Sequence Classification Example Using a LSTM Recurrent Neural Network (deeplearning4j) to train/test/forecast the network model on subject: if the sequence of double is matching certain criteria.
Namely learns how to classify univariate time series as belonging to one of six categories. Categories are: Normal, Cyclic, Increasing trend, Decreasing trend, Upward shift, Downward shift. The original code was reworked to not use the file-system
#The Data is the UCI Synthetic Control Chart Time Series Data Set
The deatils are avaliable as well as the raw raw dataset
#The Network configuration:
1 x input; init of weights - XAVIER; gradient update - NADAM;
Layer1: LSTM activation - TANH;
Layer2: RNN activation - SOFTMAX; loss function - MCXENT