Feature : Learning speed and backtesting

[raresjuretchi]

I am not sure if the model is learning or if it is a slow learner.
I am thinking if there is a way to speed up the learning (increase the frequency of the tuning?!) ; not sure

My concern is, each time you start the app, the model will have to start training itself and it can take days ...weeks ...months ? (not sure how much time) until it will start making money ?

The problem is - in a real world scenario, you cannot afford each time you start the app to wait for it to get better (train) ...you are going to lose money and perhaps by the time it gets better there will be no more money to trade :)

some suggestions

1. save the model (first time, then append)
2. load the model (when the program starts)
3. Speed up the training (by using 1 min OHLC over a period of a few days)
   4.(if possible) Train the model by backtesting before trading, and select the optimum strategy parameters to start with, and adjust overtime
4. add "scalping" - technical indicators may or may not work because there are many factors involved in the market ; however you can have a function running in parallel that is looking to do just that "scalping" meaning to take the profit (sell) if the stock price is up by XX% (a variable to be defined) compared with what you paid for it. This way if the technical analysis fails, you at least take some profit.

[ProgramingPro2]

Yeah. So I would love if someone made a separate backtesting script for this, because I'm currently busy fixing bugs in the main one, I will get around to this soon anyways, but in previous projects, it has taken a very long time to add this feature.
You currently are able to change the learning frequency of the model using:

    def __init__(self, alpha=0.4, gamma=0.7, epsilon=0.5):
        self.alpha = alpha      # Learning rate
        self.gamma = gamma      # Discount factor
        self.epsilon = epsilon  # Exploration rate

but I do believe there is a bug with the reward calculation so that if the script is restarted during open hours the reward will reset.