Skip to content

Latest commit

 

History

History
37 lines (25 loc) · 2.73 KB

README.md

File metadata and controls

37 lines (25 loc) · 2.73 KB
Raw

Modelling of k-growing graphs and privacy for eta-AD

Binder

Models

The models we fit in this example are built in the following way:

\begin{align*} M_1(n,k) &= \alpha \ln(\beta n) + \frac{\gamma}{e^{\delta k}} + \epsilon, \ M_2(n,k) &= \frac{\alpha \ln(\beta n)}{e^{\gamma k}} + \delta \ln(\eta n) + \frac{\zeta}{e^{\gamma k}} + \epsilon, \ M_3(n,k) &= \frac{\alpha \ln(\beta n)}{e^{\gamma k}} + \epsilon, \ M_4(n,k) &= \frac{\alpha \ln(\beta n)}{e^{\gamma k}} - \frac{\alpha \delta k}{e^{\gamma k}} + \frac{\zeta}{e^{\eta k}} + \frac{\theta \ln(\iota n)}{(\kappa n)^{\lambda n}} +\nu \ln(\xi n). \end{align*}

Structure

Distribution Model Notebook

Followalong with the Distribution Model section of the paper. (You can run the interactive versin on Binder here: https://mybinder.org/v2/gh/vs-uulm/eta-adaptive/main?filepath=Distribution%20Model.ipynb )

Prepared Datafiles

There are three files provided in 'data/': 'normal_approx_' 1 through 3. These files are pickle files of normal distribution fits for various experimental results. Number 1 was used to fit the models, while 2 and 3 are used for validation.

These files are loaded via 'load_data' from 'helpers.py'. Their contents are also available convieniently via 'data.py' as prepared objects base, val1 and val2.

Python Files

  • 'models.py': implementation of the fittable and fitted models we use for calculating mu and sigma.
  • 'data.py': Provides the 3 datasets as canonical objects base, val1 and val2.
  • 'helpers.py': Provide some internal helper methods to process, load, transform etc. data.
  • 'generate.py': Generate a pickle file of normal distribution fits for random graphs. Results will be of compatible format to be loaded via 'load_data' from 'helpers.py'