$ python3 -m venv fund
$ . ./fund/bin/activate
$ pip -r requirements.txt
Run below command at the project root, then server will open at http://localhost:1357
$ tensorboard --logdir saved/log/ --port 1357
$ python train.py -c config.json -d 0
$ python test.py -c config.json -d 0 --resume saved/models/StockNet/1010_013058/model_best.pth
-
If I does not normalize the data, model will always predict the same value like 54.
-
My train curve (loss and difference): http://meow2.csie.ntu.edu.tw:8787/#scalars
-
prediction curve (training + testing phase)
-
prediction curve (only testing phase)
-
Performance is better when only use 1 layer GRU and don't use batch normalization.
-
Use technical index (47 dim) result in better performance.