In the actual work, the timing class data modeling analysis is the more important part, we can use machine learning to build a general regression model to predict the value of the analysis may be based on neural network to build the network model to predict the complete time series data analysis in a relatively simple task, using machine learning to build regression models generally meet the needs, but when the actual problems faced relatively complex, simple models are often difficult to perform competently, there need to be more efficient, more model skills to do the job, LSTM this type of model is one of the best, LSTM RNN essentially a special case, the focus here is not to explain in detail the principles on LSTM of this column series mainly emphasis on practice and applications.
For this type of analysis of time-series data modeling and interested friends can go to look at my previous two articles:
This article is the practice of convolution integration model + LSTM structure based on a number of papers which have seen a number of related implementation, integration and convolution LSTM might really be a good timing modeling and analysis tools, because the volume the product will be free to dig and computing features, LSTM itself is to time-series data modeling and raw, a combination of both to some extent, there may be a better role in promoting, but this is not absolute, for the time-series data , it is very important that the need to model time series data capture feature, so the model has some memory, but could undermine the role of the convolution timing characteristics inherent in the original data, which may give LSTM bring learning negative impact, of course, these are just my personal understanding and analysis, then we have concrete data to actual combat.
The following is a sample data about one day:
"2016-03-01 00:00:00"," 7