[WEEK 3] Make World Green Again!
This week, we have worked on methodology. We didn’t want to start with a randomly chosen algorithm that’s why we have searched about the past implementations of the subject. We have seen that in most of the projects, multilayer neural network is used, at least in the experimental phase. These readings made us start with neural networks.
At the first attempt, we have tried to estimate the peak value of NO2 using concentrations of NO and NO2 from early hour measurements. The trial was not satisfactory for evaluation and we have decided to extend our dataset from another source.[1] The data has missing values which have to be eliminated. Size of the data is very big and it takes time to scan the all of it and combine with wanted parameters.
We are now intended to make a series of experiments in order to analyze our network implementation.
A BASIC INTRODUCTION TO NEURAL NETWORKS
A Neural network is a biologically-inspired supervised learning method. A neural network is organized in layers. Each layer takes the output of the previous layer as its input and produces an output to be used in next layer or as the result itself for the last layer. Neural networks become nonlinear classifiers with help of the used activation functions.
There are lots of tutorials on neural networks for beginners and there are several python frameworks that can be used. Some of them are Theano, TensorFlow, Caffe, Lasagne and scikit-learn. We are currently using scikit-learn.
A recommended post to write a simple multilayer neural network
A video tutorial to understand the process of a neural network
