changing tutorials (adding part)

tutorials/1_Single_Frequency_Estimation.py

@@ -0,0 +1,180 @@
+#!/usr/bin/env python
+# coding: utf-8
+
+# ## Common Libraries
+
+# In[1]:
+
+
+import matplotlib.pyplot as plt
+import matplotlib
+params = {'axes.titlesize':'14',
+          'xtick.labelsize':'14',
+          'ytick.labelsize':'14',
+          'font.size':'14',
+          'legend.fontsize':'medium',
+          'lines.linewidth':'2',
+          'font.weight':'normal',
+          'lines.markersize':'10'
+          }
+matplotlib.rcParams.update(params)
+plt.rcParams["mathtext.fontset"] = "cm"
+plt.rc('font', family='serif')
+
+import time
+import numpy as np
+import pandas as pd
+from sklearn.metrics import mean_squared_error
+from sklearn.preprocessing import LabelEncoder
+
+
+# ## Importing Pure Frequency Oracles from multi_freq_ldpy
+
+# In[2]:
+
+
+from multi_freq_ldpy.pure_frequency_oracles.GRR import *
+from multi_freq_ldpy.pure_frequency_oracles.UE import *
+from multi_freq_ldpy.pure_frequency_oracles.ADP import *
+from multi_freq_ldpy.pure_frequency_oracles.LH import *
+
+
+# ## Reading Adult dataset with only 'age' attribute
+
+# In[3]:
+
+
+df = pd.read_csv('datasets/db_adults.csv', usecols=['age'])
+df
+
+
+# ## Encoding values
+
+# In[4]:
+
+
+LE = LabelEncoder()
+
+df['age'] = LE.fit_transform(df['age'])
+df
+
+
+# ## Static Parameteres
+
+# In[5]:
+
+
+# number of users (n)
+n = df.shape[0]
+print('Number of Users =',n)
+
+# attribute's domain size
+k = len(set(df['age']))
+print("\nAttribute's domain size =", k)
+
+print("\nPrivacy guarantees:")
+
+# range of epsilon
+lst_eps = np.arange(0.5, 5.1, 0.5)
+print('Epsilon values =', lst_eps)
+
+
+# ## Comparison of frequency oracles
+
+# In[6]:
+
+
+# Real normalized frequency
+real_freq = np.unique(df, return_counts=True)[-1] / n
+
+# Repeat nb_seed times since DP protocols are randomized
+nb_seed = 50
+
+# Save Mean Squared Error (MSE) between real and estimated frequencies per seed
+dic_mse = {seed: 
+               {"GRR": [],
+               "SUE": [],
+               "OUE": [],
+               "BLH": [],
+               "OLH": [],
+               } 
+               for seed in range(nb_seed)
+          }
+
+starttime = time.time()
+for seed in range(nb_seed):
+    print('Starting w/ seed:', seed)
+
+    for eps in lst_eps:
+
+        # GRR protocol
+        grr_reports = [GRR_Client(input_data, k, eps) for input_data in df['age']]
+        grr_est_freq = GRR_Aggregator(grr_reports, k, eps)
+        dic_mse[seed]["GRR"].append(mean_squared_error(real_freq, grr_est_freq))
+
+        # SUE protocol
+        sue_reports = [UE_Client(input_data, k, eps, optimal=False) for input_data in df['age']]
+        sue_est_freq = UE_Aggregator(sue_reports, eps, optimal=False)
+        dic_mse[seed]["SUE"].append(mean_squared_error(real_freq, sue_est_freq))
+
+        # OUE protocol
+        oue_reports = [UE_Client(input_data, k, eps, optimal=True) for input_data in df['age']]
+        oue_est_freq = UE_Aggregator(oue_reports, eps, optimal=True)
+        dic_mse[seed]["OUE"].append(mean_squared_error(real_freq, oue_est_freq))
+
+        # BLH protocol        
+        blh_reports = [LH_Client(input_data, eps, optimal=False) for input_data in df['age']]
+        blh_est_freq = LH_Aggregator(blh_reports, k, eps, optimal=False)
+        dic_mse[seed]["BLH"].append(mean_squared_error(real_freq, blh_est_freq))
+
+        # OLH protocol       
+        olh_reports = [LH_Client(input_data, eps, optimal=True) for input_data in df['age']]
+        olh_est_freq = LH_Aggregator(olh_reports, k, eps, optimal=True)
+        dic_mse[seed]["OLH"].append(mean_squared_error(real_freq, olh_est_freq))
+print('That took {} seconds'.format(time.time() - starttime))        
+
+
+# ## Plotting metrics results
+
+# In[7]:
+
+
+plt.figure(figsize=(8,5))
+plt.grid(color='grey', linestyle='dashdot', linewidth=0.5)
+plt.plot(np.mean([dic_mse[seed]["GRR"] for seed in range(nb_seed)], axis=0), label='GRR', marker='o')
+plt.plot(np.mean([dic_mse[seed]["SUE"] for seed in range(nb_seed)], axis=0), label='SUE',marker='>',linestyle='dashed')
+plt.plot(np.mean([dic_mse[seed]["OUE"] for seed in range(nb_seed)], axis=0), label='OUE',marker='s',linestyle='dotted')
+plt.plot(np.mean([dic_mse[seed]["BLH"] for seed in range(nb_seed)], axis=0), label='BLH', marker='D', linestyle=(0, (3, 10, 1, 10)))
+plt.plot(np.mean([dic_mse[seed]["OLH"] for seed in range(nb_seed)], axis=0), label='OLH',marker='d',linestyle=(0, (5, 10)))
+
+plt.yscale('log')
+plt.xlabel('$\epsilon$')
+plt.ylabel('Mean Squared Error (MSE)')
+plt.xticks(range(len(lst_eps)), lst_eps)
+plt.legend(ncol=2)
+plt.show()
+
+
+# ## Example of Real vs Estimated Freqencies
+
+# In[8]:
+
+
+plt.figure(figsize=(12, 5))
+
+barwidth = 0.4
+x_axis = np.arange(k)
+
+plt.bar(x_axis - barwidth, real_freq, label='Real Freq', width=barwidth)
+plt.bar(x_axis, olh_est_freq, label='Est Freq: OLH', width=barwidth)
+plt.ylabel('Normalized Frequency')
+plt.xlabel('Age attribute with domain size = {}'.format(k))
+plt.legend()
+plt.show();
+
+
+# In[ ]:
+
+
+
+