Where is DTW performed during training? What is the Signal_To_Event binary doing? Still maintained?

[maximilianmordig]

Hi
Thank you for your work. I am interested in using it and have a few questions:

1. I cannot identify the code of the context-dependent model where you do the time warping. Looking at main_train.py, it calls regression_model in model_graph.py, which takes a mean square error here:
   

   DeepSimulator/pore_model/src/model_graph.py

   Line 158 in d5113cc

   pred = model_graph(input_seq, input_seq_3, input_seq_5, kr, phase)

   
   I don't see where you do the time warping. You preprocess the data with cwDTW to have one basepair per signal value, but this is not what you described in the paper in Figure 3 (for DeepSim1.0), where you described it as performing DTW at training time. Where is the second neural network f2 in the code?
2. What is the binary Signal_To_Event doing exactly? Is the source code available? In particular, what are can_value, adp and fix_value?
3. Is this tool still being actively developed? Python2.7 has reached end-of-life and the python dependencies (including TensorFlow) are very old.