Input of Time2Vec embed

[MBelniak]

I have a question that bothers me a lot recently - it appears to me like some people on the web get time2vec concept wrong. What you described in the article is correct from my point of view, but the implementation is incorrect, at least for NY-TX dataset.
As I understand, time2vec is used to create an embedding of time. Thus, the input of time2vec layer should be some time information, like day number since beginning of data, timestep, year etc. To support my claim, the excerpt from time2vec paper:

To answer Q3, we trained a model on our synthesized dataset where the input integer (day) is used as
the time for Time2Vec

I've run your model with ny-tx dataset and from what I see the input of time2vec is just the RAW temperature data, which doesn't contain any time-related information.
I will be grateful if you dispel my doubts, maybe I misunderstand sth. And, thank you for a great article :)

[MBelniak]

After further analysis I find out that what I thought was an input tensor (x) was in fact tensor of date features (year, month, day etc.), so everything is clear now. Sorry for disturbing.