyze 430000 events. To speed things up I would like to test a step function instead of a sigmoid. Unfortunately I wasn’t able to find it in the TMVA documentation. Is it implemented and just not documented, or is there no chance at all to get it working?