When first learning about LSTMs I created some toy models to help me learn how LSTMs are like the more basic types of neural networks, such as RNNs and MLPs. I was also teaching myself Lua and Torch at the same time, and the result is this tutorial.
We'll first first attack a toy modeling problem using a multi-layer perceptron, pretty much the simplest type of neural net. Then we'll progress to recursive neuran lets (RNN) and finally the long-term short-term memory (LSTM). But to begin, we'll need a toy problem to model.
All the source code is available at kbullaughey/lstm-play.