Coding with AI | Page 4 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

I Will Learn Coding in 1 Month - Give me the 80/20 of MT4 Coding 14 replies

Auto Pivot Point in MetaTrader coding assist please 10 replies

Coding Example Wanted 1 reply

Primer for coding MT4 3 replies

Overly Complex Coding and Trading Systems. 32 replies

Attachments: Coding with AI
Exit Attachments

Coding with AI

  • #61
  • Feb 6, 2025 9:45am Feb 6, 2025 9:45am
Quoting FxArt.Trader
...Can AI teach you how to drive a car? No. Can AI teach you how to fly a plane? No. Can AI teach you how to swim? No. Can AI teach you how to play music? No. If we see all similar questions that AI is not capable of as opportunities and with an open mind, they are all opportunities to make money....
Last thing I'll respond to on this thread lol

AI can teach you how to fly a plane if you know about planes. If you have flown a plane already or have a license, it could definitely instruct you on the differences between the plane you're used to flying and a newer one.

Same with the car and music scenarios. If you have your license but don’t know the small details of another car, it can fill in the gaps. If you already know how to play the piano but you want to learn a new piano piece faster, it can help!

It's a tool, so it's really about the intended purpose of the user. I use CoPilot every day in all of my projects!! Without my coding background/experience, CoPilot would be useless. I don't use it to create new ideas and scripts, I use it to address issues and do mundane tasks (like filling switch cases, copy-and-pasting code blocks, auto-completing my statements, etc).
Software Developer and Music Producer
News Fundamentals All Time Return: 1,159.1%
  • #62
  • Feb 6, 2025 11:57am Feb 6, 2025 11:57am
Quoting jblanked
{quote} Last thing I'll respond to on this thread lol AI can teach you how to fly a plane if you know about planes. If you have flown a plane already or have a license, it could definitely instruct you on the differences between the plane you're used to flying and a newer one. Same with the car and music scenarios. If you have your license but don’t know the small details of another car, it can fill in the gaps. If you already know how to play the piano but you want to learn a new piano piece faster, it can help! It's a tool, so it's really about...
Okay, my friend, no problem. I think we're stuck in a loop, and maybe because my English isn't great, I can't explain what I mean. This will be my last answer, and I won't explain it anymore after this.

If I told you right now that I've created an AI that teaches people how to fly a plane, what would your reaction be? I can guess – you'd probably show me the shortest route to a psychiatrist's office.

Something like that isn't real. Even if it *were* real, and I *had* built such an AI, I wouldn't trust it myself.

I feel like you think I'm looking at things in a childish way.

The current AI we have is the result of gathering the best experts from the best companies. An AI that's supposed to teach flying would *have* to come from the best military pilots, the best flight instructors in the military, the best private pilots, and the best private flight instructors.

Then, not just me, but everyone would trust it, and it definitely wouldn't be easy – it's a completely specialized job.

The same goes for the other examples.

If I wanted to explain that in my previous posts, I would have had to type 100 pages, and it still wouldn't have been enough.

It's a specialized job, so we have to leave it to the experts in that field. It's very simple, and I don't think it even needs to be said.

And about coding:
It's 100000000% important to have coding knowledge and be able to read the code produced by AI. You need that ability to guide the AI to the right point.

Is it possible to do this without coding knowledge? Yes, but it's very difficult, and you should expect to fail. But I recommend you don't give up and try to learn a little bit about programming. Or, find examples of the same code and give them to the AI. 90% of those codes have already been coded, and you just need to find the correct code or something close to what you want so that the AI uses the same coding logic to create the code you want.

If what you want is part of that special 10%, none of the AIs will work, and you'll have to solve the problem yourself. You can get help from AI for the coding part, but not for solving the problem.

I'm saying this based on experience. I coded one of the codes that falls into that special 10% (which I had previously asked about in this community, and the opinion of more experienced people in coding was that it couldn't be done) with the help of AI.

But the AI was literally stupid and kept messing up the code and couldn't understand. Even though I explained it to it many times that we had four conditions, it kept deleting one and couldn't solve the problem. In the end, the code successfully reached the point I wanted it to reach thanks to the solutions that came to my mind and that I coded with the help of AI.
Theory of Everything is Liquidity...https://www.youtube.com/@FxArt.Trader
  • #63
  • Feb 7, 2025 4:37am Feb 7, 2025 4:37am
Maybe you have seen my trading systems. I started coding high level complex systems long before the AI came to the scene.

Now AI basically taught me how to build my own little AI by trial and error. In the process I have learned many new things about machine learning, neural networks, different aspects of probability theory and statistics, hidden Markov model, all kinds of pattern recognition methods, efficient methods for working with large data sets by using compressions, and things like symbolic aggregate approximations and many many more things.

In the end I have found that the markets are indeed very efficient from a technical perspective and the small edges that you can find here and there are very fragile and tend to decay over time. The profitable strategies come and go and you constantly have to stay on top of the game and play catch up. Sometimes old profitable models will re-emerge, so don't dismiss the failed strategies because at some point they can become profitable again.

The market is constantly changing and every single pattern and shape that you see on the chart is as unique as the fingerprints on your hand! The patterns never repeat exactly. So you have to allow some percent of variation and error in your models otherwise they will never work. The sweet spot appears to be in the 60-80% accuracy range.
  • #64
  • Feb 7, 2025 9:45am Feb 7, 2025 9:45am
Quoting alphaomega
Maybe you have seen my trading systems. I started coding high level complex systems long before the AI came to the scene. Now AI basically taught me how to build my own little AI by trial and error. In the process I have learned many new things about machine learning, neural networks, different aspects of probability theory and statistics, hidden Markov model, all kinds of pattern recognition methods, efficient methods for working with large data sets by using compressions, and things like symbolic aggregate approximations and many many more things....
"Unfortunately, I don't remember if I've seen your trading systems before.

I'm glad people like you with coding skills are here, so we can learn something from you sometimes.

I don't know much about the things you're talking about. I only know a little about some of them, and that's just a very basic, surface-level understanding – just from hearing about them.

I don't have a lot of experience. I've just spent a little time in AI forums asking questions. I'm not sure if my opinion is correct, but I think that things like machine learning or working with AI require not only good programming skills, but also time and money.

In AI, they mostly turn everything into vectors so the GPU can calculate it, and that's expensive. It's likely that machine learning won't be free, and if you do it with regular systems, it will take a lot of time. I'm not sure if that's right, but if it is, I'd advise people who don't have strong programming skills or can't afford machine learning to not get too involved. The time and money it takes probably isn't worth it.

Of course, maybe I'm wrong. First, I think you should find a specific strategy that's consistently profitable. The amount of profit doesn't matter, what's important is that it's consistent. Then you can try new things.

I haven't really looked into the market with AI, robots, or anything like machine learning, so I don't have much of an opinion on it.

But about the market...
This market is designed so that all previous strategies both work and don't work. Patterns have to work enough for you to stay in the market, but they also shouldn't work so well that you leave the market. The market has to reward you so you stay, and sometimes it also has to take your profits. That's why you'll see a lot of people saying that everything is 50/50.

If someone has a strategy with a 60-80% win rate and they've tested it with a robot, I'm skeptical. I only trust a strategy that I can see on the chart and analyze for at least a few months and over 100 trades.

A few words about this topic...
For those who want to start coding and working with AI, I can say this: Maybe you've seen an indicator that draws something like an object with a specific color on the chart, and there's no setting in the indicator to change that color. That's where I started, and that's how I learned – just by adding a new input to an indicator so you can change the color of an object."
Theory of Everything is Liquidity...https://www.youtube.com/@FxArt.Trader
  • #65
  • Feb 11, 2025 1:29pm Feb 11, 2025 1:29pm
Some of the responses seem to mix up two concepts:

  1. LLMs: Large Language Models like ChatGPT or CoPilot "understand" written texts, and can generate pieces of source code. These could be used to generate the logic of your trading strategy.
  2. Machine Learning: Linear Regression, Random Forest, Support Vector Machine (SVM), Neural Networks, etc. are algorithms that can find complex relationships between data values. These can be used to predict future price action to use an input for your strategy.

Nowadays both are often called "AI" but these are entirely different things!

  • #66
  • Feb 12, 2025 1:44am Feb 12, 2025 1:44am
Quoting jblanked
{quote} Last thing I'll respond to on this thread lol AI can teach you how to fly a plane if you know about planes. If you have flown a plane already or have a license, it could definitely instruct you on the differences between the plane you're used to flying and a newer one. Same with the car and music scenarios. If you have your license but don’t know the small details of another car, it can fill in the gaps. If you already know how to play the piano but you want to learn a new piano piece faster, it can help! It's a tool, so it's really about...
To sum up everything you've said, AI is a tool just like a pencil. You give a pencil to a regular person and they can write. You give a pencil to a skilled artist and they will create art. Just don't forget that the artist started out as a regular person who could only just use the pencil to write. To add to your point, unfortunately many people lack any conviction to learn and understand and simply abuse and misuse AI which always backfires. But you can't simply rule out everyone that has no background on something and say they shouldn't use AI to learn. If you know how to use AI tools correctly, it doesn't matter what you don't know, you will learn and understand very quickly.
  • #67
  • Feb 14, 2025 2:36am Feb 14, 2025 2:36am
Quoting arteks
{quote} That's how it is. Not just for traders but for everyone. We pay to train the greatest enemy of humanity. Whoever invented it..........bravo.
No one and nothing have the power to hurt you. It's you who give this power to them.
And you will leave pain to take care of itself.
  • #68
  • Feb 15, 2025 5:10pm Feb 15, 2025 5:10pm
okay guys, lets continue with more useful attempts using MS windows Edge's copilot
as I said you must explain a clear trading plan to analyze and conduct with AI
then the AI will answer you as an expert code for automated trading or a trading signal!
Youtube has already good videos about this process, especially producing an EA
we can continue with our attempt too, and improve the code or approach
here we go!
  • #69
  • Feb 17, 2025 9:47pm Feb 17, 2025 9:47pm
Coding AI is not new. its always been present before the advent of Windows operating system or the internet. Its simply code written to carry out function at all times without to easy the coding of a programmer or user . it has the potential of carrying out multiple functions all at once. The only difference in the now AI , is additional Network , Sound and the basic programming have in cooperated as one entity. the AI code is been written to make it think more of its own process more resulting in a newly generated function .
Your google has AI code function embedded in it that sometimes can be annoying. Could there be a sense of smell in future, no one knows.......
  • #70
  • Feb 17, 2025 9:52pm Feb 17, 2025 9:52pm
Quoting yoriz
Some of the responses seem to mix up two concepts: LLMs: Large Language Models like ChatGPT or CoPilot "understand" written texts, and can generate pieces of source code. These could be used to generate the logic of your trading strategy. Machine Learning: Linear Regression, Random Forest, Support Vector Machine (SVM), Neural Networks, etc. are algorithms that can find complex relationships between data values. These can be used to predict future price action to use an input for your strategy. Nowadays both are often called "AI" but these are entirely...
Definitely, reflect back at university education that is translating NN etc... into strategies or embedded functions for AI applications.
  • #71
  • Feb 18, 2025 1:23am Feb 18, 2025 1:23am
Quoting FF Index
okay guys, lets continue with more useful attempts using MS windows Edge's copilot as I said you must explain a clear trading plan to analyze and conduct with AI then the AI will answer you as an expert code for automated trading or a trading signal! Youtube has already good videos about this process, especially producing an EA we can continue with our attempt too, and improve the code or approach here we go!
The tricky part is having a clear trading plan, which requires human intelligence. You can ask AI for a plan, but you need to realize that the AI's plan will be a copy of a human idea, and might not be profitable. The most important part is the initial trading plan idea itself.
Theory of Everything is Liquidity...https://www.youtube.com/@FxArt.Trader
  • #72
  • Feb 18, 2025 2:04am Feb 18, 2025 2:04am
Quoting FxArt.Trader
{quote} The tricky part is having a clear trading plan, which requires human intelligence. You can ask AI for a plan, but you need to realize that the AI's plan will be a copy of a human idea, and might not be profitable. The most important part is the initial trading plan idea itself.
"which requires human intelligence"

That's the point. The AI is everything but intelligent.
And you will leave pain to take care of itself.
  • #73
  • Feb 18, 2025 10:50am Feb 18, 2025 10:50am
just a bit off topic. i asked deepseek to make me an mt4 indi that plots the autoscroll button, zoom in and out on the chart. And he did. i like using the whole screen so i press F11 to go fullscreen on mt4. took a while because deepseek giving "server busy" but it got it right.
Attached File(s)
File Type: mq4 buttons-as-zoom 1.2.mq4   7 KB | 256 downloads
  • #74
  • Edited 12:23pm Feb 18, 2025 12:04pm | Edited 12:23pm
I've been experimenting with AI to write some MQL4 indicators, and though not perfect, it has been helpful to create for me a starting outline that is close to doing the right thing, and I can take it from there. (Good luck, non-programmers).

One request was to create an indicator that would use trendlines to mark the O,H,L,C for, say 50, "sessions" of Yearly/Monthly/Weekly/Daily. Also, to provide an option to shift those sessions forward by one session length.

For this request, and just like another indicator went, the dialogue is usually:
I ask the question and get an answer.
It won't necessarily compile, or does so with errors, so I give that feedback.
It says, "Oh I know that problem, here's a better version".
I say, ok it compiles, but I asked for... 50 sets of lines, you gave me only one set.
"Oh I know that problem, here's a better version".
Ok, it mostly works, but is highly inefficient, calculating every single line with every single tick.
"Ok, I can make it more efficient".
Yeah, it does, but though it calculates the 49 past sessions only once, it still calculated the current session with every tick, which is far less efficient than I can do.

I also noticed a major bug with Yearly lines. I tell it, "Yearly" lines don't work because you have a reference to the timeframe "12 * PERIOD_MN1" which is not an official supported timeframe.
"Ok, I can rewrite it so that Yearly lines use PERIOD_MN1 and find the session High and Low for 12 months."

Yeah, but it uses the end of this month as the end of the yearly session, rather than "January 1, YYYY". Rather than fight with AI, I fixed it myself.

There is another major bug with the DrawYearlyLines() function also. (There is a similar line in the 'DrawTimeframeLines()' function too; also needs fixing).

I asked for '50 sessions' (which in hindsight, I would have preferred to give a starting date, but still go earlier than that to get each of session data that includes the starting date).
In the code is this:
Inserted Code
   // Get enough monthly bars to cover the yearly sessions
   if(CopyRates(Symbol(), PERIOD_MN1, 0, sessions * 12 + 1, monthly_rates) < sessions * 12 + 1) return;

This is asking for 50 years of Monthly bars! Well my broker doesn't have bars back to 1975, so it returns without doing anything. What it should have done is get as many bars as it can, up to 50-sessions worth, and calculate whatever FULL sessions it can.

Some other glitches too: For the start of each session, AI looks for the current bar to match the start of the session bar *exactly*, which may or may not occur. For example, in a Monthly chart, every Month begins on the 1st. When you are on a Daily chart, there's no guarantee there is a Daily bar that starts on the 1st, rather it might be the 2nd or 3rd of the month. So, the AI version might not work properly. I implemented my own method to start a new session if a bar "crosses the starting threshold" of a new session, instead of requiring an exact match.

I customized my own version using my own private functions, but I don't mind posting the flawed 8th version of AI code (from Claude.ai) before I'd already gone off on my own to fix the bugs. By the way, by the time I got to this version, I'd used up my "free" access, and would have to wait a few hours to get more.

Next time, I will start up front with a lot more specific criteria, to include: Compile without errors or warnings; Run as efficiently as possible and don't calculate past data more than once, unless old bars change/update; I'll provide all my starting 'input' variables up front, so it builds in all the features I'll eventually want.

P.S.
I don't want to pay each AI site $20/month until I'm sure to use it often, and convinced the AI does the best compared to others (until of course they leapfrog each other). So for now, I chose to pay $30/month to just one site, "you.com" which is a collection of AI models. The UI is not great (I preferred Claude.ai to create an easy way to download the code, or to left/right arrow to review previous versions). For the fee, it says it provides unlimited access to a bunch of different AI models I can choose; it also says it does not retain my data, though I doubt I believe that entirely.

Any other multi-AI model site I should consider?


P.S.S. I did not describe all of the bugs in the AI code. Another bug: All the lines are clrBlack because its use of 'StringToColor(color_name))' returns 0 for 'DodgerBlue', but 'clrDodgerBlue' would work.
But you get the point, it does create code with bugs, then you tell it so, and it says, "Oh yeah, I know that, let me fix it".
Attached File(s)
File Type: mq4 mtf-session-lines-ver8-by-ClaudeAI-UNFIXED.mq4   11 KB | 281 downloads
  • #75
  • Feb 18, 2025 3:56pm Feb 18, 2025 3:56pm
Quoting pips4life
I've been experimenting with AI to write some MQL4 indicators, and though not perfect, it has been helpful to create for me a starting outline that is close to doing the right thing, and I can take it from there. (Good luck, non-programmers). One request was to create an indicator that would use trendlines to mark the O,H,L,C for, say 50, "sessions" of Yearly/Monthly/Weekly/Daily. Also, to provide an option to shift those sessions forward by one session length. For this request, and just like another indicator went, the dialogue is usually: I ask...
I didn't understand one thing: what exactly do you want to do?
Let's assume we're only considering the daily time frame.
For example, do you want to show the open, close, etc., for today or the current time along with the data from 50 days ago?
Or do you want the indicator to analyze all days and plot the open, close, etc., for each of those days from 50 days ago?

When working with AI, keep a few things in mind:
Try to break the code into different sections and avoid creating redundant lines and patterns. Instead, convert those patterns into a function or try to summarize them in the shortest way possible.
Encourage the AI to optimize the code and create functions for some of the repetitive sections.
For several reasons:
- You will have better and more readable code.
- Editing the code will become easier.
- You might not be interested in these reasons, but when working with AI, it's better to consider these points for a few reasons:
The length of the text that AI analyzes will decrease, which means you'll pay less for tokens.
On the other hand, when the number of tokens or text length is large, there's a higher chance of mistakes and inappropriate responses from AI.
This is my experience; that's why I try to optimize the code as much as possible to reduce the number of lines that AI has to analyze so it can focus more on the main issue.

I've attached a few images below.

Regarding different models:
I wanted to record a video, but I haven't had the opportunity for a while now.
The safest method is to use offline models for things you don't want to be widely shared.
The process is very simple.
There are several programs available, but I use this one because it’s much easier for me.
I recommend the following models: Qwen2.5 Coder 14B or Qwen2.5 Coder 32B
The models Qwen2.5 Coder 0.5, 1.5, 3, 7 billion parameters are not suitable for coding and are very weak. To receive appropriate coding responses, you should at least use models with 14 billion parameters; I realized this through research and inquiries on Discord about language models.
The models Qwen2.5 Coder 1.5,...7B will only be good if they are retrained and upgraded for a specific language.
I saw a model that has 1.5 billion parameters optimized specifically for Python, but there’s nothing available for financial markets that is free and has been trained; you need to use general models trained on many programming languages.
The model FuseO1-DeepSeekR1-Qwen2.5-Coder-32B-Preview-i1 is also good but not cost-effective for me because it's very slow for coding; it takes a long time to think and uses system resources for responses, and it's usually used when other models cannot provide correct answers.
I usually use Qwen2.5 Coder 14B because it is faster; however, if I don’t get a correct response, then I switch to a higher model.
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-02-19_00-21-27_notepad++.png Size: 88 KB
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-02-19_00-22-06_LM_Studio.png Size: 86 KB
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-02-19_00-24-49_LM_Studio.png Size: 92 KB
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-02-19_00-26-46_notepad++.png Size: 88 KB
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-02-19_00-28-34_LM_Studio.png Size: 66 KB
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-02-19_00-33-45_LM_Studio.png Size: 73 KB
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-02-19_00-35-47_LM_Studio.png Size: 69 KB


also here is the code
Attached File(s)
File Type: mq4 mtf-session-lines-ver8-by-ClaudeAI-UNFIXED_01.mq4   10 KB | 274 downloads
Theory of Everything is Liquidity...https://www.youtube.com/@FxArt.Trader
  • #76
  • Feb 18, 2025 6:01pm Feb 18, 2025 6:01pm
guys how about upload a history of price to AI like copilot and want to predict another wave of price with complete the data of uploaded file!!
  • #77
  • Feb 18, 2025 7:30pm Feb 18, 2025 7:30pm
Quoting FxArt.Trader
{quote} I didn't understand one thing: what exactly do you want to do? Let's assume we're only considering the daily time frame. For example, do you want to show the open, close, etc., for today or the current time along with the data from 50 days ago? Or do you want the indicator to analyze all days and plot the open, close, etc., for each of those days from 50 days ago? When working with AI, keep a few things in mind: Try to break the code into different sections and avoid creating redundant lines and patterns. Instead, convert those patterns...
Thanks, these are useful comments to me.

Re: what exactly do you want to do?

First, what I want looks something like this (just the lines; I've manually circled reactions of interest) :

Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 42 KB


For all session TF's, I'm only plotting OHL, and not C because C vs the next O are usually too close to each other to matter, though not always.
Blue=Daily, Green=Weekly, Orange=Monthly, Crimson=Yearly

I've circled most of the places on the chart where the *previous* session lines (that is, plotted 1 session forward), caused an observed reaction.
In a couple of places, I've marked and labelled what look like reactions to lines from 2 sessions prior.

I think these demonstrate the possible usefulness of the lines in trading, maybe just by themselves... or when combined with other signals. It would be VERY helpful to have code that watches for some sort of detection that an initial reaction has occurred, and before it gets too far away, consider taking (or suggesting) a trade in the direction of the bounce (or, a break and enter on retest). By "some sort of detection", I can think of time restrictions, like give it 60 seconds of time where it has approached but not exceeded by X pips; or, an M1/M5 pinbar (at close) must occur; or ... something else I haven't mentioned/thought of yet.

If I were to start from scratch and rewrite my indicator, I would scrap the "50 sessions" criteria. Instead, I would likely use MinDays, (or MinBars), *however*, quite unlike most interpretations of these, I want to see the correctly calculated *previous* session lines that occurred entirely before the starting point that MinDays defines.

That's confusing to read, but as an example:
If I say MinDays=28, and today is Tuesday Feb 18, 2025, then what I want would be that on that day, *Tuesday Jan 21st, every previous session line is plotted.

For each session timeframe, this means:
Daily sessions calculated 29 days (Monday Jan 20th) ago but each plotted one session forward. Therefore, starting 28 days ago, for each day I see a set of previous Daily session lines. (29 sets of Daily lines, the previous 28 days, plus 1 set for lines plotted today).

Weekly sessions: 28 days ago was Tuesday, Jan 21st. That week started Sunday Jan 19th, so I need the data for the week of Sunday Jan 12th, plotted forward to the week of Jan 19th. (5 sets of Weekly sessions lines).

Monthly: 28 days ago was Jan 21st, so for that month, I need December data plotted forward 1 month to January, then next would be January plotted this month. (2 sets of Monthly session lines).

Yearly: Clearly, 2024 lines plotted starting Jan 1st, which includes 28 days ago. (1 set of Yearly session lines).

For live trading, MinDays=3 is more than enough lines. For historical review, a much larger MinDays would be great to go back and study how well the lines work.



Re When working with AI, keep a few things in mind:
You make good points about guiding AI in parts/pieces, rather than as big chunks. I'll have to learn by more experience.

My detailed criteria above is so specific, I doubt its worth the effort to get AI to do it all, but rather just the foundation and I can tailor the specifics. (But I don't know AI well yet!)

I hadn't much considered installing and using the AI model locally, but now I might do so. I have a fast machine, and could easily boost my 32G to 64G or more.

Regarding code changes in your _01.mq4:
A single line ObjectDeleteAll( args... ) would be better, but as you appeared to already know... it is a complete waste to delete ALL sessions and re-create all again. So that code needs to take into account the OnCalculate(... prev_calculated ...) variable, to optimize calculations for usually the current bar 0 changing, but on occasion, updates to all Bars-1 bars.

After that is implemented, a much more efficient CreateOrUpdateLine(...) will make sense to use, though I would still cluster the color/style/width/ray properties to only run when ObjectCreate runs, because -- apart from changing Input vars -- they'll never change. Only the time1,2/price1,2 will change, which your code will do.

The main purpose of the indicator is to plot session OHL(C) lines *forward* by 1 session width, and since we're talking about old bar info plotted forward, every line should only be plotted one-time-only, and never need changing (apart from MT4 loaded *changed* old bars, which we certainly hope is rare; *Just remembered scrolling far back in time loads new bars, which is not rare at all).

Not mentioned before would be an additional feature to plot ONLY the current session: Open line (just once), and the High & Low lines (only when new extremes occur), and don't bother with the current session Close, which is just the current price. These High & Low lines are the only ones which might get changed prices with live data, therefore they're the only two lines that need bother with incoming new ticks. I didn't try to ask AI to do this, and probably will just write it myself.

Another bug (with previous Daily session lines) I just noticed is... with Daily lines, at the market close of Friday, it draws those lines forward 1 session width to Saturday. But since Saturday doesn't get real bars, the lines show up as either dots, or 1-bar wide. The time2 needs adjusting forward to Monday. An easy check of the DayOfWeek(time2) should do the trick.
  • #78
  • Edited 3:21am Feb 19, 2025 3:11am | Edited 3:21am
Quoting pips4life
{quote} Thanks, these are useful comments to me. Re: what exactly do you want to do? First, what I want looks something like this (just the lines; I've manually circled reactions of interest) : {image} For all session TF's, I'm only plotting OHL, and not C because C vs the next O are usually too close to each other to matter, though not always. Blue=Daily, Green=Weekly, Orange=Monthly, Crimson=Yearly I've circled most of the places on the chart where the *previous* session lines (that is, plotted 1 session forward), caused an observed reaction....
The first priority is the graphics card. My graphics card is a GTX 1080 Ti. It’s old and I bought it for 3D rendering. Now, the latest generation of budget graphics cards will be significantly better than this model, but still, please check.

In the beginning, I struggled to run the two models below with my graphics card, and sometimes the responses were so slow that it took about 15-20 minutes when I used the 32 billion model. That’s why I preferred to use the online model. The 14 million model was also relatively coded, which I later found out was due to settings. With the new settings, the speed improved to a bearable level, making it worthwhile to continue with the offline model.

Now, regarding your indicator: Your explanation about using dates was a bit confusing; however, I can guess what you want to do since I have done similar coding before. I can say that at least 80% of it can be easily coded by artificial intelligence, provided you have a plan; otherwise, the results will not be satisfactory.

To reach a clear plan, you need to ask yourself a few questions about how it will work in your mind. Will the color or other properties of the line in the indicator change later? The code I sent is for that purpose; if not, that section should be moved under the part that creates a new line. This means that later on, the line color will be set, but the price and time will always be updated.

Usually, since the time frame is high, there is no need to clear the lines, especially in this type of indicator unless in special conditions. There are three scenarios where lines need to be deleted and redrawn: when the indicator input changes (as one of the options may turn off or the session range changes), when a new time range is created, and finally when the indicator is removed from the chart.

You can consider a time range for calculations; however, new systems are strong enough that this shouldn’t be an issue and doesn’t require this condition often. In some indicators, it needs to be managed correctly, but here consider one variable for all of them—for example, updating every hour makes more sense since it’s a high time frame and doesn’t put pressure on the system. There’s a chance that an issue might arise if a line hasn’t been updated; therefore, it’s better for the indicator to update its drawings every once in a while.

Now let’s move on to calculations. Assuming that index 0 always means the current session: there needs to be a condition that if it’s session 0, then no closes should be drawn across all relevant time frames. This condition ensures that all closes are drawn except for the current close.

In this example, we’re considering a range of 50. In daily time frames, if it's not a crypto pair, we don’t have candles over the weekend. We only have two scenarios: either we’re looking at the last 50 daily candles (which changes how we code) or we’re considering 50 days including weekends. We don’t care how many days there were in each month; for instance, 50 days ago could mean 42 daily candles in one range and 43 daily candles in another due to different weekends. Of course, we can also consider how many days were in each month.

In these conditions, my coding method changes and it’s better to use the method below if there’s a need for specific timing or if you want to filter weekends in some way. The best method is to use `MqlDateTime`, which gives you complete management for this task.
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-02-19_12-28-17_metaeditor.png Size: 22 KB


There is also a solution for issues with annual time frames. If we had data from the last 50 years that follows one trend, if we didn’t have data from those 50 years but had data from 30 years instead, then it would follow another trend or simply disregard it and not consider anything or take into account just those 30 years.

If I were you, I would first get the current time from `TimeCurrent()` and then transfer it to `MqlDateTime`, extract the year from it and subtract 50 years from it before checking whether there is data for that date or not. Then based on that information, I would follow through with those steps mentioned above.

I call this guiding AI to a desired point—AI doesn’t know what to do or what’s in our minds; we need to guide it along the path we want. If we don’t have an idea about how to guide it, that’s okay… there are solutions; we just need to find similar code samples so AI can use their logic with minor adjustments for our code.

I tested this approach myself; I had previously created a dashboard but it had bugs and my current experience told me that code likely had many issues. I searched on Google and found two codes which I provided to AI and asked it to write a dashboard that gives a buy signal if the price is above average and a sell signal if it's below average. This was really done easily as I honestly didn’t have the patience or ability to recode everything again. However, I reached the code I wanted and even improved upon it!

If I were to write such code, I would 100% use AI.
For example, if I only want to change the calculation section, I would tell it to only change the following functions and only show me those:
Inserted Code
OnCalculate()
DrawYearlyLines()
DrawTimeframeLines(int timeframe, string prefix, color color_name)

Why? For several reasons:

  1. The code length is reduced.
  2. When the code length is large, it might change some functions and mess up the code.
  3. Besides, the answer will be faster.
  4. Also, the usage of free tokens is reduced, and I have a better chance of getting more answers.

You should also tell it how the output answers should be in the chats and manage it as much as possible.

Theory of Everything is Liquidity...https://www.youtube.com/@FxArt.Trader
  • #79
  • Feb 19, 2025 6:28am Feb 19, 2025 6:28am
Hello friend how are you doing today hope all those projects are coming up fine as you desired seems you are super busy these days however i read through some pages here and used the chatgpt link you gave here to create an ea that can work on 4hrs chart using the arrow signal from the attached LWMA The input setting for the arrow was 5FAST AND 5 SLOW but the setting i used in generating the EA is LWMA 1 FAST AND LWMA SLOW 2, i think the chatgpt did it the other way round by using 1 as SLOW AND 2 AS FAST please when you are a bit free kindly help in checking it out so that it can work as expected. You are kool friend please dont forget me i wont trade with it until you touch it with your expertise. The ea is even yet to generate a buy or sell i dont know why and i dont know how to tweak it.
Attached Image (click to enlarge)
Click to Enlarge Name: FOR FX ART.png Size: 61 KB
Attached File(s)
File Type: mq4 LWMA_Crossover_EA.mq4   2 KB | 268 downloads
File Type: mq4 LWMA-Crossover_Signal.mq4   4 KB | 269 downloads
  • #80
  • Feb 22, 2025 8:57am Feb 22, 2025 8:57am
Quoting Slingshots1
Hello friend how are you doing today hope all those projects are coming up fine as you desired seems you are super busy these days however i read through some pages here and used the chatgpt link you gave here to create an ea that can work on 4hrs chart using the arrow signal from the attached LWMA The input setting for the arrow was 5FAST AND 5 SLOW but the setting i used in generating the EA is LWMA 1 FAST AND LWMA SLOW 2, i think the chatgpt did it the other way round by using 1 as SLOW AND 2 AS FAST please when you are a bit free kindly help...
Hello,
Thanks for asking about my well-being; I'm doing well and hope you are too.
You can easily check that expert advisor. Just place two moving averages on the chart so everything matches the settings of the EA, then run it in the tester like in the picture below. If there's an issue or it doesn't work exactly as you want, go to ChatGPT and explain the problem so they can fix the code.
In my MetaTrader, it works perfectly without any changes.
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-02-22_18-10-34_terminal.png Size: 35 KB

Attached Image (click to enlarge)
Click to Enlarge Name: 2025-02-22_18-26-54_terminal.png Size: 40 KB
Theory of Everything is Liquidity...https://www.youtube.com/@FxArt.Trader
0 traders viewing now