expose pretrain decay factor function

small fixes to random negative samples

Perform imputation in training script

Added unit for training model with imputed data

adding test to make sure imputation improves predictive accuracy on a small allele, added allow_unknown_amino_acids to from_hyperparameters

fixed neural network test, fixed length filtering of data loading, started on adding imputation to training script

AlleleData objects now have numpy arrays for all attributes

moving helpers for performing dataset imputations from experiments 

Minor fixes

Copying imputation and soft-pretraining logic from experiments to main training script

 - Add default values and usage to argparse help messages
 - Require theano >=0.8.2 (was getting errors instantiating a model at 0.7.0)