lecture 10
Practical AI:
Recommendation system - Movie Recommender
with CreateML

SeAts APp SEAtS ApP SEaTS APP

🧘

my fav 10-min video to wake us up

after today's lecture:
-- Introduction to recommendation system 🤑
-- A simple movie recommender

money talks 🤑

money-related AI:
-- recommendation system
-- market price prediction
-- etc.

art work prices prediction

art analytics: artnome

Recommendation System!
(RS)

before talking about what RS is, let's start by examining the data

Training:
Data
Netflix Prize data

two entities in RS: user and item (movie)

one quantity: rating

to organise user, item and ratings together: tabular data with use-movie ratings
(sketch on the whiteboard)

recommendation system paradigms:
-- content-based: aka find similar user OR item
-- collaborative filtering: similarities between users and items simultaneously
etc.

keyword: embedding(aka some sort of represenation of user/item) and similarity
-- more explanations here

let's do some data pre-processing and training!
all code here 🥰

open a terminal:

cd to/the/code/folder 

conda create -n RecPrepEnv python=3.6 

conda activate RecPrepEnv 

put "netflix-prize-data" to the same directory as preparation.py (why?)

open a terminal:

cd to/the/code/folder 

conda create -n RecPrepEnv python=3.6 

conda activate RecPrepEnv 

python preparation.py 

then, a new "netflix-prize-data.csv" will appear in the folder "netflix-prize-data", check it out!

go to CreateML

select recommendation

add the "netflix-prize-data.csv" to the training data

assign users/items/ratings dropdown menu 

(what is in the drop downmenu and why are they there?)

training and download the model 

next:
run the training playground to see what the output of the recommender ML model is

some philosophy noodling
-- recall chatgpt's success is partially attributed to its "abstraction" and "compression" power ?
-- this adds more spice to the discussion

lil open question:
-- where else have you seen the tree-like structures?

for example,
platonic ideal form, recipes, western music notation, etc.

lil open question:
what have we lost during the hyped compression and abstraction process?

today we talked about:

- Recommendation system
- Movie Recommender as an example

!!!IMPORTANT!!!

No MLTwo lecture next week
Double lectures next next week (25th)
Don't worry, I'll make the agenda on the 25th more relaxing 😎
drop me a message if you have any concerns/feedbacks

Next:
Sketch some ideas for the final presentation:
a bespoke ML model
a trained model using techniques we have covered so far
-- if you want to train some model of your own, start think about dataset curation 🤠
-- if you want to use a trained mode, focus on the application 👾