Machine Learning NoteBook 0810

8/10: to-do

• finish the training of model version 5 which keep training until manually stop (make some graphs)
• start a new training for classification
• continue working on method

current status

• the 4th version of model already preform quite well in the inter-test set:
  

  • according to the test, $r^2$ value is $0.74$ almost hit the $0.75$, but it starts to have potential of overfitting

• Because the 4th version of model stop at 200 epochs and I forgot to save the $r^2$ value for training set. So, I start a new training with 2000 epochs and show the status in real-time with TensorBoard and also start to save the model every 100 epoch.

need to do tomorrow

• evaluate the new model and draw figures with the model

notice

• start to use TensorBoard in the code, so when you want to check the figures, pls use ssh to translate the port first with:
  • notice: when starting the TensorBoard don’t forget to use --logdir=dir to set the dir which save the data

1
2

ssh -L yourport:yourip:serverport username@serverip
ssh -L 10060:127.0.0.1:6006 wbm001@39.103.182.77

• when run the code with nohup the full sentence should be:

1

nohup cmd > outputfile 2>&1 &

figures and reference for paper

• NaN