Read this details of the usecase on this link.
Prerequisites for this environment
- Log on to Azure
- Create a new Linux DSVM (Choose NC6 at least), this runs on GPU
- Log in via putty (or something similar)
- Run the following commands on the linux terminal
- Open your jupyter server, upload the files and run them.
The following files are in this repo:
- First model with explanation of approach (01_pred_maintainance_LSTM_GPU.ipynb)
First attach to a terminal for connectivity issues:
tmux attach -t 0
Environment configuration, replace py36-mh with your environment name:
conda create -n py36-mh python=3.6
conda activate py36-mh
conda install numpy pandas matplotlib tensorflow-gpu keras h5py scikit-learn -y
conda install -c anaconda-nb-extensions nb_conda -y
conda install -c conda-forge jupyter_contrib_nbextensions -y
pip install azureml-sdk[notebooks]`
jupyter nbextension install --py --user azureml.train.widgets
jupyter nbextension enable --py --user azureml.train.widgets
Finally, register your kernel. Replace Python (py36-mh) with your kernel name
python -m ipykernel install --user --name py36-mh --display-name "Python (py36-mh)"
If you want to monitor GPU while training, use this command, it is on a loop to update every second:
nvidia-smi -l 1