Menu
Sister Sites

Options

Bookmark Thread

First Page First Unread Last Page Last Post

Print Thread

Similar Threads

Machine Learning with algoTraderJo 939 replies

Algorithmic Quant Trading (Machine Learning + Stat-Arb) 25 replies

Machine Learning + Retail Forex = Profitable? (Quant) 1 reply

Potential new machine learning style software. 79 replies

My most recent advancements into machine learning 16 replies

Attachments: Ideas for Data & applications of Machine Learning with Azure ML
Exit Attachments

Ideas for Data & applications of Machine Learning with Azure ML

Absolutely, it's possible to try different hyperparameters of the network. This is what provided in h2o.deeplearning function:

Inserted Code
   activation = c("Tanh", "TanhWithDropout", "Rectifier", "RectifierWithDropout", "Maxout", "MaxoutWithDropout"),
   loss = c("Automatic", "CrossEntropy", "Quadratic", "Huber", "Absolute", "Quantile"),
   distribution = c("AUTO", "bernoulli", "multinomial", "gaussian", "poisson", "gamma", "tweedie", "laplace", "quantile", "huber")

I am not an expert in deep learning yet ... may be I am becoming one by doing research and discussing here ... however in my opinion the best way is simply to try different options. h2o has quite efficient code and training for my dataset typically takes about 4-5 minutes using 4 cores so what I typically would do is to write a for loop to try different options and record results of tests.

The way I do test is a bit specific (see above -1/+1 index). What it means is that model is used to perform predictions of price change on un-seen observations and provided that we know the ground truth it's a perfect way to estimate a simplified trading strategy: virtually buy or sell whatever model suggests and compare it with real results... do index is derived by dividing obtained model generated trades results by maximum possible results from the trades assumed we knew the ground truth... for example:

  1. absolute possible maximum gain is 14000 units. Model generated trades results in -7000 units then index is - -0.5 (bad)
  2. idem for +7000 units will result in 0.5 (good)

It does not provide exact results nor accounting for spreads however it's a fast way to evaluate each model and choose the best sets of parameters... Additionally pipes in R tidyverse package makes this as easy as to eat a piece of ...

What type of implementation do you have?

Hi Excelx,

I just read your post, played a little around with Azure ML and MT4 in the last months in my leisure and this is what I have so far:

  1. Communicate in realtime (via C++ library) from MT4 EA, Indicator ... to Microsoft Azure ML Web Service
  2. I use tickdata from Tickstory - importet to MT4, written to mySql database via MT4 Indicator through the strategy tester and ported to MS Azure via database export (csv-file)
  3. In my first try I made the mistake with the shift and - of course - if I make a dataset with +-1 from the actually candle or time series or whatever I got pretty good results in the matrix (becaus ML predicts the past) but it did not work
  4. Several tries without relevant accuracy in the confusion matrix (all indicators with different settings and so on)

My thoughts about it were:

  1. Use Tickstory data to create the dataset
  2. Communication via MT4 EA in realtime to MS Azure
  3. Get a relevant accuracy in the confusion matrix for prediction
  4. Test how long the prediction will work after the "testdata" (e. g. 1 day)
  5. Switch from MS Azure Web-Service to "retraining Web-Service" to hold the good accuracy

At the moment I did not now exactly how the "Retraining Web-Service" will work as I did not get the good results you have in the confusion matrix. Maybe I have to enhance the dll to retrain the web-service but if your setup can give the results of over 90% for a short period after the "training data" it might give these good results on a retraining web-service continously.

If you like to share your setup (indicators/MS Azure ML), maybe we can give it a try with the tickdata and a retraining web service?

A few additional questions:

  1. In your Image 1 it looks like there are no counts for either up nor down candles - how you handled the "nothing" ones?
  2. Did you get this kind of accuracy with data of time series now and "+-1" for the future time series to predict (sorry about this question - I like to ask because my results looked like your discriptions as I missed to take +-1 for the next candle in the first run)?
  3. You said "unknown data is good as random". What are the results of ML prediction - is the probability over 90% and will not work or does ML predicts something like "Up" with probability "20%" continously?


Regards,
Martin

Quoting Martin79
Hi Excelx, I just read your post, played a little around with Azure ML and MT4 in the last months in my leisure and this is what I have so far: Communicate in realtime (via C++ library) from MT4 EA, Indicator ... to Microsoft Azure ML Web Service I use tickdata from Tickstory - importet to MT4, written to mySql database via MT4 Indicator through the strategy tester and ported to MS Azure via database export (csv-file) In my first try I made the mistake with the shift and - of course - if I make a dataset with +-1 from the actually candle...
Can you share the MT4 EA for realtime communication between MS Azure ?Or can you share with private message that would be also very helpful.
Quoting patara
{quote} Can you share the MT4 EA for realtime communication between MS Azure ?Or can you share with private message that would be also very helpful.
How can I send private messages? My profile says: "Private Messages Restricted Per Member Status".
I see... well can you upload it here?https://expirebox.com it says'' your files will be automatically deleted after 2 days, it will no longer be downloaded''
  • Post #26
  • Quote
  • Edited at 3:59am Jul 24, 2018 3:42am | Edited at 3:59am
Quoting Excelx
What is more interesting is that if i test it with known data, it works perfectly. it doesn't get anything wrong, so the problem is not in the input or output, I actually thought it was perfect for a second until i realized it was known data i was back-testing with, but as soon as i test with unknown data it's as good as random. So please.. does anyone have any perspective on it and can help me figure out what is wrong with it.
Hi Excelx,

I use machine learning (ML) for my extreme scalping strategy with success. I have played a lot with different ML paradigms for Forex trading. So I have some experience with ML based Forex trading, but I am not going to tell about myself in this reply.

In a nutshell, one major reason for the problem you are describing is that your ML model overfits to the training data such that it cannot generalize the model to predict better than random choices on unseen data. In general it is a very hard task for any ML model to learn a random walk with short term trends, as Forex market is known to be. My advice is start with D1 data which represents more trend in price and hence makes it a bit easier for traditional ML paradigms to learn a generalized model of the market. And do NOT forget to split your known data (training data) into two subsets of train/validation to perform a cross-validation.

Best,
Matt
Trading is the hardest way to make easy money...
Quoting MathTrader7
{quote}I use machine learning (ML) for my extreme scalping strategy with success. I have played a lot with different ML paradigms for Forex trading. So I have some experience with ML based Forex trading
Hi Matt,
did you automated the process?
Regards,
Martin
  • Post #28
  • Quote
  • Edited at 8:01am Jul 25, 2018 7:36am | Edited at 8:01am
Quoting Martin79
{quote} Hi Matt, did you automated the process? Regards, Martin
Hi Martin79,

Yes, it is fully automated running on my VPS. I employ mt4R to use ML packages of R language with MQL4.

Best,
Matt
Trading is the hardest way to make easy money...
Quoting MathTrader7
{quote} Hi Martin79, Yes, it is fully automated running on my VPS. I employ mt4R to use ML packages of R language with MQL4. Best, Matt
Hi Matt,
Thank you for the link!
Regards,
Martin
Since İ am not a coder yet ,does anyone know are these codes:. https://github.com/T-Umezaki/ML_FX comunicate realtime with Azure machine learning web service and Metatrader 4 ?
Quoting Martin79
Hi Excelx, I just read your post, played a little around with Azure ML and MT4 in the last months in my leisure and this is what I have so far: Communicate in realtime (via C++ library) from MT4 EA, Indicator ... to Microsoft Azure ML Web Service I use tickdata from Tickstory - importet to MT4, written to mySql database via MT4 Indicator through the strategy tester and ported to MS Azure via database export (csv-file) In my first try I made the mistake with the shift and - of course - if I make a dataset with +-1 from the actually candle or time...
Hi,

Have you managed to switch from MS Azure Web-Service to retraining web-service ?
Hey excelx could i contact you for some help with my own ai trading robot pelase
I have a really good solution for the interaction between Azure ml and mt4.
Quoting sasschrist
I have a really good solution for the interaction between Azure ml and mt4.
Hi sasschrist,

Sorry for the delay, i just got back to working with Forex again. I have to put together my old material so i can carry on with my research and programming for ML. What i remember is that my biggest problem with getting the accuracy i was looking for, with unknown and real-time data seemed to be the data i was using and how i structured it.

I look forward to discuss it further and in more details as soon as i get up to date with everything i worked with in the past. I am currently looking into new tech in AI and ML, there are definitely some interesting new stuff.

I was contacted by a few people about how to integrate Azure ML with an Expert Advisor, i will share how it can be done ( although the way i did it at the time was a bit complicated for people unfamiliar with coding ). It would be appreciated an easy and practical way to do it, if anyone can share.

So, understanding that the way most of us are trying to use it, Machine Learning is just a probabilistic prediction based on historical data using a predefined ML algorithm ( which you don't need to know much about its detailed mechanics to work with ) i can say that the data you put in and how you attempt to predict data is the key point to make a decent ML model and an ML based EA.

So, my burning questions for everyone using this method and curious about using Azure ML or any other ML tool is: What data are you using ? How do you structure the data ? what are you trying to predict ? and how ?
and if you feel comfortable sharing your solutions and findings, that would be of help to improve a possible good end result.
You are welcome to PM me anytime too.

I will soon share more details and info about this. So people feel free to leave your questions and comments about it.

New ideas, observations, insights and new info will always also be welcome.

This is definitely a vast field to be explored, and one with great potential.
Could i get your info to PM you please ?
Quoting Excelx
{quote} Hi sasschrist, Sorry for the delay, i just got back to working with Forex again. I have to put together my old material so i can carry on with my research and programming for ML. What i remember is that my biggest problem with getting the accuracy i was looking for, with unknown and real-time data seemed to be the data i was using and how i structured it. I look forward to discuss it further and in more details as soon as i get up to date with everything i worked with in the past. I am currently looking into new tech in AI and ML, there...
Very interesting thread Excelx!

You may have mentioned this already, but what time horizon are you looking at for trades? I think the data you use would depend on that. If you're going intraday I'd probably do a lot of intermarket analysis and maybe implement a stat arb strategy. Longer-term would really expand your options on data. FRED and Quandl would both be great resources.
Self-sufficiency is the greatest of all wealth. - Epicurus
  • Post #37
  • Quote
  • Edited at 8:58am Dec 5, 2019 8:28am | Edited at 8:58am
Quoting Ef5
{quote} Very interesting thread Excelx! You may have mentioned this already, but what time horizon are you looking at for trades? I think the data you use would depend on that. If you're going intraday I'd probably do a lot of intermarket analysis and maybe implement a stat arb strategy. Longer-term would really expand your options on data. FRED and Quandl would both be great resources.
Hi Ef5,
I get my data from MT available values from any indicator or chart info, and i can build it on any desired time frame. So it's possible to build any term-based strategy or technical analysis based strategy on the data available.

That's what i think is my problem with getting a good result, a well built strategy. So that's mostly why i asked help here, i can build it backwards, have a strategy thought out and create a database and a form of how and what to predict from there.

That's mostly what i lack i think, a well-based strategy.
And a good knowledge of data science would also be very helpful for this.

About FRED and Quandl, i am honestly surprised i could get that kind of info, since i was always mostly focused on technical analysis. But using that kind of info we can use Machine Learning based on Fundamental Analysis, and that could be awesome by itself or even combined.
So i thank you very much for that info. I will definitely try something out with that.

So please if someone have a good idea or an strategy, we could share and combine our knowledge to build something with great potential.

Thank you for anyone who contributed so far.

And as always anyone feel free to take part if you have anything to add or take from here.
Hello again,

So like i said before I've been looking into new tech for AI and ML.
And getting a little bit out of the main topic here i found something pretty interesting that goes more into the field of AI, but if we could use it and apply it to Forex trading, i think it could really do some unexpected magic.

Although, it's very hard and complex to get it out of its main applied field and put it in trading.

I really do not have much knowledge on it and it is a pretty new technology, there's not much info on it on the web yet. and i think if i am to get into it it will take quite a while and a lot of hard studying to be qualified to bend it to my will. I might try to find more material on it and look into it deeper after i do something with what i already know, and test out my ML ideas.

I am sure someone will find interest in it too.

So here it is : https://hackernoon.com/augmented-ran...t-e0e3e765808a

Take a look at it, and share and thoughts.

What i think is: if we can have a drawing learn how to walk in a virtual space, why not have a MT robot learn how to trade profitably !?
Quoting Excelx
{quote} Hi Ef5, I get my data from MT available values from any indicator or chart info, and i can build it on any desired time frame. So it's possible to build any term-based strategy or technical analysis based strategy on the data available. That's what i think is my problem with getting a good result, a well built strategy. So that's mostly why i asked help here, i can build it backwards, have a strategy thought out and create a database and a form of how and what to predict from there. That's mostly what i lack i think, a well-based strategy....
My pleasure Excelx, glad I could help. I'm a big fan of integrating fundamentals into longer term systems and using them as a filter. There's a lot of value in that in my experience. If you want something purely technical I'd look into a momentum based strategy. The Turtle Traders or Davas might be good examples.
Self-sufficiency is the greatest of all wealth. - Epicurus
Quoting Excelx
Hello again, So like i said before I've been looking into new tech for AI and ML. And getting a little bit out of the main topic here i found something pretty interesting that goes more into the field of AI, but if we could use it and apply it to Forex trading, i think it could really do some unexpected magic. Although, it's very hard and complex to get it out of its main applied field and put it in trading. I really do not have much knowledge on it and it is a pretty new technology, there's not much info on it on the web yet. and i think if i am...
Hello Excelx, I would like to speak to you about some little details on your Stddev Zig zag thread. I've sent you a message. Can you please send me a message. So I don't distract your thread. Your prompt reply will be Much appreciated

Thank you.
0 traders viewing now