Default parameters for model training should be extracted from Persephone

[shuttle1987]

Consider this situation:

def post(modelInfo):
    """Create a new transcription model"""
    current_corpus = DBcorpus.query.get_or_404(modelInfo['corpusID'])

    min_epochs = modelInfo.get('minimumEpochs', 0)
    max_epochs = modelInfo.get('maximumEpochs', None)
    if max_epochs and min_epochs > max_epochs:
        return "minimum number of epochs must be smaller than maximum", 400

    early_stopping_steps = modelInfo.get('earlyStoppingSteps', None)
    num_layers = modelInfo.get('numberLayers', 3)
    hidden_size = modelInfo.get('hiddenSize', 250)
    beam_width = modelInfo.get('beamWidth', 100)
    decoding_merge_repeated = modelInfo.get('decodingMergeRepeated', True)

As you can see default parameters are provided here if the request doesn't contain them. These currently match up with what occurs in Persephone but they could get out of sync easily.

[shuttle1987]

PR #94 addresses this somewhat for the model.train calls.

[shuttle1987]

I can see a case for loading these from a config file too, the hardcoded nature that exists here is merely a interim hack to keep other work unblocked.

[shuttle1987]

Could do something like this https://stackoverflow.com/questions/12627118/get-a-function-arguments-default-value