mohammad95labbaf / eurusd_lstm_attention Goto Github PK

I have a question about the preprocessing section in the ipynb.

It seems that the preprocessing of min-max scaling is applied to the entire dataset, but it is not distinguished between training data and validation data, which may cause data leakage. How are you addressing this issue?

So I pulled this down to work it through with your video and the predictions are coming from the middle of my csv. For instance, I have an AUD/USD df that is about 418 rows long post cleaning. I run through all the steps, dealt with the overfitting based on your suggestions. Then I added a table to print out the last 20 close values along with the predictions and graphs. When I do that I get a last close value that when I look it up it is in line 131 of the csv, has a value of 0.68362. The very last entry in my civ has a value of 0.65876. The prediction and value it is based off is not even close.

Any suggestions on how to fix this?