You need to have theano and sklearn installed for training.
Input files should have following format:

<reference sequence>
<mean of first event> <std of first event> <lenght of first event>
<mean of second event> <std of second event> <lenght of second event>
...

See examples.

Get alignment of events:
python realign.py ../networks/r94.pkl <list of files>

For each file with <name> it creates a file <name>r, so for file 1.txt it will create 1.txtr.

For training:
python train.py <optionally name of starting network like ../networks/r94.pkl> <list of files>

Remember to put files ending with r into training.

After each epoch the training will output:
<number of epoch> <current training log likelihood> <accuracy of first softmax> <accuracy of second
softmax> <approximate overall sequence identity> <time>
<distribution of outputs from first softmax>
<distribution of outputs from second softmax>
<confusion matrix for first softmax>
<confusion matrix for second softmax>

In folder with timestamp of training start there will be file names latest.pkl and each 20
epoch will be saved current snapshot of the network. 

