multi stock data training about finrl-trading HOT 4 OPEN

I think you need to change parameter named "STOCK_DIM" which is total number of stocks in our portfolio in Env train, validation,trade files.

from finrl-trading.

I don't think so, in this repo all stocks in STOCK_DIM form a stock portfolio, in other words, after a prediction, people must do every ops(sell, buy, nothing) according to prediction result, but if I have STOCK_DIM=3000, I can't buy all of them, it may need a lot of money.

from finrl-trading.

From my point of view buy and sell actions are ranked according to the ammount to buy at each time step. So you are right if you have 3000 stocks in your action space hence possibly buy every stock of the 3000 then it can happen that not every stock is bought because the buy function is intitialized in a for loop and the actions are sorted according to the amount to buy maybe you should increase the intitial account balance from 100.000 to a higher amount or lower the HMAX parameter to control the rescaling of the actions because the PPO/A2C... rely on a gaussian distribution and are scaled to [-1,1]
you can also verify this by creating a variable (buy_activity/sell_activity) which is returend after testing to track the stocks bought and sold at each time step
just modifiy the testing environment if you want i can help you

from finrl-trading.

@emjay032 thank you, I'm trying to another method, that is every stock is one episode even in environment, so one episode is done, another one again, I'm doing it

from finrl-trading.