def calculate(self, model_func):
self.w_store.append(self.w)
algo = algo_GD.SGD(w_init=self.w, t_max=self.t, a=self.a)
for i in algo:
f = model_func(noise_value=self.noise_data[algo.t - 1])
algo.update(f)
self.w_store.append(algo.w)
tmp = 0
total_w = 0
for k in range(algo.t):
tmp += (1 - self.eta)**k
total_w += ((1 - self.eta)**k) * algo.wstore[k]
self.w = total_w / tmp
self.w_store.append(self.w)
return self.w
import helper
import noise
if __name__ == "__main__":
args = sys.argv
t = int(args[1])
w_init = np.array([3, 3])
_t_max = 3000
f = model_opt.RosenBrock()
w_star = f.w_star
last_w_store = []
iqr_store = []
for i in tqdm(range(t)):
a = 5
noise_data = noise.Pareto(dim=2, n=_t_max, a=a).generate() * 300
iqr = helper.iqr(noise_data)
algo = algo_GD.SGD(w_init=w_init, t_max=_t_max, a=0.00078)
for i in algo:
noise_value = noise_data[algo.t - 1]
f = model_opt.RosenBrock(noise_value=noise_value)
algo.update(model=f)
last_w_store.append(algo.w)
iqr_store.append(iqr)
dt_now = datetime.datetime.now()
last_w_store = np.array(last_w_store)
data = np.array([iqr_store, last_w_store[:, 0], last_w_store[:, 1]]).T
df = pd.DataFrame(data=data, columns=['iqr', 'w_0', 'w_1'])
df.to_csv('pareto_noise/pareto_noise{}_last_w{}.csv'.format(t, dt_now),
header=True)