write_to_file = exp_dir + 'exp_stdout.txt'
###########################################################################################
###########################################################################################
# LOAD DATA
if args.dataset == 'clevr':
# # CLEVR DATA
dataset = load_clevr(batch_size=args.batch_size,
vws=args.vws,
quick=args.quick)
elif args.dataset == 'cifar':
# CIFAR DATA
# train_image_dataset = load_cifar(data_dir=args.data_dir)
dataset = load_cifar(data_dir=home + '/Documents/')
print(len(dataset), dataset[0].shape)
###########################################################################################
###########################################################################################
# Init Model
# ------------------------------------------------------------------------------
sampling_batch_size = 64
shape = dataset[0].shape
model = Glow_((sampling_batch_size, shape[0], shape[1], shape[2]), args).cuda()
# print(model)
print("number of model parameters:",
sum([np.prod(p.size()) for p in model.parameters()]))
# fasdfad
# model = nn.DataParallel(model).cuda()
# # CLEVR DATA
if args.machine in ['vws', 'vector', 'vaughn']:
data_dir = home + "/vl_data/two_objects_large/" #vws
else:
data_dir = home + "/VL/data/two_objects_no_occ/" #boltz
train_x, test_x = load_clevr(batch_size=args.batch_size,
data_dir=data_dir,
quick=args.quick)
shape = train_x[0].shape
elif args.dataset == 'cifar':
# CIFAR DATA
# train_image_dataset = load_cifar(data_dir=args.data_dir)
train_x, test_x = load_cifar(data_dir=home + '/Documents/',
dataset_size=args.dataset_size)
shape = train_x[0].shape
# print (len(test_x), 'test set len')
svhn_test_x = load_svhn(data_dir=home + '/Documents/')
# svhn_test_x = test_x
# dataset = train_x
elif args.dataset == 'flickr':
train_x, test_x = load_flickr(
data_dir='/scratch/gobi1/ccremer/Flickr_Faces/images1024x1024/',
data_dir_test=
'/scratch/gobi1/ccremer/Flickr_Faces/images1024x1024_test/',
dataset_size=args.dataset_size)
else:
result_list = average_weights_crossover(crossover, data, x_train,
y_train, x_test, y_test,
num_transplants,
batch_size_activation,
batch_size_sgd, work_id)
return result_list
if __name__ == "__main__":
data = "cifar10"
if data == "cifar10":
x_train, x_test, y_train, y_test = load_cifar()
elif data == "cifar100":
x_train, x_test, y_train, y_test = load_cifar_100()
elif data == "mnist":
x_train, x_test, y_train, y_test = load_mnist()
num_processes = 1
start = timer()
pair_list = [pair for pair in range(num_processes)]
results = crossover_offspring(data, x_train, y_train, x_test, y_test,
pair_list)
pickle.dump(results, open("crossover_results.pickle", "wb"))
print("**************** Training on MNIST *****************")
model = Model()
model.train(x_train_mnist, y_train_mnist)
_ = model.test(x_test_mnist, y_test_mnist)
# To empty the RAM
del x_train_mnist
del y_train_mnist
del x_test_mnist
del y_test_mnist
del model
time.sleep(5)
print("Loading CIFAR-100 ...")
x_train_cifar, y_train_cifar, x_test_cifar, y_test_cifar = load_cifar()
print("Done")
# Training for the cifar images.
print("**************** Training on CIFAR-100 *****************")
model = Model()
model.train(x_train_cifar, y_train_cifar)
_ = model.test(x_test_cifar, y_test_cifar)
# To empty the RAM
del x_train_cifar
del y_train_cifar
del x_test_cifar
del y_test_cifar
del model
# -*- coding: utf-8 -*-
from __future__ import division
import tensorflow as tf
import numpy as np
import os
import shutil
import time
import load_data
x_train, x_validation, x_test, y_train, y_validation, y_test \
= load_data.load_cifar('./data/cifar/', seed=0, as_image=True, scaling=True)
BOARD_PATH = "./board/lab08-10_board"
INPUT_DIM = np.size(x_train, 1)
NCLASS = len(np.unique(y_train))
BATCH_SIZE = 32
TOTAL_EPOCH = 100
ALPHA = 0
INIT_LEARNING_RATE = 0.001
ntrain = len(x_train)
nvalidation = len(x_validation)
ntest = len(x_test)
image_width = np.size(x_train, 1)
image_height = np.size(x_train, 2)
n_channels = np.size(x_train, 3)
import load_data
import plot_helper
x_train, x_validation, x_test, y_train, y_validation, y_test=load_data.load_cifar('./data/cifar/', seed=0, as_image=True, scaling=True)
plot_helper.plot_cifar(x_train, 100)
x_train, x_validation, x_test, y_train, y_validation, y_test=load_data.load_mnist('./data/cmnist/', seed=0, as_image=True, scaling=True)
plot_helper.plot_mnist(x_train, 100)
