Open wrhm opened 6 years ago
I'm actively thinking about different language stuff to look into. From the NLP side, would you say the main goal is to engineer a number of feature-detectors, that would then be trained on by the ensemble?
I guess my question is: instead of \<NLP stuff from an article> leading directly to \<trading action>, should our NLP tools/scripts produce derived data (like sentiment scores, author-validity, crypto-currency name/alias aggregation, performance prediction/retrospection)?
I think I answered my own question here. It's honestly just helpful to write some of this stuff out and stare at it.
I don't see how \<NLP stuff from an article> is so different from \<derived data>? Can you elaborate?
Meta-comment: using carets ("<" / ">") was a poor choice on my part, since they only render when escaped/quoted. Did you use them in your reply?
I don't see how is so different from ?
I'm putting together lists of goals and tasks in src/kraken_brain/nlp/README.md
which I will push soon.
I don't see how
is so different from ? Can you elaborate?
Yeah whoops my whole comment there was pretty rambly. You're right; there is no distinction - our NLP tools will be a collection of different metrics to feed model training.
Check out nlp/README.md for a centralized overview of the currently-planned goals tasks.
https://yajasd.github.io/2018/06/20/Bullish-Bearish-or-Consolidation/
Looks like a good resource :)
Also I'm still on holiday but wanted to post this before I forget.
1) something to consider is also whether an article is predictive or reactionary - whether it sees a drop coming up, or if it seeks to explain why it's falling. Understandably both would "result" in different things
2) once you've got an outline more or less set up tag Bryan and I so we're all on the same page
3) break up the outline into actionable plans. When we meet in person we're going to talk about how to distribute the load for news scraping + linear model construction.
Feel free to PM Bryan and I if you need help. Idk what he's interested in but I believe he can contribute as long as he has the background reading listed out