[D] How to do One-to-Many in Keras?

Is it possible to setup a proper one-to-many RNN in Keras without using RepeatVector to "cheat"? Let's assume a simple example: I have an alphabet of only 5 letters, each letter is represented as one hot. I want to train an RNN which takes as input the first letter, and generates predictions for what the next 4 letters of that word will be (let's forget for a moment about words of different amounts of letters). My training X would be a 5x5 matrix, representing each letter and its one hot, and the Y would be a 5x4x5 matrix, representing the next four letters in one hot for each row of X.

[D] Are there any tools for getting deep learning models mathematical equation?

is there a way to get deep learning models equation especially in keras. For example consider a neural network with one input and one output. the model looks some thing like this y=mx where y is predicted value, x is the input value and m the parmaeter. For smaller models we can use pen and paper to solve this but larger models with 100k parameters its not possible. are there tools which solves this problem?

Model Suggestions for Sparse Data

Hi, I'm currently trying to implement logistic regression on a data set which consists of about 6000 entries, with each entry having 5 features, 1 of which hardly varies among the entries, and the other 4 being categorical data (lots of variability). At best, the data set has a 90%/10% split between two labels: 0 and 1. Currently, my model is choosing to just predict 0's as it still minimizes its overall error, so I'm trying to consider any modifications I might want to make to the logistic regression model, or figure out an alternative model. Let me know if you have any suggestions!