Does this code set the learning rate？when i use my own dataset，the loss becomes nan and i want to decrease the learning-rate but i cant find it

jaungiers / LSTM-Neural-Network-for-Time-Series-Prediction

LSTM built using Keras Python package to predict time series steps and sequences. Includes sin wave and stock market data

GNU Affero General Public License v3.0

4.85k stars 1.96k forks source link

Does this code set the learning rate？when i use my own dataset，the loss becomes nan and i want to decrease the learning-rate but i cant find it #77

Open zhangzhang-vvv opened 4 years ago

VincieD commented 4 years ago

No, it takes the defaulf one by adam. If you want to change it, add it in the config.json and change it in the model.py

dongdongrj commented 3 years ago

Change the LR by below code: K.set_value(model.optimizer.lr, 0.0001)

xFireBerg commented 3 years ago

Also faced similar issues, but then realized the preprocessing work is insufficient, and null values in any part of your dataset can break the model, thus giving you nan losses.