Hands-on Tutorial on Deep Learning and Tensors
October 5, 2018
Information
What is it: a half-a-day detailed hands-on tutorial on deep learning and tensors, given by Jean Kossaifi. The tutorial is aimed to be beginner friendly and accessible to all Caltech people and AI4science enthusiasts. You will get to install Jupyter notebooks, run pytorch and tensorly, in addition to understanding the concepts.
When is it: October 5, 2018, 8:30am - 12:30pm (Note: Chandler cafe opens at 7am for breakfast)
Where is it: Beckman Institute Auditorium (Lecture Room).
How to participate: Please RSVP here
Schedule:
- 08:30 - Setup and introduction
- 09:00 - Introduction to deep learning and deep convolutional neural networks
- 10:00 - break
- 10:30 - Introduction to tensor methods
- 11:30 - Tensor decomposition and tensor regression
- 12:00 - Combining tensor methods and deep learning
Tutorial Materials
Available online at https://github.com/JeanKossaifi/caltech-tutorial
Prerequisites
You will need a laptop to get the most of the session, with Python installed, along with PyTorch, TensorLy and sckit-learn.
You also need to have the jupyter notebook installed.
Installation instructions
If you are new to Python or simply want a pain-free experience, I recommend you install the Anaconda distribution. It comes with all you need shipped-in and ready to use!
Once you have anaconda installed, you want to get either the Jupyter Lab, or the Jupyter Notebook. Typically, you can simply run:
conda install jupyterlab
For PyTorch, simply follow the instructions. With conda, it should be something like:
conda install pytorch torchvision -c pytorch
Finally, to install TensorLy to get the latest version, you can clone the Github repository. In the command line, run:
Or, if you have an issue during installation, you can also use conda:
conda install -c tensorly tensorly
Finally, you can install Scikit-Learn using pip. In the command line, just type:
pip install scikit-learn
Checking whether you have the correct versions
First, open a jupyter lab (just type jupyter lab in the command line and open the link that appears) . You can also run the code directly in Python or IPython but it's not as nice. Generally, if you are not already familiar with jupyter lab or jupyter notebook, I recommend you spend a little time exploring as it's a very useful tool, especially for data science.
To check you have all you need, you can run the following:
import torch
import numpy as np
import tensorly as tl
import sklearn
print(torch.__version__)
print(tl.__version__)
For pytorch, you should have version of at least 0.4.0