SMCJB 

Would depend what algo you are trying to train but consider instead of using RSI maybe use (Is RSI < 20) as a binary input. While I can't claim to have great results my results did seem to improve when I did things like that.

I am sure that works(especially given the last decades data) because it is basically a large % drop over a short period of time feature.

You could just calculate that directly though as a feature. A rolling 2 week window with a certain threshold -% return as a binary input. No one would call that a technical indicator though. It is just feature engineering from the data.

I already got bored with the astronomy class lol. Just not worth studying over De Prado's book.

Agreed.

I only used RSI as an example. You can obviously make your features anything you want. (C > 200MA, Inside Day, New 20 Day High etc). I should have made the point that sometimes binary feature like that can be more successful than continuous ones as the continuous can involve so much noise.

rleplae 

as soon as possible i will drop a draft version in a PM for your review as agreed later the work will become public

Has this become public yet? If so, can you point to a link?

srgtroy 

Has this become public yet? If so, can you point to a link?

Yes it was published
Please find the work attached