SONM for Machine Learning: TensorFlow/Keras RNN as Jupyter notebook on 6 GPU mining rig
Jupyter is a modern tool for conducting data mining and machine learning experiments. It provides simple GUI for wiki notepad, inline program code snippets, and inline execution results. It transforms a remote mining rig into a powerful workplace for a data scientist. Ready to use solution with SONM!
We continue practical exercises with a 6 GPU rig, provided by Mining Union.
Today we selected a sample educational task for machine learning from Github, that just predicts Apple stock prices. It models Recurrent Neural Network using TensorFlow and Keras libraries. This project is presented as a Jupyter notebook.
We used:
- Publicly available Machine Learning educational task hosted on the GitHub: https://github.com/nerush/aind2-rnn/blob/master/RNN_project.ipynb
- Official Docker image with Jupyter and TensorFlow.
- Keras added manually according to project dependencies.
- UPD more details on CUDA+TensorFlow+Docker in SONM blog by Anastasiya Ashaeva: https://blog.sonm.io/machine-learning-on-sonm-is-now-live-f91f996da057
Now to the screenshots. Jupyter notebook start page on SONM:
Recurrent Neural Network project notebook start page:
Recurrent Neural Network training execution on Jupyter on 6 GPU mining rig in SONM:
More examples of Jupyter notebook execution on 6 GPU mining rig in SONM:
Working on this article I would like to give thanks to:
- Mining Union for provided equipment and for use case idea.
- Yevgen Nerush for RNN code from https://github.com/nerush.
- SONM team Eugene Manaev and Alexander Sigaev for preparing and testing this use case.
- Anastasiya Ashaeva for helping with TensorFlow setup in Docker: https://blog.sonm.io/machine-learning-on-sonm-is-now-live-f91f996da057.
