search

blacksnail789521 / TimeDRL

Official repository for TimeDRL: Disentangled Representation Learning for Multivariate Time-Series, accepted at ICDE 2024.
MIT License
11 stars 2 forks source link

[BUG] Why are my results poor? #2

Open ipc-robot opened 3 months ago

ipc-robot commented 3 months ago

I'm troubled by the model's performance. I didn't change anything, I just modified the config according to the json file. 2024-04-06 13-52-10屏幕截图 I choose the classification task and the dataset is "HAR".

blacksnail789521 commented 3 months ago

This is my result just before the pre-training stage's epoch 9. image

Remember, when utilizing the hyperparameter settings defined in exp_settings_and_results, you should execute the following command: python main.py --overwrite_args.

blacksnail789521 commented 3 months ago

Yes, the fixed and tunable params would be updated. image

ipc-robot commented 3 months ago

Now the neural network is running and looks great! Can I save the best model parameters? I passed the path to the early stopping module, but the saved one doesn't seem to be the best one, which bothers me.

blacksnail789521 commented 3 months ago

Ok, I'll try to fix it in the future, currently busy with another paper.