Open ccoffrin opened 6 years ago
Are we talking about the default behavior of the learning algorithm or the parser for the data structure? If it is the earlier we can imagine having a different default behavior depending on the algo. (Inverse Ising can convert to bool or int, gaussian methods can convert to float, etc...).
The core issue here is when a user reads discrete data into Julia as floats, instead of int or bool.
You propose a good solution. If the user passes a floating point samples into a discrete model solver, we can first try to convert the floating point data to discrete data and throw a warning if it works and an error if it does not work.
How should we deal with float vs int data types?