ForwardTest class

[einarbmag]

Hi, thank you for your excellent work, this is very interesting stuff.

I am eager to test this on the live market, but having trouble moving from backtesting to forwardtesting. Any chance that an update with a ForwardTest class is on the way, or that you could advise on how to implement it? I understand it roughly, i.e. the generate_history_matrix( ) function needs to update the datamatrix with the newest market data (with "online" = True in the config file), and return that. And the trade_by_strategy( ) clearly needs a slight rewriting compared to BackTest as we don't know the future price. Any help on how to correctly return the newest market data would be appreciated.

[dexhunter]

possible related to this comment

[gittaylor]

Yes, I am having the same problem. I was able to modify the Backtest class to get new data, but tflearn will not accept the new history for training. Any help would be appreciated.

[redzhepdx]

In backtesting there are two options to test model online or offline.

Online : Takes data from poloniex and updates database
Offline : Using pre-collected datas in Data.db

Backtesting module using rollingtrainer's datamatrix and it takes all datas from database at the begginning of the backtest.

But I couldn't find a way to get new data and process it .
It is using day based date to take data in given day based interval.
Is there any module to update database dynamically for livetrading without resetting learning of the model ?

[ZhengyaoJiang]

Is there any module to update database dynamically for livetrading without resetting learning of the model ?

A simple(while not elegant) method is to append new data to the global_data (Panel in the memory) and modify all the related indexes.

[redzhepdx]

I am doing this for real time data ;

• last input X: global_data[: , :, -window_size:]
• apply action according to X
• Wait n-minutes
• get input again
• y : newX[:, :, -1] / newX[:, :, -2]
  But couldn't solve the how to use thid x, y data to update RL agent.
  I think RL training must be always active for the understand every new situation , but it seems it don't train model in backtesting part adding None object to experience tracker module.
  Also thank you for first fast reply :)

[einarbmag]

Thank you for the hints @ZhengyaoJiang and @redzhepdx, will give it another go now!