Review and refactor DateTime encoder

[breznak]

Follow up to #462 ,

Date time encoder is only available in python:

• [ ] make a c++ version + bindings?

Python datetime:

• [ ] consider removing "weekend" field, replaced by cusomDays
  • same with holidays?
• [ ] make more general purpose!
  • current DateTime is an old encoder, heavily specified for "hotgym" dataset, that has weekly period, and is tied to human behavior (such as weekend, ...)
  • [ ] specify encoding width for minutes, seconds, years? (currently seasons, days only)
• [ ] extend "date string" format of inputs, such as for customDays ("Mon", "Tuesday") to use full datetime's format
  • so we can specify "Mon, 13:05" etc
• [ ] test quality of encoding using RDSE as the internal encoder, instead of ScalarEncoder
• [ ] allow "smoothing of upcoming events"
  • currently, say Wed is a special day, output looks like (day:special): Mon:0, Tue:0, Wed:1,Thr:0, ..
  • change to smoothing: Mon:0, Tue:0.5, Wed:1.0, Thr:0, ...

[ctrl-z-9000-times]

I dont think that there is very much demand for this encoder in C++

I dont want to remove fields from the encoder. I think that the weekend field is a great idea. The purpose of this encoder is to demonstrate the capabilities of the TM. It is not supposed to be biological, its for fun examples.

Minutes seconds years? I thought that it already did some of this but if not then sure it could be added, +1

Date strings sounds good too, although i hope there is a library which already does this.

[dkeeney]

I dont think that there is very much demand for this encoder in C++

Probably not but if such a thing existed it would make it easier to make this encoder as a built-in region for NetworkAPI. Also, C++ only folks would be able to use it.

[breznak]

Now, how can we fold this into the networkAPI? Is there going to be a C++ version of this?

I would also say c++ version would find it's use, as some form of date-time is used in almost every other example, see hotgym dataset that we want to replicate for c++ in #30

On the other hand, David is right the c++ impl would likely be more complicated, as python has the neat datetime library.

Also, I'd wait a bit as I'm unsatisfied with this as a "general purpos" time/date encoder (see the problem with years/months), so I'll seek forums for thoughts on more generic implementation of this. So for now I'd suggest c++ on hold.

[breznak]

FYI this should make porting to c++ feasible: https://github.com/coin-au-carre/datetimepp
maybe there's a better c++ implementation of python's datetime .