Goals

• We should have a Github repo

• Each person should be able to…

• Run the code from last project on their laptop

• Access DeepThought and/or Nathan’s Desktop and run intensive training code

• Access and contribute to the training data repository

Instructions

Things to install

• Python 2.7

  • Test: in terminal, run python --version

• ROS Kinetic

  • Test by running rosversion -d in a terminal

• ML Libraries:

  • TensorFlow

    • Don’t try to set up GPU support, it will end in fire and pain.

    • sudo -H pip install tensorflow

    • Test: in terminal, run python -c "import tensorflow; print('Works!')"

  • Keras

    • sudo -H pip install keras

    • Test: in terminal, run python -c "import keras" It should say “Using TensorFlow backend.”

• Misc

  • Bcolz

    • Test: run python -c "import bcolz; bcolz.test()"

  • Numpy: The preferred way to run numeric computations in python. All hail Numpy.

    • Test: run python -c "import numpy as np; np.test()"

  • Glob: Returns names of items in a directory as a list.

    • Test: run python -c "from glob import glob; print('Works')"

  • PIL

  • Matplotlib: Plots in python, based off of MatLab’s plotting commands

    • sudo apt-get install python-matplotlib for Ubuntu Linux

    • Test: run python -c "import matplotlib.pyplot as plt; print('Works!')"

  • Seaborn: Prettier plots on top of matplotlib

    • Test: python -c "import seaborn as sns;print('Works')"

  • Tqdm: Adds progress bars to loops

    • Test: python -c "import tqdm; print('Works')"

Working on a remote computer:

Notes

• The examples here assume the remote computer is called ‘nathanDesktop’, the username is ‘nathan’, and DNS resolving works properly on your network.

• We are connecting to the remote computer because it’s a better way than using a shared user account. If you have a different method that works better for you, follow that.

Steps:

1. Get access to the remote computer using SSH public keys (once)

   1. Generate them (Ubuntu Guide)

   2. Transmit them to the owner

      1. Your public key is found at ~/.ssh/id_rsa.pub

      2. Send the contents of that file over email, Google Docs, etc.

      3. While you’re at it, you should add your SSH key to Github, if you haven’t already. It’ll allow you to access your repos without entering your u/n and password every time.

   3. Owner adds the keys to server

      1. All public keys should be stored in .ssh/authorized_keys

      2. http://stackoverflow.com/questions/12392598/how-to-add-rsa-key-to-authorized-keys-file

2. Access and edit your files!

   1. Connect to the computer via ssh

      1. ssh nathan@nathanDesktop

   2. [Optional] If you want to use tmux rather than your normal terminal:

      1. Try $ tmux ls

         1. Should print out that jupyter_server is running.

      2. tmux a -t jupyter_server

         1. Navigate tmux using ctrl+b, arrow key

         2. https://learnxinyminutes.com/docs/tmux/

         3. https://danielmiessler.com/study/tmux/#gs.=j=4xnE

   3. Open the remote jupyter server in your browser to start editing files

      1. $ ssh -N -f -L localhost:8890:localhost:8890 nathan@nathanDesktop

      2. Go to localhost:8890/tree in browser

      3. Change kernel to your deep learning one. On Jupyter’s top bar, Kernel>’Python [condaenv:deepLearning]’

   4. Want a file explorer - nautilus

      1. sftp://nathan@nathandesktop/home/nathan/olin/spring2017/line-follower

   5. Try broadcasting a message: wall _____

Code to download

Clone our repository into