def main(_):
tf.set_random_seed(1234)
sess = tf.Session()
print('---Prepare data...')
### YOUR CODE HERE
data_dir = './'
### END CODE HERE
x_train, y_train, x_test, y_test = load_data(data_dir)
x_train_new, y_train_new, x_valid, y_valid = train_valid_split(
x_train, y_train)
model = Cifar(sess, configure())
### YOUR CODE HERE
# First step: use the train_new set and the valid set to choose hyperparameters.
# Second step: with hyperparameters determined in the first run, re-train
# your model on the original train set.
model.train(x_train, y_train, 50)
# Third step: after re-training, test your model on the test set.
# Report testing accuracy in your hard-copy report.
model.test_or_validate(x_test, y_test, [30, 35, 40, 45, 50])
def main(_):
sess = tf.Session()
print('---Prepare data...')
### YOUR CODE HERE
#data_dir = "/content/drive/My Drive/Colab Notebooks/code/HW3/data/"
data_dir = "/content/drive/My Drive/data"
### END CODE HERE
x_train, y_train, x_test, y_test = load_data(data_dir)
x_train_new, y_train_new, x_valid, y_valid = train_valid_split(
x_train, y_train)
model = Cifar(sess, configure())
### YOUR CODE HERE
# First step: use the train_new set and the valid set to choose hyperparameters.
#model.train(x_train_new, y_train_new, 200)
#model.test_or_validate(x_valid, y_valid, [10,20,30,40,50,100,150,200])
# Second step: with hyperparameters determined in the first run, re-train
# your model on the original train set.
# model.train(x_train, y_train, 200)
# Third step: after re-training, test your model on the test set.
# Report testing accuracy in your hard-copy report.
model.test_or_validate(
x_test, y_test,
[100, 110, 120, 130, 140, 150, 160, 170, 180, 190, 200])
def main(_):
sess = tf.Session()
print('---Prepare data...')
### YOUR CODE HERE
data_dir = "data/"
### END CODE HERE
x_train, y_train, x_test, y_test = load_data(data_dir)
x_train_new, y_train_new, x_valid, y_valid = train_valid_split(
x_train, y_train)
model = Cifar(sess, configure())
### YOUR CODE HERE
# First step: use the train_new set and the valid set to choose hyperparameters.
#
#model.train(x_train_new, y_train_new, 200)
#model.train(x_train_new, y_train_new, 3)
#model.test_or_validate(x_valid, y_valid, [160, 170, 180, 190, 200])
#model.test_or_validate(x_valid, y_valid, [10])
# Second step: with hyperparameters determined in the first run, re-train
# your model on the original train set.
model.train(x_train, y_train, 150)
def main(_):
sess = tf.Session()
print('---Prepare data...')
### YOUR CODE HERE
data_dir = '../cifar-10-batches-py/'
### END CODE HERE
x_train, y_train, x_test, y_test = load_data(data_dir)
x_train_new, y_train_new, x_valid, y_valid = train_valid_split(
x_train, y_train)
model = Cifar(sess, configure())
### YOUR CODE HERE
# First step: use the train_new set and the valid set to choose hyperparameters.
# model.train(x_train_new, y_train_new, 200)
# model.train(x_train_new, y_train_new, 10)
# model.test_or_validate(x_valid, y_valid, [160, 170, 180, 190, 200])
# model.test_or_validate(x_valid, y_valid, list(range(80, 201, 10)))
# model.test_or_validate(x_valid, y_valid, [150])
# Second step: with hyperparameters determined in the first run, re-train
# your model on the original train set.
model.train(x_train, y_train, 140)
# Third step: after re-training, test your model on the test set.
# Report testing accuracy in your hard-copy report.
model.test_or_validate(x_test, y_test, [140])
def main(_):
sess = tf.Session()
print('---Prepare data...')
### YOUR CODE HERE
data_dir = "cifar-10-batches-py"
### END CODE HERE
x_train, y_train, x_test, y_test = load_data(data_dir)
x_train_new, y_train_new, x_valid, y_valid = train_valid_split(
x_train, y_train, 8000)
parse_record(x_train[0], True)
model = Cifar(sess, configure())
### YOUR CODE HERE
# First step: use the train_new set and the valid set to choose hyperparameters.
def del_all_flags(FLAGS):
flags_dict = FLAGS._flags()
keys_list = [keys for keys in flags_dict]
for keys in keys_list:
FLAGS.__delattr__(keys)
# model.train(x_train_new, y_train_new, 3)
sess_test = tf.Session()
del_all_flags(tf.flags.FLAGS)
model_test = Cifar(sess_test, configure())
model_test.test_or_validate(x_valid, y_valid, [1, 2, 3])
def main(_):
sess = tf.Session()
print('---Prepare data...')
### YOUR CODE HERE
data_dir = "/Users/tiandi03/Desktop/dataset/cifar-10-batches-py"
### END CODE HERE
x_train, y_train, x_test, y_test = load_data(data_dir)
x_train_new, y_train_new, x_valid, y_valid = train_valid_split(x_train, y_train)
model = Cifar(sess, configure())
### YOUR CODE HERE
# First step: use the train_new set and the valid set to choose hyperparameters.
model.train(x_train_new, y_train_new, 200)
model.test_or_validate(x_valid, y_valid, [160, 170, 180, 190, 200])
def main(_):
sess = tf.Session()
print('---Prepare data...')
### YOUR CODE HERE
data_dir = os.path.join(os.path.abspath(os.getcwd()),"ResNet/data")
### END CODE HERE
x_train, y_train, x_test, y_test = load_data(data_dir)
x_train_new, y_train_new, x_valid, y_valid = train_valid_split(x_train, y_train)
model = Cifar(sess, configure())
### YOUR CODE HERE
# First step: use the train_new set and the valid set to choose hyperparameters.
model.train(x_train_new, y_train_new, 200)
def main(_):
sess = tf.Session()
print('---Prepare data...')
### YOUR CODE HERE
# Download cifar-10 dataset from https://www.cs.toronto.edu/~kriz/cifar.html
data_dir = "cifar-10-batches-py"
### END CODE HERE
x_train, y_train, x_test, y_test = load_data(data_dir)
x_train_new, y_train_new, x_valid, y_valid = train_valid_split(
x_train, y_train)
model = Cifar(sess, configure())
### YOUR CODE HERE
model.train(x_train, y_train, 40)
model.test_or_validate(x_test, y_test, [5, 10, 15, 20, 25, 30, 35, 40])
def main(_):
sess = tf.Session()
print('---Prepare data...')
### YOUR CODE HERE
data_dir = '../cifar-10-batches-py'
### END CODE HERE
x_train, y_train, x_test, y_test = load_data(data_dir)
x_train_new, y_train_new, x_valid, y_valid = train_valid_split(
x_train, y_train)
model = Cifar(sess, configure())
### YOUR CODE HERE
# from Network import ResNet
# network = ResNet(1, 3, 10, 16)
# ips = tf.placeholder(tf.float32, shape=(100, 32, 32, 3))
# sess.run(tf.global_variables_initializer())
# sess.run(tf.local_variables_initializer())
# net = network(ips,training=True)
# from tensorflow.keras import Model
# model = Model(inputs=ips, outputs=net)
# print(model.summary)
# # print(sess.run(network(ips,training=True)))
# writer = tf.summary.FileWriter('output', sess.graph)
# writer.close()
# First step: use the train_new set and the valid set to choose hyperparameters.
# model.train(x_train_new, y_train_new, 200)
# while True:
# model.train(x_train_new, y_train_new, 600)
# model.test_or_validate(x_valid,y_valid,[i*10 for i in range(1,11)])
# model.test_or_validate(x_valid,y_valid,[20])
# model.test_or_validate(x_valid, y_valid, [160, 170, 180, 190, 200])
# model.test_or_validate(x_valid,y_valid,[10])
# Second step: with hyperparameters determined in the first run, re-train
# your model on the original train set.
# model.train(x_train, y_train, 200)
# Third step: after re-training, test your model on the test set.
# Report testing accuracy in your hard-copy report.
model.test_or_validate(x_test, y_test, [170])
def main(_):
os.environ["CUDA_VISIBLE_DEVICES"] = '3'
sess = tf.Session()
print('---Prepare data...')
x_train, y_train, x_test, y_test = load_data()
x_train_new, y_train_new, x_valid, y_valid \
= train_valid_split(x_train, y_train)
model = MNIST(sess, configure())
### YOUR CODE HERE
# First run: use the train_new set and the valid set to choose
# hyperparameters, like num_hid_layers, num_hid_units, stopping epoch, etc.
# Report chosen hyperparameters in your hard-copy report.
num_hidden_layers = [1, 2, 3]
num_hidden_units = [256, 512, 1024]
batch_sizes = [32, 64, 128]
num_epochs = [1, 5, 10]
for model.conf.num_hid_layers in num_hidden_layers:
for model.conf.num_hid_units in num_hidden_units:
for model.conf.batch_size in batch_sizes:
for epochs in num_epochs:
print("Hidden layers: {}, Hidden units: {}, Batch size: {}, Max epochs: {}".format(model.conf.num_hid_layers, \
model.conf.num_hid_units, model.conf.batch_size, epochs))
model.train(x_train_new,
y_train_new,
x_valid,
y_valid,
epochs,
validation=True)
# Second run: with hyperparameters determined in the first run, re-train
# your model on the original train set.
model.train(x_train, y_train, None, None, 10, validation=False)
# Third run: after re-training, test your model on the test set.
# Report testing accuracy in your hard-copy report.
model.test(x_test, y_test, 10)
def main(_):
os.environ["CUDA_VISIBLE_DEVICES"] = '3'
sess = tf.Session()
print('---Prepare data...')
x_train, y_train, x_test, y_test = load_data()
x_train_new, y_train_new, x_valid, y_valid \
= train_valid_split(x_train, y_train)
model = MNIST(sess, configure())
### YOUR CODE HERE
# First run: use the train_new set and the valid set to choose
# hyperparameters, like num_hid_layers, num_hid_units, stopping epoch, etc.
# Report chosen hyperparameters in your hard-copy report.
model.train(x_train_new,
y_train_new,
x_valid,
y_valid,
max_epoch=1,
validation=True)
def main(_):
os.environ["CUDA_VISIBLE_DEVICES"] = '3'
#sess = tf.Session()
print('---Prepare data...')
x_train, y_train, x_test, y_test = load_data()
x_train_new, y_train_new, x_valid, y_valid \
= train_valid_split(x_train, y_train)
# model = MNIST(sess, conf())
### YOUR CODE HERE
conf = configure()
# First run: use the train_new set and the valid set to choose
# hyperparameters, like num_hid_layers, num_hid_units, stopping epoch, etc.
# Report chosen hyperparameters in your hard-copy report.
params = {
'num_hid_layers': [0, 1, 2, 3, 4, 5],
'num_hid_units': [64, 128, 256, 512],
'max_epoch': [50, 100, 125, 150, 175, 200],
'batch_size': [128, 256, 512, 1024, 2048]
}
#lowest supported by python
best_accuracy = -sys.maxsize -1
best_params = None
for batch_size in params['batch_size']:
conf.batch_size = batch_size
for num_hid_units in params['num_hid_units']:
conf.num_hid_units = num_hid_units
for num_hid_layers in params['num_hid_layers']:
conf.num_hid_layers = num_hid_layers
max_epoch = max(params['max_epoch'])
sess = tf.Session(graph=tf.get_default_graph())
model = MNIST(sess, conf)
model.train(x_train_new, y_train_new, x_valid, y_valid, max_epoch, validation=False)
for epoch in params['max_epoch']:
accuracy = model.test(x_valid, y_valid, epoch)
print ("Accuracy with", num_hid_units, "hidden Units,", batch_size, "batch_size,",
num_hid_layers, "hidden Layers and", epoch, "epoch:", accuracy)
if accuracy > best_accuracy:
best_params = (batch_size, num_hid_units, num_hid_layers, epoch)
best_accuracy = accuracy
sess.close()
tf.reset_default_graph()
print ("Best Accuracy with", best_params[1], "hidden Units,", best_params[0], "batch_size,",
best_params[2], "hidden Layers and", best_params[3], "epoch:", best_accuracy)
# Second run: with hyperparameters determined in the first run, re-train
# your model on the original train set.
sess = tf.Session(graph=tf.get_default_graph())
conf.batch_size, conf.num_hid_units, conf.num_hid_layers, max_epoch = best_params
model = MNIST(sess, conf)
model.train(x_train, x_train, x_valid, y_valid, max_epoch, validation=False)
# Third run: after re-training, test your model on the test set.
# Report testing accuracy in your hard-copy report.
accuracy = model.test(x_test, y_test, max_epoch)
sess.close()