Result of programming mistakes! I looked up the code. the data was first denoised by wavelet soft method, so nearby data affect each other. data must be first cropped and then each one denoised independently. I fixed the code and the result accuracy is around 50% like flipping a coin. (also testing the ResNext model in with #"torch_model.training == True" and shuffling the train-test data for LightGBM which makes the model see the previous model's train data as test_dataset are two other bugs in the code! But thanks for sharing your data anyway!!
The results are absolutely amazing. I am pretty surprised by the accuracy on such a lower time frame. But definitely it would have been much interested to see where his model predicted wrong results.
Result of programming mistakes! I looked up the code. the data was first denoised by wavelet soft method, so nearby data affect each other. data must be first cropped and then each one denoised independently. I fixed the code and the result accuracy is around 50% like flipping a coin. (also testing the ResNext model in with #"torch_model.training == True" and shuffling the train-test data for LightGBM which makes the model see the previous model's train data as test_dataset are two other bugs in the code! But thanks for sharing your data anyway!!
can you share your code with me?
The results are absolutely amazing. I am pretty surprised by the accuracy on such a lower time frame. But definitely it would have been much interested to see where his model predicted wrong results.
Awesome! Have you tried to predict for longer time such as the next 1 day