In [2]:

    

import seaborn as sns
import pandas as pd
%pylab inline


    

    




Populating the interactive namespace from numpy and matplotlib

In [143]:

    

dset = pd.read_csv('diffp2-perc.csv', sep=';',header=None, names=['dataset', 'niter', 'epsilon', 'accuracy'])


    

In [144]:

    

baseline = pd.read_csv('standard.csv', sep=';',header=None, names=['dataset', 'niter', 'loss', 'accuracy'])


    

In [145]:

    

set(baseline.loss)


    

    
Out[145]:





{'hinge', 'log', 'perceptron'}

In [ ]:

    




    

In [146]:

    

jnd = pd.merge(dset, baseline[baseline.loss == 'perceptron'], on=['dataset', 'niter'])
jnd['relative accuracy'] = jnd.accuracy_x/ jnd.accuracy_y
jnd['privacy level'] = 1/(jnd.niter*jnd.epsilon)
#jnd


    

In [147]:

    

f, ax = plt.subplots(figsize=(10, 10))
sns.set_context("poster")
sns.set(font_scale=1.5)
ax.set(xscale="log")
g = sns.regplot(x='privacy level', y='relative accuracy',ax=ax, data=jnd)
g.set(ylim=(0, 1.6))
g.set(xlim=(10**-2, 500))


    

    
Out[147]:





[(0.01, 500)]






    

In [113]:

    

noisy = pd.read_csv('more_noise.csv', sep=';',header=None, names=['dataset', 'noise', 'niter', 'loss', 'accuracy'])


    

In [114]:

    

#noisy


    

In [131]:

    

jnd = pd.merge(noisy[noisy.noise <=7], baseline[baseline.loss=='hinge'], on=['dataset', 'niter', 'loss']) 
jnd['relative accuracy'] = jnd.accuracy_x  / jnd.accuracy_y


    

In [132]:

    

f, ax = plt.subplots(figsize=(10, 10))
sns.set_context("poster")
sns.set(font_scale=1.5)
ax.set(xscale="log")
g = sns.regplot(x='noise', y='relative accuracy',ax=ax, data=jnd)
g.set(ylim=(0, None))
g.set(xlim=(10**-2, 10**1))


    

    
Out[132]:





[(0.01, 10)]






    

In [ ]: