Deepsignal

make `--reference_path` not a required input in *extract* and *call_mods* module

handle gz file in call_freq script

make sure results of each read be written together in call_mods' output

enable multi-class (>=3) training/predicting,

fix bug of extrating contig name from fast5s

modify denoise module
fix success_file bug
update README

Prevent Queue.qsize() from raising NotImplementedError on Mac OS X (github: vterron/lemon9ca6b4b)
covert raw signals to pA values before normalization in extract_feature module
add denoise module
add module-chosen options of model (rnn/base/cnn), re-train human model

add option –positions in extract_features module, add option/function of binary_format feature file to speed up training