Lab5

Setting up

• Do the following from the unix prompt of your VM

  • Go to the directory you "cloned" the module files last time
  • Do `git pull origin master' to bring the new files

• Do the following tasks using your windows share or your unix account in the VM

  • Copy the lab files from the module directory into your own github lab directory, in "lab5" folder
  • Remove everything from the copied README.md

ipython/jupiter

• Start ipython/jupiter by typing ipython notebook --ip='*'
• Start a browser and connect to http://mlvm:8888/
  • You will need to input the token that was provided to you when you started ipython (looks like "c3fad33a4d227d5f395f6b2ce5de34c05b2dfa0ca516b36f" (NOT THIS ONE))
• Using the web page, go to lab5

Ipython notebooks

• Inside lab5 you will see Rec_correct.ipynb and Rec_features.ipynb

• Create a new Ipython notebook

Lab Exercises

• Load the data from the file ``jester-data-1.csv''

  • The data is from http://eigentaste.berkeley.edu/dataset/ and it contains the ratings of 101 jokes from 24,983 users
  • You can find the jokes in the website http://eigentaste.berkeley.edu/dataset/jester_dataset_1_joke_texts.zip

• Split the data into validation, test and training set with 80:10:10 proportions

• Use latent factor modelling to infer the hidden ratings of the users (they are labelled as "99" in the dataset) on the training set

• Calculate the performance of the algorithm in the validation dataset by looping through the dataset without training

• Change hyper-parameters (i.e. learning rates, number of iterations etc) as needed so you can get good results

• Report the MSE on the test dataset

• (if you have time) Use pandas to find the best and the worst rated jokes

• Once you are done, save your changes in github

  • Go inside your lab directory and do
    • git add -A -v
    • git commit -m <message>
    • git push origin master