I want a set of well-commented, well-documented functions in R, to speed things up to implement an Autoencoder in R, with Sparse, Denoising, Regularised, and Higher Order Contractive capability.
In-line rpp code may be used where necessary.
Input : Data frame or matrix. Data frame to be converted to a matrix, with categoricals becoming numerics the usual way, and options for handling missing values. Option to scale or rank numeric inputs should also be available.
All rescaling/recoding to [-1,1], and the network implemented with tanh activation functions.
results : an autoencoder which allows extraction of all connections, or just the input to hidden connections, and their conversion to a function for use in easily transforming current and new data to encoded coordinates. The transformation will be a function taking the created autoencoder as input.
This will be implemented in stages, and hence subject to an hourly rate. Appropriate testing to ensure performance will be part of the task.
I want to see any resumes with actual R and machine learning capability.
Other open jobs by this client
- Fixed-Price – extending the gbm package in R