Think about normalizing the stock prices. Your 30 Stock are on completely different nominal ranges. I always though ML models don't like large ranges. Further recommend to ensure the agent can make his decision not only on the current last candle but is able to see the the last x candles in different time frames and makes the buy / hold / sell decision on this. I tried finRL a while ago, was not able to make it work but now your videos motivate me to get a working agent. I still looks for the simpler to use frame work since the problem statement should be simple. You have three actions and stock data plus indicators. Make a call.
I recently did another project that is a bit more robust than this for feature selection on Bitcoin predictions. FinRL is definitely fun, but still yet to make something remotely ready for real world implementation.
Not sure if FeatureEngineering separates the data by tic but the return & change in volume that you calculated is totally off because pandas is working row wise and consecutive rows have data for different tickers….
Good catch, I didn't even think to double check that. I'll be taking a look under the hood over the next few months to see what exactly some of these systems are doing.
Good intentions on video channel 👍👍. However, it's useful to realise some things... (IMHO) All stocks are like flotsam floating on the sea. That is to say, they are primarily subject to large forces outside of the business. Primarily risk free return of government bond interest rates. Ie things like changes in fed rate is the primary stock market driver. Secondary things that affect inflation, like price of oil, jobs/unemployment figures, CPI (consumer price index) , PPI (producer price index). Then there's government policies that can affect certain industries. SEC edgar database holds financial filings for individual companies, news for individual companies. So the best ai is something that can process text and say if the news is good, bad or neutral.
I do think that technical indicators alone won't actually predict price movement. From my limited understanding, good trading is based on real world cycles of cause and effect measured by some of the things you mention. I plan to make future videos where NLP helps play a part in making entering and exiting positions.
![[Live] : ONE 168 วันนี้!! "โจนาธาน vs ซุปเปอร์เล็ก"](http://i.ytimg.com/vi/0jIN65KgJsI/mqdefault.jpg)
![[Re:all TREASURE] EP.6 in 방콕ㅣ🚩 쩡캠의 방콕 투어 빵 마이 와이 👍](http://i.ytimg.com/vi/uyFw-nItA_s/mqdefault.jpg)
Clearly an important point of view! Well done! Keep 'em coming! 👍
Thanks for watching!
would be interesting to see how to get feature importance in the finrl ensemble model
Think about normalizing the stock prices. Your 30 Stock are on completely different nominal ranges. I always though ML models don't like large ranges.
Further recommend to ensure the agent can make his decision not only on the current last candle but is able to see the the last x candles in different time frames and makes the buy / hold / sell decision on this.
I tried finRL a while ago, was not able to make it work but now your videos motivate me to get a working agent.
I still looks for the simpler to use frame work since the problem statement should be simple. You have three actions and stock data plus indicators. Make a call.
I recently did another project that is a bit more robust than this for feature selection on Bitcoin predictions. FinRL is definitely fun, but still yet to make something remotely ready for real world implementation.
Not sure if FeatureEngineering separates the data by tic but the return & change in volume that you calculated is totally off because pandas is working row wise and consecutive rows have data for different tickers….
Good catch, I didn't even think to double check that. I'll be taking a look under the hood over the next few months to see what exactly some of these systems are doing.
Great video, thanks!
Glad you liked it!
Good intentions on video channel 👍👍. However, it's useful to realise some things... (IMHO)
All stocks are like flotsam floating on the sea. That is to say, they are primarily subject to large forces outside of the business. Primarily risk free return of government bond interest rates. Ie things like changes in fed rate is the primary stock market driver. Secondary things that affect inflation, like price of oil, jobs/unemployment figures, CPI (consumer price index) , PPI (producer price index). Then there's government policies that can affect certain industries. SEC edgar database holds financial filings for individual companies, news for individual companies. So the best ai is something that can process text and say if the news is good, bad or neutral.
I do think that technical indicators alone won't actually predict price movement. From my limited understanding, good trading is based on real world cycles of cause and effect measured by some of the things you mention. I plan to make future videos where NLP helps play a part in making entering and exiting positions.
Nice , Can we have the google colab of this ?
Dude your videos are top quality...can you do one on FinGPT if you get a chance
I will 100% do a video on FinGPT!
PERFECT
thank you