week 4

Question 1

Collab filtering

Answer 1

suggesting complex items wihtout understanding the nature of them but by seeing similarties in users (amy liked it and she has similar prefrence to bob so will recomend to bob)

Question 2

content filtering

Answer 2

find similarities in items so recomend based on them

Question 3

collab filter issues

Answer 3

lots data , as much as possible
millions of items need lots computing power
requires one user who has seen movie or done variable

Question 4

content filter requires little data to start what are some issues

Answer 4

 Can be limited in scope
 Not present something different just similar items that user already likes

Question 5

clustering helps come up with better

Answer 5

predictive models

Question 6

2 types learning algorithim

Answer 6

supervised- : gives you some outcome
 Trying to predict something
unsupervised- don’t predict anything just group data into similar groups to build better predictive models or market segmentation etc

Question 7

cluster is repped by what normally

Answer 7

cnetroid

Question 8

how to find distance between centorid

Answer 8

take mean of coordinates of all the points in cluster and will give you the centroid for that cluster

or use any other distance metrics we dicussed

Question 9

why normalize data

Answer 9

so all distance is relevant as distance highly influenced by scale of variables so normalize it

can greatly change clusters if normalized different

Question 10

• Hierarchal closeting starts as

Answer 10

assume each point own cluster then merge into one eventually

Question 11

vertical lines in dendogram smallest at

Answer 11

bottom once move along height will incease

Question 12

how to decide how many cluster we want

Answer 12

 To do this draw horizontal line and number of vertical lines crossed is amount of clusters to have

Question 13

cluster process

Answer 13

o 1st thing is to calculate distance between all points
o Then find out point that has minimum distance between them and merge. Continue to do so until 1 cluster
 calculate 2 points centroid then use centroid as distance for next point on dendogram

Question 14

where should we draw line to decide clusters

Answer 14

1-depends on problem itself move line slight and changes cluster= very sensitive to info
2- if no specific want use dendogram and choose cluster in this case more robust to inaccuracy in data. good to have some robustness