Overview of deep learning frameworks: PyTorch, TensorFlow, and what comes next
See DeepLearningFrameworks.pptx in this repository for a the slides.
https://towardsdatascience.com/which-deep-learning-framework-is-growing-fastest-3f77f14aa318
https://www.fast.ai/2019/01/10/swift-numerics/
Try out TensorFlow! Experience with simple neural nets http://playground.tensorflow.org
https://colab.research.google.com/ is the simplest way to get started with
https://www.fast.ai/ has guidelines on setting up environments for PyTorch. Remember: virtual environments are your friend.
To use ACCRE for GPU computing, you need an ACCRE account. You can sign up for an ACCRE account here, and make sure to satisfy their class requirements.
Using your VUNetID and password, log onto ACCRE via the ACCRE portal. Navigate to the top menu and click on Interactive Apps. From the dropdown menu, choose Jupter notebook (GPU). Make sure you choose this GPU option. If you're not sure about you may use the defaults here or customize according to your needs:
- GPU Enabled Slurm Account:
enter your slurm group here - Number of hours:
4 - Number of requested GPU resources:
1 - GPU Architecture:
Pascal - Working directory:
${ACCRE_RUNTIME_DIR} - Python version:
Python 3.6.3 / Anaconda 5.0.1 - Use a virtual environment checkbox:
uncheck - Python or Conda Virtual Environment:
leave blank
Now that you've made your options for your Jupyter work environment click Launch. This will lead you to another page where ACCRE's scheduler (called SLURM) will try to schedule the time and resources you requested in your notebook. It will go from a Queued state to a Starting state and finally be ready when the Connect to Jupyter button appears. Click this button when it appears.
On virtual environments. In general, you should use virtual environments when you're developing, since modifying your base environment can result in package clashes for different projects. If you're interested in using a fast.ai virtual environment, you can modify the settings above with the following:
- Python version:
Python 3.6.3 / (GCC/6.4.0-2.28) - Use a virtual environment checkbox:
check - Python or Conda Virtual Environment:
/labs/accre_public/pascal/fast.ai.v2
More information on creating custom virtual environments using Anaconda can be found in the documentation under Anaconda at the bottom.
https://www.vanderbilt.edu/datascience/events/data-science-workshops/
Sign up for consults with our Data Science team!
Undergraduates: https://appoint.ly/s/vu-dsi-consult/student-consult
Researchers: https://appoint.ly/s/vu-dsi-consult/student-consult