Python - Torch - multilayer perceptron API worse than manually coded?

Question

I\'m taking a course on Coursera on machine learning and one of the labs was to code the multilayer perceptron manually and compare it to the Torch API. I did so; however, a