Hi, I stumbled across this repo looking for an LSTM implementations for FreqAI and the PyTorch regressor you've created is great, so many thanks for sharing your work on that.
I'm just confused about the implementation in the example strategy really... the set_freqai_targets method uses a lot of indicators to create the score to be used as a target variable but these all appear to be based on rolling windows of past data, essentially an aggregation of standard indicators. Where is it looking ahead so it can make predictions about the future score? e.g. All of the other example FreqAI strategies I've seen use some variation of shift(-n) for target variables.
Hi, I stumbled across this repo looking for an LSTM implementations for FreqAI and the PyTorch regressor you've created is great, so many thanks for sharing your work on that.
I'm just confused about the implementation in the example strategy really... the
set_freqai_targets
method uses a lot of indicators to create the score to be used as a target variable but these all appear to be based on rolling windows of past data, essentially an aggregation of standard indicators. Where is it looking ahead so it can make predictions about the future score? e.g. All of the other example FreqAI strategies I've seen use some variation ofshift(-n)
for target variables.