I don't understand almost anything you just did but it's amazing how you can use math and programming to create such a tool. It's defintely something I'll try to learn about when I'm free since I want to do computer science, so thank you so much for doing this and then documenting this! You're an idol.
Hey Keith, thanks for your support! :) And yeah, people spend a lot of time and money producing tools to help but the real life market is a lot more unpredictable so it seems like no one has truly figured it out. Then again, if they had, they wouldn't say a word :P
I'm a python newbie but somehow this all made sense and i could follow it along well. Nice vid man, subbed. Will you share the code on git when you're done? i'm looking to start developing runescape bots ( the good kind ) with machine learning etc, hmu if interested.
Hey! The code and data is already available on GitHub here: github.com/chriskok/GEPrediction-OSRS Also I really appreciate your comment! I'll try to keep making sense haha. All the best with your journey into Python and ML. If you ever have any questions of just want to chat, join us on discord: discord.gg/ZummSXK
@@ChronicCoder I'm learning programming and found this to be a great tutorial of sorts. You made it easy to follow. A topic really does become easier to learn/remember when you make it interesting.
Good question, I think that stands pretty true for the real life stock market but runescapes ge prices are more consistent and stable (for example, a price will never go away fully unless people arent purchasing the item). Plus there is a much more limited amount of factors that (we think) affect the prices in the ge. Im definitely still exploring it though, so we'll see! Ill gladly discuss it more if you have questions :)
From a trader/investor's standpoint, this should not replace decision making in making trades, it should assist. Predicting based on past trend is a technical analysis in trading/investing. As traders/investors we tend to do more fundamental analysis as well (market sentiments, press release, etc). The number crunching part becomes easier to analyze as it is assisted by ML.
Fantastic video! Looking to get into tensforflow after I had a go at writing a machine learning library at uni, any advice? How do you find debugging networks, for example you identified a case of overfitting, when using an 3rd party library? Great idea using the OSRS grand exchange, will be keeping an eye on this one!
Hi Jack! I really appreciate your support :) I'd HIGHLY suggest taking a look at this book if you haven't yet: amzn.to/38Ro6EO (Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow) It taught me a whole lot about all the basics in a simple and practical way. It covers debugging the networks to great detail as well and how to identify over/underfitting and what to do about them. Especially with a 3rd party library as large as this, they have lots of pre-existing functions that help a tonne with common ML problems that you'd face. So yes, I'd definitely suggest you get into it and let me know what you think! To be honest, I'm definitely still not a master at this yet too as you'll see in my next video so take what I say with a grain of salt. There really is a lot to learn in the field, but that's what makes it exciting too!
@@ChronicCoder By the way, if you need any assistance in Runescape HMU in game, name's the same as here. I'd love to play a role in your project, be it data gathering or hypothesis testing. I've also got a compsci background so I am in love with your focus and direction.
That's a great question! To be honest, ive just begun learning about ARIMA models and the more conventional stuff for time series analysis. So yeah, ill definitely try different models soon! Right off the top of your head, do you know if its easy to implement with python?
@@ChronicCoder Yeah, shouldn't be a problem. I've worked with time series forecasting for the last 7+ years, though I mainly work in R, I also use python from time to time, though mostly with NN models. I don't know if it's overkill, because the LSTM step for forecasting with multiple variables really do some good things for keras, however with the few variables and observations you have, I think you'll do just fine with an ARIMA model. You could also try some different models, data exploration, see if there's a unit root and go the ARMA way.
Youre right, I just finished reading 8 chapters of the book i promised Id read on time series analysis and it seems like the ARIMA model would be pretty good for this case but it seems like there could be alot of added value as well with ML so ill have to try out the more basic stats models and compare them soon! My next video is focused on feature engineering and extraction so maybe the video after that will be on testing the different models. Either way, thanks for your suggestion!
Great video! just wondering do you have a degree in CS, or data science? i'm using this video as a reference for my machine intelligence course in a paper our group is writing. i hope you don't mind :)
Hey! Just to confirm, did you mean you were disappointed about it not being applied to RS3 or not having an application interface for the model in osrs?
Hey im trying to make a better price predictor. It would be "daytrading" thingy checking the price for like 2 weeks every few minutes that way it will know the times in the day when price goes up and when the mirgins are best. Im doing that in excel right now and its too hard, wanna team up and build something together?
Hey! I've tested the waters of day trading or algo trading before but it's personally not my cup of tea. If you aren't already, check out the discord (link should be in the description). There are tons of people there who are curious about that same premise :)
Absolutely! If you think you'll love anything, you should try it out. Even if you hate it, you'll at least have the time and space for something better in your life *shrugs*
Yeah, Im from Malaysia right next door :) are you from south east asia too? Ive found that we are much more attuned to singling out our accents haha. Also, i love your username, mr soviet onion.
@@ChronicCoder Thank you :p My first quess was actually you are Malaysian, I second guessed myself because your accent was not really obvious, sort of like a cross of A Malaysian who has lived in the US a long time, or from a IB institute in Singapore, and yeah I am from SEA, Burma to be precise, but lived and studied in Malaysia for 20 years actually. Anyways It's nice to see some familiarity from somewhere I have lived a long time - and based on RS, merching and coding haha. Good luck in all your endeavors man!
Thanks man! Yeah ive been living in the states for about 3 years so that explains the strange mix of accent haha. Im having a lot of fun doing this stuff, glad you all find it interesting!!
Hi. I was planning on doing a very similar thing as part of some school work I have to do. If you could get in touch with me I would really appreciate it as I would really appreciate it if I could ask for help if I get stuck Thanks. And it's a really cool video.
That's awesome! The easiest way to get to me would be through discord. Join our channel: discord.gg/ZummSXK (people there would love to answer your questions too) or hit me up personally at ChronicCoder#1667
Amazing video mate, I actually work in this area on mechanical engineering estimating Remaining Useful Life for machine components. Your improvement analysis was very well done, but I would add an extra bit of overlooked information which is 'Seasonality'. People tend to play more, ergo buy more, on weekends and holidays. You could statistically add that information for long holidays and weekends as a categorical label according to the Timestamp. Also, you could try with GRU units which are simpler than LSTM, but better than simple RNN. Finally, since you have lots of data that might be formulated into time window data, you could try to extract weekly information with a CNN input layers which will extract better information than feature engineering in this case that the data is non-stationary. If you want to be more extreme you could use state-of-the-art CNN-LSTM (implemented on TF2 in tensorflow.keras.layers.) I would love to discuss this more since i love this game since the old days and i love machine learning as well. Hit me up for discord if you like!
Hello, Mr Nacho! Thanks for your support and suggestions! It means a lot coming from someone who is actively in the machine learning field. I just got back from a long holiday but I will definitely play around a lot more with the different features and update everyone soon. I also never heard of using a CNN to extract information, that's genius haha. Let me know what your name is on discord and I'll definitely hit you up :)
Actually, if you don't want your discord out there to the public, feel free to add me ChronicCoder#1667 or join the discord server discord.gg/ZummSXK (if you aren't yet)
I don't understand almost anything you just did but it's amazing how you can use math and programming to create such a tool. It's defintely something I'll try to learn about when I'm free since I want to do computer science, so thank you so much for doing this and then documenting this!
You're an idol.
Very informative and great video, nicely done!
Thanks for your support!!
Was looking for quant videos, and got to this! Love your RS financial insights, keep up m8!
Really interesting since I did some data science at my university... Keep it up :D
Thanks!! Glad you liked it :)
Great video, and great example! More please!
Fundamental fintech, what a great analogy to explain it!
Could only fathom the arduous amount of ML logics and data collection to make modern day AI asisted cfd derivatives with these concept.
Hey Keith, thanks for your support! :) And yeah, people spend a lot of time and money producing tools to help but the real life market is a lot more unpredictable so it seems like no one has truly figured it out. Then again, if they had, they wouldn't say a word :P
I'm a python newbie but somehow this all made sense and i could follow it along well. Nice vid man, subbed. Will you share the code on git when you're done? i'm looking to start developing runescape bots ( the good kind ) with machine learning etc, hmu if interested.
Hey! The code and data is already available on GitHub here: github.com/chriskok/GEPrediction-OSRS
Also I really appreciate your comment! I'll try to keep making sense haha. All the best with your journey into Python and ML. If you ever have any questions of just want to chat, join us on discord: discord.gg/ZummSXK
@@ChronicCoder I'm learning programming and found this to be a great tutorial of sorts. You made it easy to follow. A topic really does become easier to learn/remember when you make it interesting.
Thank you!! That's definitely the hope. I think I have a long way to go but I'm gonna keep getting better :) all the best on your programming journey!
Isn't trying to predict the market based on past trends a fools' gambit?
Good question, I think that stands pretty true for the real life stock market but runescapes ge prices are more consistent and stable (for example, a price will never go away fully unless people arent purchasing the item). Plus there is a much more limited amount of factors that (we think) affect the prices in the ge. Im definitely still exploring it though, so we'll see! Ill gladly discuss it more if you have questions :)
From a trader/investor's standpoint, this should not replace decision making in making trades, it should assist. Predicting based on past trend is a technical analysis in trading/investing. As traders/investors we tend to do more fundamental analysis as well (market sentiments, press release, etc). The number crunching part becomes easier to analyze as it is assisted by ML.
Fantastic video! Looking to get into tensforflow after I had a go at writing a machine learning library at uni, any advice? How do you find debugging networks, for example you identified a case of overfitting, when using an 3rd party library?
Great idea using the OSRS grand exchange, will be keeping an eye on this one!
Hi Jack! I really appreciate your support :)
I'd HIGHLY suggest taking a look at this book if you haven't yet: amzn.to/38Ro6EO (Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow)
It taught me a whole lot about all the basics in a simple and practical way. It covers debugging the networks to great detail as well and how to identify over/underfitting and what to do about them. Especially with a 3rd party library as large as this, they have lots of pre-existing functions that help a tonne with common ML problems that you'd face. So yes, I'd definitely suggest you get into it and let me know what you think!
To be honest, I'm definitely still not a master at this yet too as you'll see in my next video so take what I say with a grain of salt. There really is a lot to learn in the field, but that's what makes it exciting too!
I don’t know if you still read these comments, but this is an AMAZING video! I really hope this channel grows further.
I definitely do, and I really appreciate it! :)
ARE THERE A PYTHON LIBRARY FOR OSRS ?
新加坡？ Awesome stuff my guy. Thanks for sharing.
Malaysian, actually! Are you from Singapore/SEA? So far only people from this region point it out cause I think it's a pretty subtle accent ;)
@@ChronicCoder Yes, I think all Malaysians/Singaporeans will 'sense' it, we've got an aura. Its so cool what you're doing, keep it up.
@@ChronicCoder By the way, if you need any assistance in Runescape HMU in game, name's the same as here. I'd love to play a role in your project, be it data gathering or hypothesis testing. I've also got a compsci background so I am in love with your focus and direction.
Love to see it!
Thanks! :)
what about sentiment analysis?
Isn't a neural network quite overkill? Don't you think you'll get a better fit with a simple ARIMA model?
That's a great question! To be honest, ive just begun learning about ARIMA models and the more conventional stuff for time series analysis. So yeah, ill definitely try different models soon! Right off the top of your head, do you know if its easy to implement with python?
@@ChronicCoder Yeah, shouldn't be a problem. I've worked with time series forecasting for the last 7+ years, though I mainly work in R, I also use python from time to time, though mostly with NN models. I don't know if it's overkill, because the LSTM step for forecasting with multiple variables really do some good things for keras, however with the few variables and observations you have, I think you'll do just fine with an ARIMA model. You could also try some different models, data exploration, see if there's a unit root and go the ARMA way.
Youre right, I just finished reading 8 chapters of the book i promised Id read on time series analysis and it seems like the ARIMA model would be pretty good for this case but it seems like there could be alot of added value as well with ML so ill have to try out the more basic stats models and compare them soon! My next video is focused on feature engineering and extraction so maybe the video after that will be on testing the different models. Either way, thanks for your suggestion!
@@ChronicCoder Cool! well if you need some help be more than welcome to write :D
Great video! just wondering do you have a degree in CS, or data science? i'm using this video as a reference for my machine intelligence course in a paper our group is writing. i hope you don't mind :)
Ooh I'm honored! I have a bachelor's in CS with a concentration in machine intelligence. Please keep me updated on the paper! :D
Great videos!
Hey! Just to confirm, did you mean you were disappointed about it not being applied to RS3 or not having an application interface for the model in osrs?
@@ChronicCoder i just didnt know about the additional videos that followed. My bad.
Ooh okay! Hope you like it :) just a quick disclaimer, the website is down again since the 8th and I'm not sure why but I'm trying to get it back up!
stonks
Nice!
Glad you liked it :) more to come!
Hey im trying to make a better price predictor. It would be "daytrading" thingy checking the price for like 2 weeks every few minutes
that way it will know the times in the day when price goes up and when the mirgins are best. Im doing that in excel right now and its too hard, wanna team up and build something together?
Hey! I've tested the waters of day trading or algo trading before but it's personally not my cup of tea. If you aren't already, check out the discord (link should be in the description). There are tons of people there who are curious about that same premise :)
Haha love the rick roll internet tabs..
;) never gonna let you down
I would love to make something like this yet I have no knowledge or experience of any coding or programming at all, I should start maybe
Absolutely! If you think you'll love anything, you should try it out. Even if you hate it, you'll at least have the time and space for something better in your life *shrugs*
Haha I love your folders in Chrome
Thank you ;) hahah
Make a stock market bot for irl pls
That's the plan ;) today GP, tomorrow USD
Are u Singaporean?
Close! I'm Malaysian :) are you Singaporean? Most of the time it's only the people from this region who can distinguish our accents hahah
Missed being Rick Rolled, thanks you /s
Never gonna let you down ;)
CK, you have a very slight Singaporean accent
Yeah, Im from Malaysia right next door :) are you from south east asia too? Ive found that we are much more attuned to singling out our accents haha. Also, i love your username, mr soviet onion.
@@ChronicCoder Thank you :p My first quess was actually you are Malaysian, I second guessed myself because your accent was not really obvious, sort of like a cross of A Malaysian who has lived in the US a long time, or from a IB institute in Singapore, and yeah I am from SEA, Burma to be precise, but lived and studied in Malaysia for 20 years actually. Anyways It's nice to see some familiarity from somewhere I have lived a long time - and based on RS, merching and coding haha. Good luck in all your endeavors man!
Thanks man! Yeah ive been living in the states for about 3 years so that explains the strange mix of accent haha. Im having a lot of fun doing this stuff, glad you all find it interesting!!
Hi.
I was planning on doing a very similar thing as part of some school work I have to do.
If you could get in touch with me I would really appreciate it as I would really appreciate it if I could ask for help if I get stuck
Thanks. And it's a really cool video.
That's awesome! The easiest way to get to me would be through discord. Join our channel: discord.gg/ZummSXK (people there would love to answer your questions too) or hit me up personally at ChronicCoder#1667
Amazing video mate, I actually work in this area on mechanical engineering estimating Remaining Useful Life for machine components.
Your improvement analysis was very well done, but I would add an extra bit of overlooked information which is 'Seasonality'. People tend to play more, ergo buy more, on weekends and holidays. You could statistically add that information for long holidays and weekends as a categorical label according to the Timestamp.
Also, you could try with GRU units which are simpler than LSTM, but better than simple RNN.
Finally, since you have lots of data that might be formulated into time window data, you could try to extract weekly information with a CNN input layers which will extract better information than feature engineering in this case that the data is non-stationary. If you want to be more extreme you could use state-of-the-art CNN-LSTM (implemented on TF2 in tensorflow.keras.layers.)
I would love to discuss this more since i love this game since the old days and i love machine learning as well. Hit me up for discord if you like!
Hello, Mr Nacho! Thanks for your support and suggestions! It means a lot coming from someone who is actively in the machine learning field.
I just got back from a long holiday but I will definitely play around a lot more with the different features and update everyone soon. I also never heard of using a CNN to extract information, that's genius haha. Let me know what your name is on discord and I'll definitely hit you up :)
Actually, if you don't want your discord out there to the public, feel free to add me ChronicCoder#1667 or join the discord server discord.gg/ZummSXK (if you aren't yet)