LSTMfactors (Q7384241)

A method for factor retention using a pre-trained Long Short Term Memory (LSTM) Network, which is originally developed by Hochreiter and Schmidhuber (1997) <doi:10.1162/neco.1997.9.8.1735>, is provided. The sample size of the dataset used to train the LSTM model is 1,000,000. Each sample is a batch of simulated response data with a specific latent factor structure. The eigenvalues of these response data will be used as sequential data to train the LSTM. The pre-trained LSTM is capable of factor retention for real response data with a true latent factor number ranging from 1 to 10, that is, determining the number of factors.