def get_problem():
# return gauss_model.get_problem(dim=dim, n=n)
return banana_model.get_problem(dim=dim, a=20, n0=1000, n=n)
import banana_model
import dp_barker
import numpy as np
import matplotlib.pyplot as plt
np.random.seed(4673228)
dim = 2
problem = banana_model.get_problem(dim=dim, a=20, n0=1000, n=100000)
n, data_dim = problem.data.shape
epsilon = 4
delta = 0.1 / n
params = dp_barker.BarkerParams(prop_sigma=np.repeat(0.008 * 15, 2),
batch_size=1300)
res = dp_barker.dp_barker(problem, problem.theta0, epsilon, delta, params)
print("DP mcmc acceptance: {}".format(res.acceptance))
print(f"Clipped: {res.clipped_r}")
mean_error = res.mean_error
cov_error = res.cov_error
mmd = res.mmd
print(f"Mean error: {mean_error}")
print(f"Cov error: {cov_error}")
print(f"MMD: {mmd}")
fig, axes = plt.subplots(dim)
for i in range(dim):
axes[i].plot(res.chain[:, i])
plt.show()
import numpy as np
import scipy.stats as stats
import pandas as pd
import banana_model
import mmd
import argparse
import clipping
parser = argparse.ArgumentParser()
parser.add_argument("clip_bound", type=float)
parser.add_argument("dim", type=int)
parser.add_argument("index", type=int)
parser.add_argument("output", type=str)
args = parser.parse_args()
problem = banana_model.get_problem(dim=args.dim, a=20, n0=None, n=100000)
n, data_dim = problem.data.shape
dim = args.dim
iters = 500
eta = 0.001
L = 15
mass = np.ones(dim)
mass[1] = 0.5
if dim > 2:
mass[2:] = 0.4
clip_bound = args.clip_bound
theta0 = np.zeros(dim)
theta0[1] = 3
# Set the seed for the starting points only based on index
def get_problem(self):
return banana_model.get_problem(self.dim, self.n0, self.a, self.n)