LogFAQs > #926498874

LurkerFAQs, Active DB, DB1, DB2, DB3, DB4, Database 5 ( 01.01.2019-12.31.2019 ), DB6, DB7, DB8, DB9, DB10, DB11, DB12, Clear
Topic List
Page List: 1
TopicMy AI predicts these are the future prices of bitcoin
Yellow
08/24/19 5:44:44 PM
#8:


After a little tweaking, it works a little better.

BitCoin at January 02, 2018 03:15:00 (am)
Current price: 13452
Predicted price: 13280.0
Actual future price: 13415
Error is 0.049328848597268805

It correctly predicted it would go down. Not as much as it thought, but it did.

Main caveat: it takes 40 minutes to predict the next 15 minutes. (That's why I only have this one example) I can optimize this better, though. I can also transform the data so values between 7000-13000 are much further apart, meaning it might guess more accurately in that range.

I also have a backup plan to have it only analyze price differences, as I only care about their prices relative to each other.

I also know why (certain) other projects have failed at this point. The reason why is incredibly stupid. :P

People use OHLC (open high low close) data. The close value is the same as the next open. People use both the columns in their training data, allowing the NN to "cheat" and just take the value from the day's close as the next "open".

I read a whole article explaining how machine learning couldn't predict the market, and they all had this exact problem, they just chucked the whole OHLC dataset in the algorithm, and concluded that it's an alchemist's task.

Anyone familiar with the market should know a little what I'm trying to say here.
---
... Copied to Clipboard!
Topic List
Page List: 1