Machine Learning with algoTraderJo | Page 46 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

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

Forex with Machine Learning Software Project 1 reply

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: Machine Learning with algoTraderJo
Exit Attachments

Machine Learning with algoTraderJo

  • #901
  • Jul 12, 2017 12:40am Jul 12, 2017 12:40am
Sorry - ignore me - it was obviously too late at night when I was posting!

So - the drawdown on the charts isn't the drawdown in terms of equity of the currently open position - it's a measure of the drop from the peak of the balance graph!
  • #902
  • Aug 11, 2017 4:08pm Aug 11, 2017 4:08pm
Very interesting thread. I've just started working on mining patterns based on candlestick formation from historical data and extract those that repeat often and generate function to create footprints for those patterns.
Enter Signature
  • #903
  • Aug 25, 2017 11:42am Aug 25, 2017 11:42am
Quoting R.a
Very interesting thread. I've just started working on mining patterns based on candlestick formation from historical data and extract those that repeat often and generate function to create footprints for those patterns.
Hi R.a,

I am curious about which ML platform and coding language you are using for your pattern mining for the candlesticks...?

Are you using a turn key package or building the modules yourself...?

And would you please share a little more about "creating footprints" for the candlestick patterns and how you might use them...?

In the meantime...as a novice with ML...I have played a little bit with ML and candlesticks...using Orange ML, Encog, and Deep Thought ML.

The last two did have API's to connect with MT4 for Forex trading...but neither had a good GUI to help build and test the algorithm's. They required a lot of command line operations each time you wanted to change something...so I didn't get too far with the candlestick testing with those programs.

Also...time series forecasting was not too successful for me...so I think you are on the right track to look for specific patterns in the market with the candlesticks...and not just try time series analysis by itself.

To help your candlestick research...I am including my short collection of candlestick related ML articles for your review.

I especially like Bulkowski's Candlestick website at the bottom of the list...as he provides an excellent starting place for candlestick research...including a great statistical analysis of recurring candlesticks...and you can use those stats to help determine which candlestick patterns you might want to look for with your ML algo's.

I hope they help you in your ML projects...

Take care,
Robert

CANDLESTICKS Research
Advanced Candlestick Patterns | Investopedia
Candlesticks - 5 Day Stock Market Prediction Into Future | Heaton Research
Candlesticks - Stock Market Forecast: Automated Recognition of Candlestick Patterns
Candlesticks - Stock Market Prediction Model Using Candlestick Pattern Feature
Candlesticks - NN Nirvana Club
Candlesticks Patterns - indentifying-short-term-market-turns.pdf
Candlesticks - Proceedings Template - WORD - p1533.pdf
Candlesticks - Artificial Intelligence and Soft Computing, Part I: 10th International ... - Google Books
Candlesticks Patterns - Intelligent Trading: Quantitative Candlestick Pattern Recognition (HMM, Baum Welch, and all that)
Candlesticks - Predicting Stock Prices Using Technical Analysis and Machine Learning - FULLTEXT01.pdf
Candlesticks - Fuzzy Candlestick Approach to Trade S&P CNX NIFTY 50 Index using Engulfing Patterns
'Simple Methods of Forecasting Directions of the Japanese Candlesticks - MQL4 Articles
Candlesticks Patterns - Genetic Algo Market Technicians Association
Candlesticks, Condensed - TRADERS’ TIPS - FEBRUARY 2015
Encog Candlesticks - Basic Market Forecasting with Encog Neural Networks
Bulkowski's Candlestick Patterns
  • #904
  • Aug 25, 2017 1:53pm Aug 25, 2017 1:53pm
Have you seen this :http://traders.com/Documentation/FEE...adersTips.html
  • #905
  • Aug 26, 2017 3:59pm Aug 26, 2017 3:59pm
Quoting cosmicbeing
{quote} Hi R.a, I am curious about which ML platform and coding language you are using for your pattern mining for the candlesticks...? Are you using a turn key package or building the modules yourself...? And would you please share a little more about "creating footprints" for the candlestick patterns and how you might use them...? In the meantime...as a novice with ML...I have played a little bit with ML and candlesticks...using Orange ML, Encog, and Deep Thought ML. The last two did have API's to connect with MT4 for Forex trading...but neither...
I haven't used any ML platform yet. I just started with a dukas copy and C# but had to pause it for now.

My approach basically is not to mine all patterns but based on predefined targets. I first segregate the OHLC data based on the size, something like 30-40 pips, 40-50 pips and so on.

The idea is to identify only the important events and reduce the complexity. With the targets identified and filtered (it would be very few), backtrack 'n' candles to check the build-up/formation of the target and extract all such series(those that match the direction of the 'n' prior candles ). Then each of the pattern would be compared statistically to identify how similarly close they are. The final step would be to generate the footprints that actually caused the build-up of the targets.
Enter Signature
  • #906
  • Aug 28, 2017 12:03pm Aug 28, 2017 12:03pm
Quoting R.a
{quote} I haven't used any ML platform yet. I just started with a dukas copy and C# but had to pause it for now. My approach basically is not to mine all patterns but based on predefined targets. I first segregate the OHLC data based on the size, something like 30-40 pips, 40-50 pips and so on. The idea is to identify only the important events and reduce the complexity. With the targets identified and filtered (it would be very few), backtrack 'n' candles to check the build-up/formation of the target and extract all such series(those that match...
Hi R.a.,

Thanks for sharing your set up. I was hoping to find some Python packages for Forex trading with a good GUI. Good luck on developing your C# system and please share any updates you get excited about...!

The article posted by Olsen-yersen (thanks Olsen) had an interesting candle set up that might work for you finding your "candle footprints"...

Not sure how ML was used for finding that type of candle pattern...but the article shows the candle code for a number of trading platforms including Metatrader.

And I did find a trading system that uses that candle pattern which shows a good backtest profit curve......called The Gandalf Project Research System.
https://www.tradingview.com/script/Z...esearch-System

For comparisons of the candle code for a few trading platforms... I am just showing the CANDLE ENTRY codes below...

TradeStation Code
//******* SETUP ENTRY GANDALF PATTERN *******
if ( AveragePrice[AP1] < MedPrice[MP1] and MedPrice[MP2] <= AveragePrice[AP2] and MedBodyPrice[MBP1] <= AveragePrice[AP3] )
or ( AveragePrice[AP4] < MedPrice[MP3] and MedBodyPrice[MBP2] < MedPrice[MP4] and MedBodyPrice[MBP3] < MedBodyPrice[MBP4] ) then

MetaTrader Code
Entry Formula:
avgp:= (open + high + low + close)/4;
medp:= (high + low)/2;
medbp:= (open + close)/2;
entrySignal:=
(Ref(avgp < medp, -1) AND ref(medp,-2) <= ref(avgp,-1) AND ref(medbp,-2) <= ref(avgp, -3))
OR (ref(avgp,-1) < ref(medp, -3) AND ref(medbp,-1) < ref(medp, -2) AND ref(medbp,-1) < ref(medbp, -2));
ref(entrySignal, -1)

In some respects it looks like some basic candle code for comparing past candle sizes...

For the ML...I am assuming that a genetic algo was used...but I have not seen the original article to be sure (needs paid subscription?)...

Also thanks to algoTraderJo for this ML thread. I hope to be able to replicate his good work as others have done.

Good luck and good fun to all...!
Robert
  • #907
  • Dec 21, 2017 8:34am Dec 21, 2017 8:34am
Are there some Asikury members here? Any idea where to get a documentation of the F4 framework to get an overview?
  • #908
  • Dec 21, 2017 11:59pm Dec 21, 2017 11:59pm
Quoting algoTraderJo
This is an algo I like, it has been trading for a bit more than a year now for me: {image} It uses a two algo linear classifier ensemble. I post the csv for this system, if anyone wants to explore it more. {file}
This is my system's return over the period 1988-2017. I have still to work a bit for reducing the nasty drawdowns, but you can see this is significantly better performing than the Kantu-generated one.

Open to discussion
Attached Image (click to enlarge)
Click to Enlarge Name: a4ec2ffa835966ca96c135259cdc935a.png Size: 35 KB
πάντα δοκιμάζετε τὸ καλὸν κατέχετε
  • #909
  • Dec 23, 2017 6:26am Dec 23, 2017 6:26am
Nobody has a hint here?
πάντα δοκιμάζετε τὸ καλὸν κατέχετε
  • #910
  • Mar 3, 2018 12:03am Mar 3, 2018 12:03am
My experience with neural networks is moving averages are bad inputs choice in their raw form.I get better prediction results with precent change. Apart from parameter optimization with GA ,is there any suitable machine learning algorithm to predict the moving average crosses in their raw form?
  • #911
  • Mar 3, 2018 2:26am Mar 3, 2018 2:26am
Interesting subject, my eldest Son's maths PhD is part concerned with machine learning and A.I., interesting 'sponsor' too, illustrating how financial engineering is extending its tentacles, beyond the usual 'giant squid' suspects.
"The meek shall inherit the earth.." Yeah right, good luck with that shit..
  • #912
  • Mar 12, 2018 1:16am Mar 12, 2018 1:16am
Hi there,
I'm developing an AI-based preditcion myself (from scratch) and was wondering what the best trading strategy would be if you could predict the price 1h into the future?

Simply go long/short between peeks and maybe whait if the prediction isn't profitable enought or if it's touncertail?

Is anybody else willing to discuss some of their results (prediction accuracy, ...) design of their model, ...

What would be an good average prediction error for eg 1h, 2h, ... predictions to build an solid strategy?

Thank!
  • #913
  • Mar 15, 2018 3:20pm Mar 15, 2018 3:20pm
Fellows, machine learning isn't my forte, but the topic is potentially very interesting for me. Could you give me, please, some advice what kind of literature for beginners to read on this topic.
  • #914
  • Mar 16, 2018 1:33am Mar 16, 2018 1:33am
Quoting Flieldel
Fellows, machine learning isn't my forte, but the topic is potentially very interesting for me. Could you give me, please, some advice what kind of literature for beginners to read on this topic.
What level of beginners?
At the very least you need to know about probabilities and statistics: everything in machine learning boils down to estimating a probability distribution.
A basic understanding of linear algebra is needed too: almost immediately you need to manipulate matrices.
You can read the book by Bishop. It is available online for free. The link is on the wikipedia entry for Christopher Bishop.
Nando de Freitas made available his undergraduate courses on youtube https://www.youtube.com/user/ProfNandoDF/videos . His explains very well.
No greed. No fear. Just maths.
  • #915
  • Mar 17, 2018 2:29am Mar 17, 2018 2:29am
Additionally Andrew NG's Deep learning course on Coursera is the best basic course on the subject.Trevor Hastie and Robert Tibshirani's An Introduction to Statistical Learning book is one of the best along with Bishop's book on the subject:http://www-bcf.usc.edu/~gareth/ISL/
  • #916
  • Mar 17, 2018 2:53am Mar 17, 2018 2:53am
Quoting PipMeUp
{quote} What level of beginners? At the very least you need to know about probabilities and statistics: everything in machine learning boils down to estimating a probability distribution. A basic understanding of linear algebra is needed too: almost immediately you need to manipulate matrices. You can read the book by Bishop. It is available online for free. The link is on the wikipedia entry for Christopher Bishop. Nando de Freitas made available his undergraduate courses on youtube https://www.youtube.com/user/ProfNandoDF/videos . His...
thanks for the link. interesting video courses.

you will also need good programming skills to implement machine learning or to use the freely available softwares
  • #917
  • Mar 17, 2018 3:00am Mar 17, 2018 3:00am
Quoting TheRealDave
Hi there, I'm developing an AI-based preditcion myself (from scratch) and was wondering what the best trading strategy would be if you could predict the price 1h into the future? Simply go long/short between peeks and maybe whait if the prediction isn't profitable enought or if it's touncertail? Is anybody else willing to discuss some of their results (prediction accuracy, ...) design of their model, ... What would be an good average prediction error for eg 1h, 2h, ... predictions to build an solid strategy? Thank!
try shorter timeframe like 1m, 5m etc. you will increase the chance of accuracy

not to sound pessimistic but my own experience show that you are better off learning candlestick patterns than dwelling onto machine learning.

however my few years spent on AI was worth it as a knowledge excercise.
  • #918
  • Mar 17, 2018 3:07am Mar 17, 2018 3:07am
Quoting easyx
Are there some Asikury members here? Any idea where to get a documentation of the F4 framework to get an overview?
the founder of Azirukoy has a blog with hundreds of posts giving examples on AI. i used to ready every one of them until i concluded; if the founder failed to create a consistent profitable system that works in forward testing; why bother!

I greatly thank the founder for candidly and honestly contributing to science.
  • #919
  • May 30, 2018 10:33am May 30, 2018 10:33am
Quoting alphadude
{quote} the founder of Azirukoy has a blog with hundreds of posts giving examples on AI. i used to ready every one of them until i concluded; if the founder failed to create a consistent profitable system that works in forward testing; why bother! I greatly thank the founder for candidly and honestly contributing to science.
Making similar experiences. Testing a Backpropagation Neural Network with different parameters and layers on EURUSD using high quality historical tick data for half a year (Jan-June). The NN produces very good results already after the 2nd pass and getting amazingly better and better with every other pass. However, running the same NN with the same weights for the other half of the year (July-Dec), it already fails miserably again.
  • #920
  • May 30, 2018 10:53am May 30, 2018 10:53am
Quoting mw108
{quote} Making similar experiences. Testing a Backpropagation Neural Network with different parameters and layers on EURUSD using high quality historical tick data for half a year (Jan-June). The NN produces very good results already after the 2nd pass and getting amazingly better and better with every other pass. However, running the same NN with the same weights for the other half of the year (July-Dec), it already fails miserably again.
spot on.

my experience with AI is that the vast amount of time spent on searching for the holy grail; only for it to fail in live trading; was not worth it.

what I realized; is that the human brain is much more powerful than an AI machine. I have been using my brain to find edge in the market ever since.
3 traders viewing now