Is there any smart way to perform feature selection for LSTM sequence-to-sequence regression, which does not require testing the regression for different sets of features, but rather can tell how useful is the specific feature for a memory-based regression? If there are some approaches i would appreciate some references to the literature.
not sure to understand, but, need #Labels.creations in samples, LSTM models is based on k.filtering and k.forget.gates, so that, specicific criterions, biases, weights appears when scaling, otherwise, be more accurate, if i didn't embrace?
Deep learning can be used for feature selection / extraction. You can use CNN at the initial layers or auto-encoders to get desired features before passing it on to LSTM
Feature selection is one of the most embarrassing task, you can use a bloc of Conv layers.. you can try extracting some statistical features from your data
- Badges
- Science topic