Machine Learning on Levante#
How to set up conda#
While the default python3 environment offered on Levante via modules offers a basic set of libraries, often enough ML with Python requires additional libraries. An easy way to gain access to them is to set up your own conda environment.
How to get a custom kernel into jupyterhub.dkrz.de#
Jupyter notebooks and jupyterhub.dkrz.de are suitable environments to get started with ML on Levante. However, the default kernels available on jupyterhub.dkrz.de do not include many of the standard frameworks used for Machine Learning in Python. You can, however, set up your own kernel to address this.
Tailoring ANN training for batch scripts#
usage: Model.py [-h] [-d DATA] [-e EPOCHS] [-p PATIENCE] [-bs BATCH_SIZE]
[-lr LEARNING_RATE] [-wd WEIGHT_DECAY] [-optim {Adam,SGD}]
[-nni] [--n-layers {1,2,3}]
[--units1 UNITS1] [--dropout1 DROPOUT1]
[--units2 UNITS1] [--dropout2 DROPOUT1]
[--units3 UNITS1] [--dropout3 DROPOUT1]
optional arguments:
-h Print help and exit
-d DATA Input data
-e EPOCHS Maximum number of epochs to train
-p PATIENCE Stop training prematurely if results do not improve for n epochs
-bs BATCH_SIZE Batch size
-lr LEARNING_RATE Learning rate
-wd WEIGHT_DECAY Weight decay
-optim {Adam,SGD} Optimizer
-nni Special flag to use if script is run via NNI
--n-layers {1,2,3}
--units1 UNITS1
--dropout1 DROPOUT1
--units2 UNITS2
--dropout2 DROPOUT2
--units3 UNITS3
--dropout3 DROPOUT3
This could be easily extended to, e.g., cover not only testing with an architecture with different dense and dropout layers, but also with/without convolutional or LSTM layers and different parameters for them.
Automating training and hyperparameter tuning using NNI#
pip install nni
On Levante, you are able to access the web interface if you are using NNI on a GPU node and are inside the DKRZ network (e.g. via VPN or by being in a DKRZ building). The server address and port are displayed when an experiment is started.