def kmeans_anneal(X,Y,idx_condensor,name):
import time
import pickle
import numpy as np
from subset_selection.rec_annealing import anneal_optimize
cond_anneal = []
for index in idx_condensor:
cond_anneal.append(anneal_optimize(index, X, Y))
pickle.dump(cond_anneal, open('data_sub/'+name+'_kmeans_anneal.p', 'wb'))
print('saved_kmeans_anneal', len(index))
def random_anneal(X,Y,name,size=100,frac=1.1,tl=60,trial=1):
import time
import pickle
import numpy as np
from subset_selection.rec_annealing import anneal_optimize
#X, Y = pickle.load(open('data_embed/mnist_dim32.p', 'rb'))
cond_anneal = []
while size < len(Y):
for i in range(trial):
index = np.random.choice(list(range(len(Y))), size, replace=False)
cond_anneal.append(anneal_optimize(index, X, Y, tl))
pickle.dump(cond_anneal, open('data_sub/'+name+'_random_anneal.p', 'wb'))
size = int(size * frac)
print('saved_random_anneal', size)
def flat_anneal(X_tr_emb,Y_tr,idx_condensor,name,size = 100):
import pickle
from subset_selection.rec_annealing import anneal_optimize
#data_path = 'data_sub/mnist_tiers.p'
#idx_condensor = index
#print(idx_condensor.shape)
X, Y = X_tr_emb, Y_tr
#size = 100
frac = 1.1
cond_anneal = []
while size < idx_condensor.shape[0]:
index = idx_condensor[:size]
tl = 60
#if size == 100:
# tl = 300
cond_anneal.append(anneal_optimize(index, X, Y, tl))
pickle.dump(cond_anneal, open('data_sub/'+name+'_tiers_anneal.p', 'wb'))
size = int(size * frac)
print('saved_anneal', size)
import time
import pickle
import numpy as np
from subset_selection.rec_annealing import anneal_optimize
data_path = 'data_sub/mnist_tiers.p'
idx_condensor = pickle.load(open(data_path, "rb"))
print(idx_condensor.shape)
X, Y = pickle.load(open('data_embed/mnist_dim32.p', 'rb'))
size = 100
frac = 1.1
cond_anneal = []
while size < idx_condensor.shape[0]:
index = idx_condensor[:size]
tl = 60
if size == 100:
tl = 300
cond_anneal.append(anneal_optimize(index, X, Y, tl))
pickle.dump(cond_anneal, open('data_sub/condensor_anneal.p', 'wb'))
size = int(size * frac)
print('saved', size)