My ongoing project - How I want to apply AI/ML to it - Where to learn from

[tribal quiver]

Hey guys, how are you doing, hope you are doing fine.

Today I'm starting with the learning process of AI/ML and before starting I've a few questions. Before asking the questions I'd like to explain where I'm looking to use AI/ML at least for the first time, I'll probably use it in other things in a near future.

Long story short, I'm coding a TradingView Backtester. TradingView is a website that gives you access to financial data, you can visualize it, use financial indicators, etc. This TradingView Backtester works with TradingView Strategies, which is coded by the user, and what it does is: based on different conditions you code (e.g: when price goes above the last monthly high price, sell) it executes a trade, either buy or sell. This TradingView Backtester test different values/options for the strategy you want. Then collects the obtained results and continues with other values/options.

What I mentioned above is done, I already have this code that test different values in TradingView Strategies, but I'm now looking to apply AI/ML in order to make my backtesting more effective. What I mean by effective is choosing based on previous results, the next possible better value to test.

So I'll show you an example of what this would look like:

• Backtest a few values in this TradingView Strategy
• Now that we have the first set of results for each of the values tested, try to find some more values that could perhaps give even better values than before.

Example of the collected Databa with the first backtest (which wont be related at all with ML)

profit  |  ema_lenght  |  rsi_lenght
10      |  50          |  3
-12     |  25          |  5
33      |  80          |  10
45      |  28          |  3
1       |  14          |  27

• I think that up to here it all makes sense. If something doesn't, please tell me and I'll rephrase.

So as I explained before, my idea is to wisely/effectively generate next possible set of values (ema_lenght, rsi_lenght) and then obviously backtest them in TradingView. So, knowing that I've never ever worked with ML/AI and I know absolutely nothing about it (I only know it can do awesome stuff 😄), what should I learn, and where should I learn this from?

Last question, how much time do you think it's needed in order for me to start applying some of my knowledge to this project?

If you read up to here, thanks a lot 💙

I hope this is the right place to ask btw

ChatGPT says that was a clear text so hopefully u guys understand it 😄

[night acorn]

@tribal quiver You can't implement AI into Tradingview's existing backtester. If your goal is to make AI finance strategies and backtest them, source your data from online, or just get it for free in the Finance AI competition we're holding right now. If you'd like to backtest your strategies you're going to most likely need to understand python and then use this to backtest AI strategies:

https://www.backtrader.com/

If you don't understand AI yet, I recommend practicing with a simpler project to learn the basics and learning a library to build AI with (Tensorflow and/or Pytorch).

[tribal quiver]

Im not saying it is because Im the one that doesnt know, but it seems really plausible to get AI to create new sets of values based on data I have.

I dont want AI to create strategies, for the moment, I just want it to give me, as I said before, a set of different possible values that could give a good net profit based on the values I already backtested. Makes sense?

Or perhaps (idk if its too difficult), maybe Ill try to do this when Im more advanced on AI, but giving the AI a really small range of values to test, then based on the results it gets on the go, it tries new values.

But thats for the future

And I do understand Python (or at least thats what I think 🤣 )

[tribal quiver]

@night acorn

[night acorn]

You don't need AI, you need bayesian optimization and as I said before, you can't use AI in pinescript, therefore tradingview is irrelevant.

[tribal quiver]

I know you cant use AI in pine, I just mentioned it for people to know, thought that maybe some context was good. Thanks btw 🙂

[night acorn]

Your goal still isn't clear

[tribal quiver]

rn, my goal is as I explained before, predict possible combinations of values that can give better results than before

For example these last two days I tested 16200 combinations in a strategy, so now I've a database with the values plugged into the strategy + the results - ok now with that, create more combinations that can give better results

[tribal quiver]

idk it seems pretty clear to me the explanation but I might be explaining something wrong. Do you understand this?

[night acorn]

so you're coding raw in python and nothing to do with tradingview right

If its just the EMA length and RSI, it's a bad idea to overfit to the past

But if you really want to do it

[night acorn]

https://kernc.github.io/backtesting.py/doc/backtesting/#gsc.tab=0

Just use this

And here's a quickstart guide that shows you how to do it and then auto optimize the EMA length

https://kernc.github.io/backtesting.py/doc/examples/Quick Start User Guide.html

[tribal quiver]

yup, just python

The script is coded in pine, I access tradingview but with python, and in there I change the settings of the indicator/strategy

I've no idea how to use pine so

[tribal quiver]

not that

[night acorn]

Ok to quote you

So as I explained before, my idea is to wisely/effectively generate next possible set of values (ema_lenght, rsi_lenght) and then obviously backtest them in TradingView. So, knowing that I've never ever worked with ML/AI and I know absolutely nothing about it (I only know it can do awesome stuff 😄), what should I learn, and where should I learn this from?

You should learn backtester and tensorflow

If you go to youtube and just look up backtester and tensorflow, you'll get better guides than anywhere else

[tribal quiver]

but backtester cant work with trading view !!!

[glossy heart]

Hi, nice to meet you.