def mutual_information_heatmap(self, attributes: List = None):
if attributes:
private_df = self.private_df[attributes]
synthetic_df = self.synthetic_df[attributes]
else:
private_df = self.private_df
synthetic_df = self.synthetic_df
private_mi = pairwise_attributes_mutual_information(private_df)
synthetic_mi = pairwise_attributes_mutual_information(synthetic_df)
fig = plt.figure(figsize=(15, 6), dpi=120)
fig.suptitle(
'Pairwise Mutual Information Comparison (Private vs Synthetic)',
fontsize=20)
ax1 = fig.add_subplot(121)
ax2 = fig.add_subplot(122)
sns.heatmap(private_mi, ax=ax1, cmap="YlGnBu")
sns.heatmap(synthetic_mi, ax=ax2, cmap="YlGnBu")
ax1.set_title('Private, max=1', fontsize=15)
ax2.set_title('Synthetic, max=1', fontsize=15)
fig.autofmt_xdate()
fig.tight_layout()
plt.subplots_adjust(top=0.83)
plt.savefig("plots/titanic_mutual_information_heatmap.jpg")
def get_heatmap_data(dataset_filename):
df = pd.read_csv(dataset_filename)
values = pairwise_attributes_mutual_information(df)
out = []
attributes = values.columns
for x, xattr in enumerate(attributes):
for y, yattr in enumerate(attributes):
out.append(
[x, y,
int(round(1000 * values.loc[xattr, yattr])) / 1000])
return out
