Direction Accuracy and Loss

[minion50000]

I think it would be better if we try the model training based on the direction loss
prediction_dir = np.sign(prediction[:-1] - prediction[1:])
truth_dir = np.sign(truth[:-1] - truth[1:])
dir_acc = np.mean(prediction_dir == truth_dir)
What do you think? 😉

[mltrev23]

Great
You hit the nail on the head! 💯 🙌
Actually, it could improve the model's performance, imo.
👍

[mltrev23]

could you please create pr or will I?

[minion50000]

okay, i will