tsfresh
tsfresh copied to clipboard
blue-yonder
beat the Y-1 baseline in Google Stock notebook
Hi, first I think this is an awesome project! However, I would like to see a notice on the Google Stock notebook that this is just a technical example but the results gained from this have no practical relevance whatsoever.
If we check whether the prediction got the direction of the next day correct it is just slightly below 50%.
Other than that plotting the next day's prediction will always look nice but just taking the last day's value as prediction looks even nicer ;) So we can't really gain any information from that either.
On a different note I would love to see a notebook example on how to use tsfresh with keras
Other than that plotting the next day's prediction will always look nice but just taking the last day's value as prediction looks even nicer ;)
We can the add MAE of both models, feature based adaboost and the Y-1. If the Y-1 is still better, then we need to tweak the adaboost more :D. However, this was only thought as an example how one can use feature to forecast time series and not as a investment model.
On a different note I would love to see a notebook example on how to use tsfresh with keras
You can just calculate the feature matrix and then pass it into your keras model.
out of the box, we are not beating the y-1 benchmark, which I added to the notebook.