# REMBO

import numpy as np

import math

import acquisition_function as acq

import create_random_matrix as crm

import choose_bounded_region as cbr

D = 10 # number of features

d = 3 # reducted dimension

n_training = 10

n_test = 1

max_iter = 5 # maximum number of iterations

sigma_0 = 0.1

region_bound = math.sqrt(d)

region_bound_step_size = 0.5

# Sample training input and ouput

#ytrain = np.random.uniform(-5, 5, (n_training, D))

#fytrain = acq.sample_training_output(ytrain)

# Step 1

# Generate random matrix

A = crm.random_matrix(D, d)

# Step 2

# Choose bounded region set

y = cbr.chooseBoundedRegion(d, -regionBound, regionBound, regionBoundStepSize)

#define initial mu and sigma

mu =0

sigma = np.matrix(1)

# Step 3 - 6

#ytrain : D dimensional dataset

#ytest: Y subset

#fytrain: sample from dataset (ytrain)

for t in range(0, max_iter):

number_of_samples = t+1

#Step 3 : sample out of y , and create Y

Y = acq.select_sample_set(number_of_samples,y)

# Select points from bounded box to be tested

#ytest =Y

#ytest = acq.select_test_set(n_test, Y)

# Get mu and sigma

mu, sigma, ybest = acq.gp_posterior(ytrain, sigma, ytest, fytrain, A, t, d, number_of_samples)

# Find ybest

# ybest = acq.gp_optimize(ytest, t, D, mu, sigma, n_test)

# Augment the data

# ytrain, fytrain = acq.augment_data(ytrain, fytrain, ybest, A)

print ybest