Thank you to the authors for open sourcing the code, which provides us with a great learning opportunity.
However, I've noticed that in the market scoring unit implementation, when calculating the temporal attention weights, there seems to be a missing a part compared with the expression in the paper. More specifically, in the code, only the output of the LSTM network is used when calculating the weights, and the macro market indicator is not included.
Thank you to the authors for open sourcing the code, which provides us with a great learning opportunity.
However, I've noticed that in the
market scoring unitimplementation, when calculating the temporal attention weights, there seems to be a missing a part compared with the expression in the paper. More specifically, in the code, only the output of the LSTM network is used when calculating the weights, and the macro market indicator is not included.