[Time Series BikeDemandForecasting] evaluate and predict

For the sample BikeDemandForecasting, what is the difference between using model.transform(testdata) and forecaster.predict() when predicting future values? Why am I getting different results for the first date in 2012?
Another question is that when I comment out the evaluation function(Evaluate(second...)), I actually get different results when predicting, how could the evaluate process change the model?
Thanks.