Learning noisy-XOR with Veritable

generate.py

# generate some noisy-XOR data

from random import random

N = 1000
noise = 0.1

data = []
for _ in range(N):
    x1 = random() < 0.5
    x2 = random() < 0.5
    keep = random() > noise
    y = x1 != x2 if keep else x1 == x2
    data.append({ 'x1': x1, 'x2': x2, 'y': y })

run.py

# NOTE: assumes that generate.py has been run, and that an appropriate 
# data variable is defined

import veritable
from veritable.utils import clean_data

schema = {
    'x1': {'type': 'boolean'},
    'x2': {'type': 'boolean'},
    'y':  {'type': 'boolean'}
}
clean_data(data, schema, assign_ids=True)

api = veritable.connect()
table = api.create_table('noisy_XOR', force=True)
table.batch_upload_rows(data)
analysis = table.create_analysis(schema)
analysis.wait()

pr1 = analysis.predict({'x1': False, 'x2': False, 'y': None})
pr1.credible_values('y', p=.05)
# => {False: 0.82, True: 0.18}
pr2 = analysis.predict({'x1': True, 'x2': False, 'y': None})
pr2.credible_values('y', p=.05)
# => {False: 0.12, True: 0.88}
pr3 = analysis.predict({'x1': False, 'x2': True, 'y': None})
pr3.credible_values('y', p=.05)
# => {False: 0.19, True: 0.81}
pr4 = analysis.predict({'x1': True, 'x2': True, 'y': None})
pr4.credible_values('y', p=.05)
# => {False: 0.93, True: 0.07}