ai-se/privacy_sharing

Applying LACE2 to US census data

privacy_sharing's Issues

Morphing w/ Starminer Installed

Basically, I morph it crashes. I looked all over the internet and never found a compatibility fixing. If you know how or IChun can make version 1.9.3 for 1.7.10 with compatibility let me know

privacy

http://oecdprivacy.org/#collection

implement CLIFF

run more data set data sets thru lace 2

implement MORPH

the privacy measures ection of the paepr to charlotte

for the defect data, commission the APRIORI association rule learner

for the defect data, can you commission the APRIORI association rule learner

• Rakesh Agrawal and Ramakrishnan Srikant Fast algorithms for mining association rules in large databases. Proceedings of the 20th International Conference on Very Large Data Bases, VLDB, pages 487-499, Santiago, Chile, September 1994.)?

one the things we promised the sponsers was a study to check how much our privatization damaged the data. if learn associations on the original and privitized data, then look for differences, that is a measure of damage.

notes that:

• data must be discretized for the association rule learning to work.e.g. (max - min)/10
• perhaps we should ignore the dependent variable during rule learning.

Phishing dataset

• https://archive.ics.uci.edu/ml/datasets/Phishing+Websites (for training)

  • contains binary attributes

• http://data.phishtank.com/data/online-valid.csv

  • current websites which are phishing and online. (for different geographical regions)
  • problem is we will have to manually retrieve attributes based on 1st dataset.

• there is nothing as private which cant be shared so dont know how useful this dataset will be.

readings

• Fayola TSE13-LACE
• Fayola ICSE15-LACE2

apply CLIFF+MORPH to defect datasets

implement LACE2 model

implement IPR, g-measure

lace1 & lace2 results

DEFAULT = {
    'model': ['camel-1.6', 'xerces-1.4', 'ant-1.7'],
    'test_set_ratio': 0.2,
    'CLIFF_percentage': 20,
    'Lace2_holder_number': 5,
    'MORPH_alpha': 0.15,
    'MORPH_beta': 0.35,
}

college record

dependent variables

• retention rate
• revenues.
• completion
• earnings
• less chort default rate.

collegecard::IPR and association rule

check the preccision

according to the definition of pd,pf and g
higher is better.
now since it is lower

collegecard:: prediction

code release to charlotter

a readme that says install and how to run one example

softare practice and experience on the toolkit

confirm the IPR

check whether the current IPR is correct