some question of linear_regression/linear_regression.py

[rywc2005]

question

1. cost_function :(1 / 2 * num_examples)（which is num_examples/2），actually should be (1 / (2 * num_examples))。this error maybe mislead the training .
2. When processing new datas,predict and get_cost newly call prepare_for_training without the features_mean / features_deviation ，which is saved in training stage；if normalize_data=True，it maybe let the difference of norm between train&test.