采用栈式 RBM 的深度神经网络。自训练, 自控制(This paper is a continuation of the depth of the neural network and the choice of the previous predictor. Here we will be covered by the depth of the neural network characteristic of stacked RBM initialization, and it is in the "Darch" software package to achieve. The possibility of using hidden Markoff model to improve the prediction performance of neural networks is also revealed. In short, we will implement an actionable intelligent transaction program in a programmable manner.)