[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/biz/ - Business & Finance

View post   
File: 134 KB, 1500x1176, 1696178290376122.jpg [View same] [iqdb] [saucenao] [google]
57508213 No.57508213 [Reply] [Original]

What indicators should I use to train my crypto trading neural net?

>> No.57508271

>>57508213
whether the tip of my penis is leaking cum or piss

>> No.57508292

>>57508271
I don't think I have access to that data from the charts, unfortunately.

To be clear, I don't even need to know how to use each of the indicators, I just need to know your opinion on the most reliable ones, back propagation will do the rest. Another thing to consider would be the trading period, because of how volatile crypto has been. Maybe I shouldn't use data from 2022 for instance.

>> No.57508557

>>57508213
You should train your bot to invest for dividends and then reinvest them into things that pay dividends.

>> No.57508740

>>57508213
>What indicators should I use to train my crypto trading neural net?

Can you give it a list of candle chart formations and then have the AI over lay those on charts at different time frames to see if they make close matches?

You wouldn't be using data or indicators. Just having the AI literally match shapes and patterns and give an estimate of how close it is based on a percentage.

I.E. you show it what Wyckoff accumulation looks like or something or Double Bottom and then it tries to find crypto on varying time frames that match it?

Does that make sense. I have no idea how well it would work but it would certainly be interesting or fun to try if possible.

>> No.57508746

>>57508213
>>57508740

You would be showing it pictures of textbook examples and looking for as close matches as possible essentially.

>> No.57508752
File: 300 KB, 1024x1024, zLx7RMsC3Y.jpg [View same] [iqdb] [saucenao] [google]
57508752

>>57508740
>sure that makes total sense it would be interesting to teach an ai to identify patterns and match them to historical data we could try identifying wyckoff accumulations which are patterns that may indicate a shift in market sentiment and see if we can use the ai to identify them in real-time it definitely has potential to be useful in crypto trading and i'm sure 4channers would have some creative ideas on how to use it we could try experimenting with it to see what comes out of it just remember to keep it subtle with the trolling

>> No.57508759

>>57508740
Yeah, that could certainly be done, I'll note the idea down, I will have to think about how it would work across different timeframes etc. I feel like I would need actual deep knowledge on the topic to understand how to best implement this, but the comparisons themselves are very doable. Maybe I'm wrong, and I just need to lift a bunch of patters from some trading book or something.

This would be more of a heuristic based approach, but I could implement training on the leeway it gives to pattern matching, and maybe add the aforementioned indicators to the mix.

>> No.57508767

>>57508213
You should try to build some kind of sentiment analysis on the online discussion. You will quickly find out that it's not the neural network part that is difficult but enriching the data.

>> No.57508825

>>57508752
>>57508752

Is this ChatGPT? Lmao What is going on here with this shit I'm seeing it everywhere

>> No.57508827

>>57508767
No external sources to the graphs (for now).

>> No.57508838

>>57508759

Basically I want an AI to do this for me on various time frames checking the top 20-50 crypto assets in MCAP. If you build this I'd love to test it. I'd give feed back at least to help improve it.

>> No.57508866

>>57508759
>Maybe I'm wrong, and I just need to lift a bunch of patters from some trading book or something.

Seriously was my initial thoughts but Idk enough about AI to know how effective that would be

>>57508759
>but I could implement training on the leeway it gives to pattern matching

This is key as no formation is ever as a textbook example but sometimes they are damn near close. Like 90-92%%. I'd say on average it would look like 70-75% similar to the text book examples.

>> No.57508917

>>57508827
Yeah, I understand. But my point is that the pattern matching thing has most likely been done many times by actual experts; I think that there is not much edge to be found there.

>> No.57508918

>>57508838
>checking the top 20-50 crypto assets in MCAP.
What does that measure, volume? It is unclear to me how to use volume for trading because again I don't know much about the topic. I am also unsure if I can get historical data regarding volume, but I think I can.

Basically If you can't get historical data, then your only tool becomes heuristics, and to build heuristics you need expertise.

>>57508866
The tricky part about this specific kind of pattern matching is that patterns can be wider, narrower, etc. Its trivial to create a neural network to identify letters, but a little less so to create a neural network to identify a letter that can be partially cut, stretched, and off to one corner of the image. With graphs it would be even worst because of the continuous nature. I would have to find some kind of normalization that would give me sane results in that respect.

It's probably a solved problem, though, I just have to read some shit. I think it's a good idea.

>> No.57508952

>>57508213
what type of neural net? i don't think regression and decision trees are technically neural nets are they?

>> No.57508987

>>57508952
>what type of neural net?
The basic type.
>i don't think regression and decision trees are technically neural nets are they?
No, that would be a heuristic machine.

>> No.57509018

>>57508987
heuristics can not learn from data. but those methods i mentioned can.

>> No.57509039

>>57509018
Then I don't know what you consider to be a decision tree. I'm not using trees.

>> No.57509168

>>57509039
there are 9 types of neural networks. which one do you think yours is best described by?

Perceptron
Feed Forward Neural Network
Multilayer Perceptron
Convolutional Neural Network
Radial Basis Functional Neural Network
Recurrent Neural Network
LSTM – Long Short-Term Memory
Sequence to Sequence Models
Modular Neural Network

>> No.57509308

>>57509168
It's a multilayer perceptron.