def __init__(self, trainintfile, testfile, outputfile ):
loader = DataLoader()
loader.loadTrainingData(trainintfile)
self.features_names = loader.featureNames()
self.images, self.IDs = loader.loadTestData(testfile)
self.df_output = pandas.read_csv(outputfile).drop(['Unnamed: 0'], axis = 1)
pyplot.subplot()
def main(args):
dl = DataLoader()
stem = Stemmer('porter')
# files is a list of files, which are lists of lines, which are lists of words
files = [{element[0]: stem.stem(element[1]) for element in dl.load_data(file) if stem.stem(element[1])} for file in args]
for file, arg in zip(files, args):
print('Processing file {}...'.format(arg))
file = {k: list(v) for k, v in file.items()}
print('Data Clusterer')
test_clusterer(DataClusterer(list(file.values()), 'euclidean'), file)
print('-'*64)
print('Description Clusterer')
test_clusterer(DescriptionClusterer(list(file.values()), 'cosine'), file)
import json
base_dir = '../data/'
dataset_name = sys.argv[1]
#Dataset
if dataset_name == 'bn1':
dataset = pd.read_csv(base_dir + dataset_name + '.csv')
dataset.drop(['Unnamed: 0'], axis=1, inplace=True)
params = {
'dataframe': dataset.copy(),
'continuous_features': ['x1', 'x2', 'x3'],
'outcome_name': 'y'
}
d = DataLoader(params)
train_data_vae = d.data_df.copy()
#train_data_vae.drop('y', axis=1, inplace=True)
columns = train_data_vae.columns
elif dataset_name == 'adult':
dataset = load_adult_income_dataset()
params = {
'dataframe': dataset.copy(),
'continuous_features': ['age', 'hours_per_week'],
'outcome_name': 'income'
}
d = DataLoader(params)
# d = dice_ml.Data(dataframe=dataset, continuous_features=['age', 'hours_per_week'], outcome_name='income')
train_data_vae = d.data_df.copy()
else:
print(" >> Using Cpu")
params.device = torch.device('cpu')
random.seed(args.seed)
torch.manual_seed(args.seed)
params.seed = args.seed
params.tag = args.tag
params.save_dir = args.save_dir
params.batch_size = args.batch_size
params.epoch_num = args.num_epoch
params.save_freq = args.save_freq
params.save_checkpoints = args.save_checkpoints
dataloader = DataLoader(path_to_data=args.train_data, seed=params.seed, shuffle=True)
params.lr = args.lr
params.max_grad_norm = 1.0
params.num_total_steps = (dataloader.size()[0]// params.batch_size) * params.epoch_num
params.num_warmup_steps = args.warmup_steps
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
dataloader.pre_encode(tokenizer)
#model = DistilBertForTokenClassification(2, args.top_rnn) if args.distil else BertForTokenClassification.from_pretrained('bert-base-uncased', num_labels=2)
#model = BertForTokenClassification.from_pretrained('./temp/pytorch_model.bin', num_labels=2)
if args.restore_file is not None:
model = BertForTokenClassification.from_pretrained(args.restore_file, num_labels=2)
else:
model = BertForTokenClassification.from_pretrained('bert-base-uncased', num_labels=2)
def main():
set_random_seed(0)
parser = argparse.ArgumentParser()
parser.add_argument('--exp',
type=str,
default='GenericFastAdapt',
help='config file with parameters of the experiment')
parser.add_argument('--num_workers',
type=int,
default=4,
help='number of data loading workers')
parser.add_argument('--cuda', type=bool, default=True, help='enables cuda')
parser.add_argument(
'--imagedir',
type=str,
default='',
help='path to image directory containing client images')
parser.add_argument('--experts_dir',
type=str,
default='',
help='path to directory containing experts')
args_opt = parser.parse_args()
exp_config_file = os.path.join('.', 'config', args_opt.exp + '.py')
exp_directory = os.path.join(
'.', 'experiments',
args_opt.exp + '-' + args_opt.experts_dir.replace('/', '-'))
# Load the configuration params of the experiment
print('Launching experiment: %s' % exp_config_file)
config = imp.load_source("", exp_config_file).config
config[
'exp_dir'] = exp_directory # the place where logs, models, and other stuff will be stored
config['image_directory'] = args_opt.imagedir
data_test_opt = config['data_test_opt']
dataset_test = GenericDataset(dataset_name=data_test_opt['dataset_name'],
split=data_test_opt['split'],
config=config)
dloader_test = DataLoader(dataset=dataset_test,
batch_size=data_test_opt['batch_size'],
num_workers=args_opt.num_workers,
shuffle=False)
z = {}
algorithm = ClassificationModel(config)
for expert_fname in os.listdir(args_opt.experts_dir):
expert_path = os.path.join(args_opt.experts_dir, expert_fname)
if not os.path.isfile(expert_path):
continue
algorithm.init_network()
algorithm.load_pretrained(expert_path)
if args_opt.cuda:
algorithm.load_to_gpu()
eval_stats = algorithm.evaluate(dloader_test)
z[expert_fname] = eval_stats['prec1']
print(z)
with open(os.path.join(exp_directory, 'z.pickle'), 'wb') as f:
pickle.dump(z, f, protocol=pickle.HIGHEST_PROTOCOL)
#Main Code
base_data_dir = '../data/'
base_model_dir = '../models/'
dataset_name = args.dataset_name
#Dataset
if dataset_name == 'bn1':
dataset = pd.read_csv(base_data_dir + dataset_name + '.csv')
dataset.drop(['Unnamed: 0'], axis=1, inplace=True)
params = {
'dataframe': dataset.copy(),
'continuous_features': ['x1', 'x2', 'x3'],
'outcome_name': 'y'
}
d = DataLoader(params)
elif dataset_name == 'adult':
dataset = load_adult_income_dataset()
params = {
'dataframe': dataset.copy(),
'continuous_features': ['age', 'hours_per_week'],
'outcome_name': 'income'
}
d = DataLoader(params)
#Load Black Box Prediction Model
data_size = len(d.encoded_feature_names)
pred_model = BlackBox(data_size)
path = base_model_dir + dataset_name + '.pth'
pred_model.load_state_dict(torch.load(path))
import cv2
from dataloader import DataLoader
from harrisdetector import harris_corners
from pointTracker import PointTracker
from pointProjection import project_points
import time
from debug.PointsVisualizer import PointVisualizer
# pathen her vil ikke fungere når dette uploades til git. uploader ikke dataset
#rgbd_dataset_freiburg1_rpy
#rgbd_dataset_freiburg1_xyz
#rgbd_dataset_freiburg1_rgb_calibration
dl = DataLoader('dataset/rgbd_dataset_freiburg1_rpy'
) # Edit this string to load a different dataset
tracker = PointTracker()
vis = PointVisualizer()
# Set initial position of cameras in visualizer
initial_orientation, initial_position = dl.get_transform()
vis.set_groundtruth_transform(initial_orientation, initial_position)
vis.set_estimated_transform(initial_orientation, initial_position)
# Get points for the first frame
grey_img = dl.get_greyscale()
depth_img = dl.get_depth()
points_and_response = harris_corners(grey_img)
tracker.add_new_corners(grey_img, points_and_response)
# Project the points in the first frame
def evaluate(im0_path,
restore_path,
output_path,
n_batch=settings.N_BATCH,
n_height=settings.N_HEIGHT,
n_width=settings.N_WIDTH,
n_channel=settings.N_CHANNEL,
n_pyramid=settings.N_PYRAMID,
max_disparity=settings.MAX_DISPARITY,
n_gpu=settings.N_GPU,
n_thread=settings.N_THREAD):
"""Test function."""
# Create dataloader for computation graph
dataloader = DataLoader(shape=[n_batch, n_height, n_width, n_channel],
name='dataloader',
n_thread=n_thread,
prefetch_size=n_thread,
normalize=True,
random_flip=False,
random_gamma=False,
gamma_range=[0.8, 1.2],
random_brightness=False,
brightness_range=[0.5, 2.0],
random_color=False,
color_range=[0.8, 1.2])
# Build model
model = MonoDispNet(dataloader.next_element[0],
dataloader.next_element[1],
n_pyramid=n_pyramid,
max_disparity=max_disparity)
# Start a Tensorflow session
config = tf.ConfigProto(allow_soft_placement=True)
config.gpu_options.allow_growth = True
session = tf.Session(config=config)
# Initialize saver that will be used for restore
train_saver = tf.train.Saver()
# Initialize all variables
session.run(tf.global_variables_initializer())
session.run(tf.local_variables_initializer())
# Restore weights from checkpoint
log('Restoring from: %s' % restore_path)
train_saver.restore(session, restore_path)
# Load the files for evaluation
im0_paths = data_utils.read_paths(im0_path)
n_sample = len(im0_paths)
im0_paths = data_utils.pad_batch(im0_paths, n_batch)
n_step = len(im0_paths) // n_batch
log('Evaluating %d files...' % n_sample)
dataloader.initialize(session,
im0_paths=im0_paths,
im1_paths=im0_paths,
augment=False)
d_arr = np.zeros((n_step * n_batch, n_height, n_width), dtype=np.float32)
start_time = time.time()
for step in range(n_step):
batch_start = step * n_batch
batch_end = step * n_batch + n_batch
d = session.run(model.model0[0])
d_arr[batch_start:batch_end, :, :] = d[:, :, :, 0]
end_time = time.time()
log('Total time: %.1f ms Average time per image: %.1f ms' %
(1000 * (end_time - start_time),
(1000 * (end_time - start_time) / n_sample)))
d_arr = d_arr[0:n_sample, :, :]
output_path = os.path.join(output_path, 'disparities.npy')
log('Storing predictions to %s' % output_path)
if not os.path.exists(os.path.dirname(output_path)):
os.makedirs(os.path.dirname(output_path))
np.save(output_path, d_arr)
# read config file
with open('./configs/train_params.yaml') as config_file:
config = yaml.load(config_file, Loader=yaml.FullLoader)
config = Dict(config)
train_df = pd.read_csv('../data/train_dataset.csv')
test_df = pd.read_csv('../data/test_dataset.csv')
train_dataset = Dataset(df = train_df,
image_folder_path='../data/images',
augmentations=get_training_augmentation())
test_dataset = Dataset(df = test_df,
image_folder_path='../data/images',
augmentations=get_test_augmentation())
train_dataloader = DataLoader(dataset=train_dataset, batch_size=config.batch_size, shuffle=False)
test_dataloader = DataLoader(dataset=test_dataset, batch_size=config.batch_size, shuffle=False)
model = get_model()
optimizer = tf.keras.optimizers.Adam(learning_rate = config.learning_rate)
loss = tf.keras.losses.CategoricalCrossentropy()
model.compile(optimizer=optimizer, loss = loss, metrics=['accuracy'])
model.fit(x = train_dataloader, steps_per_epoch=len(train_dataloader), epochs=config.epochs,
validation_data = test_dataloader, validation_steps=len(test_dataloader))
def run(final_file, kaoqin_dir, jixiao_dir, fenc='utf8'):
dl = DataLoader()
dl.load(kaoqin_dir, jixiao_dir, final_file, fenc)
dl.check()
def main():
""" Train the Yolov3 Model """
# If there checkpoint is already, assign checkpoint=checkpoint_file
checkpoint = None
# Set epochs, load the data and the trainable model
start_epoch = 0
end_epoch = 7000
learning_rate = 1e-3
batch_size = 6
model = DarkNet()
data = DataLoader(416, "data/train")
dataloader = torch.utils.data.DataLoader(dataset=data,
batch_size=batch_size,
num_workers=0,
shuffle=True)
model = model.to("cuda")
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
# If there's a checkpoint, load its values
if checkpoint != None:
model.load_state_dict(torch.load(checkpoint)['state_dict'])
optimizer.load_state_dict(torch.load(checkpoint)['optimizer'])
start_epoch = torch.load(checkpoint)['epoch']
for param in model.parameters():
param.requires_grad = True
count = 0
x_y = []
w_h = []
conf_loss = []
final_loss = []
# Train the model
print("Starting Training..")
for epoch in range(start_epoch, end_epoch):
print(
"------------------------------------------------------------------------------------------------------------"
)
for batch_id, (imgs, target) in enumerate(dataloader):
imgs = imgs.cuda()
target = target.cuda()
optimizer.zero_grad()
loss = model(imgs, target)
loss.backward()
optimizer.step()
if batch_id % 10 == 0:
print(
"Epoch %d/%d || Batch %d || Overall Loss %.2f || X-Loss %.2f || Y-Loss %.2f || W-Loss %.2f || H-Loss %.2f"
%
(epoch, end_epoch, batch_id, loss.item(), model.losses[0],
model.losses[1], model.losses[2], model.losses[3]))
x_y.append(model.losses[0] + model.losses[1])
w_h.append(model.losses[2] + model.losses[3])
conf_loss.append(model.losses[4])
final_loss.append(loss.item())
# Plot the graph to check if the loss is decreasing through the epochs
# X-Y Loss
plt.plot(x_y, label='X and Y')
plt.savefig('x-y-loss.png')
plt.close()
# W-H Loss
plt.plot(w_h, label='W and H')
plt.savefig('w-h-loss.png')
plt.close()
# Confidence Loss
plt.plot(conf_loss, label='Conf')
plt.savefig('conf-loss.png')
plt.close()
# Overall Loss
plt.plot(final_loss, label='Loss')
plt.savefig('final-loss.png')
plt.show()
plt.close()
# Save the model as checkpoint
torch.save(
{
'epoch': epoch,
'state_dict': model.state_dict(),
'optimizer': optimizer.state_dict()
}, 'checkpoints/checkpoint.epoch.{}.pth.tar'.format(epoch))
dataset_train = GenericDataset(
dataset_name=data_train_opt['dataset_name'],
split=data_train_opt['split'],
random_sized_crop=data_train_opt['random_sized_crop'])
dataset_test = GenericDataset(
dataset_name=data_test_opt['dataset_name'],
split=data_test_opt['split'],
random_sized_crop=data_test_opt['random_sized_crop'])
dloader_train = DataLoader(
dataset=dataset_train,
batch_size=data_train_opt['batch_size'],
unsupervised=data_train_opt['unsupervised'],
epoch_size=data_train_opt['epoch_size'],
num_workers=args_opt.num_workers,
fillcolor=(128,128,128),
resample=PIL.Image.BILINEAR,
shuffle=True)
dloader_test = DataLoader(
dataset=dataset_test,
batch_size=data_test_opt['batch_size'],
unsupervised=data_test_opt['unsupervised'],
epoch_size=data_test_opt['epoch_size'],
num_workers=args_opt.num_workers,
fillcolor=(128,128,128),
resample=PIL.Image.BILINEAR,
shuffle=False)
dataset_train = GenericDataset(
dataset_name=data_train_opt['dataset_name'],
split=data_train_opt['split'],
random_sized_crop=data_train_opt['random_sized_crop'],
num_imgs_per_cat=num_imgs_per_cat)
dataset_test = GenericDataset(
dataset_name=data_test_opt['dataset_name'],
split=data_test_opt['split'],
random_sized_crop=data_test_opt['random_sized_crop'])
dloader_train = DataLoader(
dataset=dataset_train,
batch_size=data_train_opt['batch_size'],
unsupervised=data_train_opt['unsupervised'],
epoch_size=data_train_opt['epoch_size'],
num_workers=args_opt.num_workers,
shuffle=True)
dloader_test = DataLoader(
dataset=dataset_test,
batch_size=data_test_opt['batch_size'],
unsupervised=data_test_opt['unsupervised'],
epoch_size=data_test_opt['epoch_size'],
num_workers=args_opt.num_workers,
shuffle=False)
config['disp_step'] = args_opt.disp_step
algorithm = getattr(alg, config['algorithm_type'])(config)
if args_opt.cuda: # enable cuda
if args.train:
#---create datasets---#
dataset = Dataset(args.dataset_path,
args.index_path,
seg_len=hps.seg_len)
sourceset = Dataset(args.dataset_path,
args.index_source_path,
seg_len=hps.seg_len)
targetset = Dataset(args.dataset_path,
args.index_target_path,
seg_len=hps.seg_len)
#---create data loaders---#
data_loader = DataLoader(dataset, hps.batch_size)
source_loader = DataLoader(sourceset, hps.batch_size)
target_loader = DataLoader(targetset, hps.batch_size)
#---handle paths---#
os.makedirs(args.ckpt_dir, exist_ok=True)
model_path = os.path.join(args.ckpt_dir, args.model_name)
#---initialize trainer---#
trainer = Trainer(hps, data_loader, args.targeted_G, args.one_hot,
args.binary_output, args.binary_ver)
if args.load_model:
trainer.load_model(os.path.join(args.ckpt_dir,
args.load_train_model_name),
model_all=False)
def train(model_config,
data_config,
output_path,
device,
epoch_size,
max_epoch,
batch_size,
repeats,
decade_rate,
clip_grad,
log_every,
valid_every,
learning_rate=0.0005):
print('use device: %s' % device, file=sys.stderr)
vocab = Vocab.load(data_config["vacab_file"])
model = NMT(vocab=vocab, **model_config)
model = model.to(torch.device(device))
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
data_config.pop("vacab_file", None)
data_loader = DataLoader(**data_config)
batch_queue, loss_queue = data_loader.load_train_data(
epoch_size, max_epoch, batch_size, repeats, decade_rate)
dev_data = data_loader.load_dev_data()
hist_valid_scores = []
train_losses = []
train_iter = cum_loss = report_loss = cum_tgt_words = report_tgt_words = 0
cum_examples = report_examples = epoch = valid_num = 0
if os.path.isfile(output_path + '/speech-to-text.model'):
print('loading saved model...')
params = torch.load(output_path + '/speech-to-text.model',
map_location=lambda storage, loc: storage)
model.load_state_dict(params['state_dict'])
print('restoring parameters of the optimizers', file=sys.stderr)
optimizer.load_state_dict(
torch.load(output_path + '/speech-to-text.optim'))
dev_ppl = evaluate_ppl(
model, dev_data,
batch_size=128) # dev batch size can be a bit larger
valid_metric = -dev_ppl
hist_valid_scores.append(valid_metric)
print("saved model ppl: ", dev_ppl)
model.train()
train_time = begin_time = time.time()
epoch, voices, tgt_sents = batch_queue.get(True)
while voices is not None and tgt_sents is not None:
train_iter += 1
optimizer.zero_grad()
# print("received voices:", len(voices))
# print("tgt_sents[0]:", len(tgt_sents[0]), tgt_sents[0])
# print("tgt_sents[1]:", len(tgt_sents[1]), tgt_sents[1])
optimizer.zero_grad()
batch_size = len(voices)
sample_losses = -model(voices, tgt_sents)
batch_loss = sample_losses.sum()
loss = batch_loss / batch_size
loss.backward()
# clip gradient
grad_norm = torch.nn.utils.clip_grad_norm_(model.parameters(),
clip_grad)
optimizer.step()
batch_losses_val = batch_loss.item()
report_loss += batch_losses_val
cum_loss += batch_losses_val
tgt_words_num_to_predict = sum(
len(s[1:]) for s in tgt_sents) # omitting leading `<s>`
report_tgt_words += tgt_words_num_to_predict
cum_tgt_words += tgt_words_num_to_predict
report_examples += batch_size
cum_examples += batch_size
loss_queue.put(report_loss / report_examples)
train_losses.append({
'epoch':
epoch,
'iter':
train_iter,
'loss':
report_loss / report_examples,
'ppl':
math.exp(report_loss / report_tgt_words),
'cum':
cum_examples,
'speed':
report_tgt_words / (time.time() - train_time)
})
if train_iter % log_every == 0:
print(
'epoch %d, iter %d, avg. loss %.2f, avg. ppl %.2f '
'cum. examples %d, speed %.2f words/sec, time elapsed %.2f sec'
% (epoch, train_iter, report_loss / report_examples,
math.exp(report_loss / report_tgt_words), cum_examples,
report_tgt_words /
(time.time() - train_time), time.time() - begin_time),
file=sys.stderr)
train_time = time.time()
report_loss = report_tgt_words = report_examples = 0.
# perform validation
if train_iter % valid_every == 0:
print(
'epoch %d, iter %d, cum. loss %.2f, cum. ppl %.2f cum. examples %d'
% (epoch, train_iter, cum_loss / cum_examples,
np.exp(cum_loss / cum_tgt_words), cum_examples),
file=sys.stderr)
cum_loss = cum_examples = cum_tgt_words = 0.
valid_num += 1
print('begin validation ...', file=sys.stderr)
# compute dev. ppl and bleu
dev_ppl = evaluate_ppl(
model, dev_data,
batch_size=128) # dev batch size can be a bit larger
valid_metric = -dev_ppl
print('validation: iter %d, dev. ppl %f' % (train_iter, dev_ppl),
file=sys.stderr)
is_better = len(hist_valid_scores
) == 0 or valid_metric > max(hist_valid_scores)
hist_valid_scores.append(valid_metric)
if is_better:
patience = 0
print('save currently the best model to [%s]' % output_path,
file=sys.stderr)
model.save(output_path + '/speech-to-text.model')
torch.save(optimizer.state_dict(),
output_path + '/speech-to-text.optim')
epoch, voices, tgt_sents = batch_queue.get(True)
def submit(ypred):
from dataloader import DataLoader
dataloader = DataLoader(config)
df_test = dataloader.load_test_data()
id_test = df_test['id']
pd.DataFrame({"id": id_test, "relevance": ypred}).to_csv('submission.csv',index=False)
def __init__(self,
dataset_name="celeba",
img_size=28,
channels=3,
backup_dir="backup"):
self.dataset_name = dataset_name
self.img_size = img_size
self.channels = channels
self.backup_dir = backup_dir
self.time = time()
# Input shape
self.latent_dim = 100
self.learning_rate = 1e-3
self.gf = 64 # filter size of generator's last layer
self.df = 64 # filter size of discriminator's first layer
optimizer_disc = Adam(self.learning_rate / 10,
beta_1=0.5,
decay=0.00005)
optimizer_gen = Adam(self.learning_rate, beta_1=0.5, decay=0.00005)
# Configure data loader
self.dl = DataLoader(dataset_name=self.dataset_name,
img_res=(self.img_size, self.img_size),
mem_load=True)
self.n_data = self.dl.get_n_data()
# Build generator
self.generator = self.build_generator()
print(
"---------------------generator summary----------------------------"
)
self.generator.summary()
# Build and compile the discriminator
self.discriminator = self.build_discriminator()
print(
"\n---------------------discriminator summary----------------------------"
)
self.discriminator.summary()
self.discriminator.compile(loss='binary_crossentropy',
optimizer=optimizer_disc,
metrics=['accuracy'])
z = Input(shape=(self.latent_dim, ))
fake_img = self.generator(z)
# for the combined model, we only train ganerator
self.discriminator.trainable = False
validity = self.discriminator(fake_img)
# Build combined model
self.combined = Model(z, validity)
print(
"\n---------------------combined summary----------------------------"
)
self.combined.summary()
self.combined.compile(loss=['binary_crossentropy'],
optimizer=optimizer_gen)
sys.stdout.flush()
gt, vm = data_utils.load_depth_with_validity_map(gt_paths[idx])
gt = np.concatenate(
[np.expand_dims(gt, axis=-1),
np.expand_dims(vm, axis=-1)],
axis=-1)
gt_arr.append(gt)
print('Completed loading {} groundtruth depth maps'.format(n_sample))
with tf.Graph().as_default():
# Initialize dataloader
dataloader = DataLoader(
shape=[args.n_batch, args.n_height, args.n_width, 3],
name='dataloader',
is_training=False,
n_thread=args.n_thread,
prefetch_size=2 * args.n_thread)
# Fetch the input from dataloader
im0 = dataloader.next_element[0]
sz0 = dataloader.next_element[3]
# Build computation graph
model = VOICEDModel(im0,
im0,
im0,
sz0,
None,
is_training=False,
occ_threshold=args.occ_threshold,
occ_ksize=args.occ_ksize,
import utils
from dataloader import DataLoader
from model import Model
import tensorflow as tf
import os
import numpy as np
# set batch size
test_batch_size = 52
# create dataloader object. this will be used to get the testing examples
dataloader = DataLoader('dataset.h5', test_batch_size, train=False)
# create the model
model = Model(test_batch_size)
# create tensorflow session
sess = tf.InteractiveSession()
# create new saver to restore the model
saver = tf.train.Saver()
# if there is a checkpoint file, load it and populate the weights and biases.
if os.path.isfile('checkpoints/model.ckpt'):
saver.restore(sess, 'checkpoints/model.ckpt')
print("Model restored!")
else:
# otherwise, set all parameters to a small random value
sess.run(tf.initialize_all_variables())
# making sure read pointer is at zero
dataloader.reset_read_pointer()
class DCGAN():
def __init__(self,
dataset_name="celeba",
img_size=28,
channels=3,
backup_dir="backup"):
self.dataset_name = dataset_name
self.img_size = img_size
self.channels = channels
self.backup_dir = backup_dir
self.time = time()
# Input shape
self.latent_dim = 100
self.learning_rate = 1e-3
self.gf = 64 # filter size of generator's last layer
self.df = 64 # filter size of discriminator's first layer
optimizer_disc = Adam(self.learning_rate / 10,
beta_1=0.5,
decay=0.00005)
optimizer_gen = Adam(self.learning_rate, beta_1=0.5, decay=0.00005)
# Configure data loader
self.dl = DataLoader(dataset_name=self.dataset_name,
img_res=(self.img_size, self.img_size),
mem_load=True)
self.n_data = self.dl.get_n_data()
# Build generator
self.generator = self.build_generator()
print(
"---------------------generator summary----------------------------"
)
self.generator.summary()
# Build and compile the discriminator
self.discriminator = self.build_discriminator()
print(
"\n---------------------discriminator summary----------------------------"
)
self.discriminator.summary()
self.discriminator.compile(loss='binary_crossentropy',
optimizer=optimizer_disc,
metrics=['accuracy'])
z = Input(shape=(self.latent_dim, ))
fake_img = self.generator(z)
# for the combined model, we only train ganerator
self.discriminator.trainable = False
validity = self.discriminator(fake_img)
# Build combined model
self.combined = Model(z, validity)
print(
"\n---------------------combined summary----------------------------"
)
self.combined.summary()
self.combined.compile(loss=['binary_crossentropy'],
optimizer=optimizer_gen)
def build_generator(self):
noise = Input(shape=(self.latent_dim, ))
def deconv2d(layer_input,
filters=256,
kernel_size=(5, 5),
strides=(2, 2),
bn_relu=True):
"""Layers used during upsampling"""
u = Conv2DTranspose(filters,
kernel_size=kernel_size,
strides=strides,
padding='same',
use_bias=False)(layer_input)
if bn_relu:
u = BatchNormalization()(u)
u = Activation("relu")(u)
return u
generator = Dense(4 * self.gf * self.img_size // 4 * self.img_size //
4,
use_bias=False)(noise)
generator = BatchNormalization()(generator)
generator = Activation('relu')(generator)
generator = Reshape(
(self.img_size // 4, self.img_size // 4, self.gf * 4))(generator)
generator = deconv2d(generator, filters=self.gf * 2)
generator = deconv2d(generator, filters=self.gf)
generator = deconv2d(generator,
filters=self.channels,
kernel_size=(3, 3),
strides=(1, 1),
bn_relu=False)
generator = Activation('tanh')(generator)
return Model(noise, generator)
def build_discriminator(self):
def d_block(layer_input, filters, strides=1, bn=True):
"""Discriminator layer"""
d = Conv2D(filters,
kernel_size=3,
strides=strides,
padding='same',
use_bias=False)(layer_input)
if bn:
d = Dropout(rate=0.8)(d)
d = LeakyReLU(alpha=0.2)(d)
return d
# Input img = generated image
d0 = Input(shape=(self.img_size, self.img_size, self.channels))
d = d_block(d0, self.df, strides=2, bn=False)
d = d_block(d, self.df * 2, strides=2)
d = Flatten()(d)
validity = Dense(1, activation='sigmoid')(d)
return Model(d0, validity)
def train(self, epochs, batch_size, sample_interval):
def named_logs(model, logs):
result = {}
for l in zip(model.metrics_names, logs):
result[l[0]] = l[1]
return result
start_time = datetime.datetime.now()
valid = np.ones((batch_size, 1))
fake = np.zeros((batch_size, 1))
max_iter = int(self.n_data / batch_size)
os.makedirs(f"{self.backup_dir}/logs/{self.time}", exist_ok=True)
tensorboard = TensorBoard(f"{self.backup_dir}/logs/{self.time}")
tensorboard.set_model(self.generator)
os.makedirs(f"{self.backup_dir}/models/{self.time}/", exist_ok=True)
with open(
f"{self.backup_dir}/models/{self.time}/generator_architecture.json",
"w") as f:
f.write(self.generator.to_json())
print(
f"\nbatch size : {batch_size} | num_data : {self.n_data} | max iteration : {max_iter} | time : {self.time} \n"
)
for epoch in range(1, epochs + 1):
for iter in range(max_iter):
# ------------------
# Train Generator
# ------------------
ref_imgs = self.dl.load_data(batch_size)
noise = np.random.normal(0, 1, (batch_size, self.latent_dim))
gen_imgs = self.generator.predict(noise)
make_trainable(self.discriminator, True)
d_loss_real = self.discriminator.train_on_batch(
ref_imgs, valid * 0.9) # label smoothing *0.9
d_loss_fake = self.discriminator.train_on_batch(gen_imgs, fake)
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
make_trainable(self.discriminator, False)
logs = self.combined.train_on_batch([noise], [valid])
tensorboard.on_epoch_end(iter,
named_logs(self.combined, [logs]))
if iter % (sample_interval // 10) == 0:
elapsed_time = datetime.datetime.now() - start_time
print(
f"epoch:{epoch} | iter : {iter} / {max_iter} | time : {elapsed_time} | g_loss : {logs} | d_loss : {d_loss} "
)
if (iter + 1) % sample_interval == 0:
self.sample_images(epoch, iter + 1)
# save weights after every epoch
self.generator.save_weights(
f"{self.backup_dir}/models/{self.time}/generator_epoch{epoch}_weights.h5"
)
def sample_images(self, epoch, iter):
os.makedirs(f'{self.backup_dir}/samples/{self.time}', exist_ok=True)
r, c = 5, 5
noise = np.random.normal(0, 1, (r * c, self.latent_dim))
gen_img = self.generator.predict(noise)
# Rescale images 0 - 1
gen_img = 0.5 * gen_img + 0.5
# Save generated images and the high resolution originals
fig, axs = plt.subplots(r, c)
for row in range(r):
for col in range(c):
axs[row, col].imshow(gen_img[5 * row + col, :, :, :])
axs[row, col].axis('off')
fig.savefig(
f"{self.backup_dir}/samples/{self.time}/e{epoch}-i{iter}.png",
bbox_inches='tight',
dpi=100)
# plt.show() # only when running in ipython, otherwise halts the execution
plt.close()
# Placeholder and variables
# TODO : declare placeholder and variables
# Build model
# TODO : build your model here
# Loss and optimizer
# TODO : declare loss and optimizer operation
# Train and evaluate
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
if is_train_mode:
train_dataloader = DataLoader(file_path='dataset/track_metadata.csv', batch_size=batch_size,
label_column_name=label_column_name, is_training=True)
for epoch in range(epoch_num):
total_batch = train_dataloader.num_batch
for i in range(total_batch):
batch_x, batch_y = train_dataloader.next_batch()
# TODO: do some train step code here
print('Training finished !')
output_dir = checkpoint_path + '/run-%02d%02d-%02d%02d' % tuple(localtime(time()))[1:5]
if not gfile.Exists(output_dir):
gfile.MakeDirs(output_dir)
saver.save(sess, output_dir)
print('Model saved in file : %s'%output_dir)
else:
def main(args):
device = Config.device
print("PyTorch running with device {0}".format(device))
if args.download:
print("Downloading data")
download_required_data()
if args.lemmatize:
caption_file = 'data/Flickr_Data/Flickr_TextData/Flickr8k.lemma.token.txt'
else:
caption_file = 'data/Flickr_Data/Flickr_TextData/Flickr8k.token.txt'
print("Generating word2id")
word2id = generate_word2id(caption_file)
id2word = dict([(v, k) for k, v in word2id.items()])
print("Loading Encoder and Decoder")
encoder = Encoder(Config.encoded_size, Config.encoder_finetune)
decoder = Decoder(Config.encoder_dim,
Config.decoder_dim,
Config.attention_dim,
Config.embed_dim,
vocab_size=len(word2id),
dropout=Config.dropout,
embedding_finetune=Config.embedding_finetune)
if args.model_path:
print("Loading model from model_path")
load_model(encoder, decoder, args.model_path)
else:
# no model path, so load pretrained embedding
print("Generating embedding from pretrained embedding file")
embedding = load_pretrained_embedding(
'data/glove.6B.{}d.txt'.format(Config.embed_dim), word2id,
Config.embed_dim)
decoder.load_embedding(embedding)
if not args.test:
# train
print("Loading DataLoader and Trainer")
dloader = DataLoader(caption_file, 'data/Flickr_Data/Images')
trainer = Trainer(encoder, decoder, dloader)
print("Start Training")
loss_history = trainer.train(Config.num_epochs)
plt.plot(np.arange(len(loss_history)), loss_history, label='Loss')
plt.legend()
plt.show()
else:
# test
assert args.image_path
encoder.eval()
decoder.eval()
transform = transforms.Compose([
transforms.Resize(224),
transforms.CenterCrop(224),
transforms.ToTensor()
])
image = transform(Image.open(args.image_path))
image = image.unsqueeze(0)
# TODO
# generate caption from an image
encoder_output = encoder(image)
captions, alphas = decoder.generate_caption_greedily(encoder_output)
caption_in_word = ' '.join(list(map(id2word.get, captions[1:])))
plt.imshow(image[0].numpy().transpose(1, 2, 0))
plt.title(caption_in_word)
plt.axis('off')
plt.show()
print(caption_in_word)
import numpy
import pandas
from pandas import *
import matplotlib
import matplotlib.pyplot as plt
matplotlib.style.use('ggplot')
from dataloader import DataLoader
loader = DataLoader()
from __future__ import division
data_key = loader.get_key_data()
ascii_num = 99
raw_of_all = data_key[data_key['intent_code_point']==ascii_num]
raw_of_alphabet = data_key[data_key['intent_code_point']==ascii_num][data_key['code_point']!=ascii_num]
hands = ['Both', 'Right', 'Left']
result = []
for hand in hands:
data_all = raw_of_all[raw_of_all['input_posture']==hand]
data_all = data_all.groupby('keyboard_condition').count()['time']
data_of_alphabet = raw_of_alphabet[raw_of_alphabet['input_posture']==hand]
use_cuda = True
if use_cuda:
cuda.empty_cache()
""" training mode"""
results = []
f = 3
model = CDKT()
if use_cuda:
model = model.cuda()
optimizer = optim.Adam(model.parameters(),5*1e-4)
DL = DataLoader(read_data(f'/data/train.{f}.dat'),load_init())
for r in range(10): # 20-epochs
i = 0
for x,y in DL.samples(72):
X = tensor(x)
Y = tensor(y)
if use_cuda:
X = X.cuda()
Y = Y.cuda()
loss = model.forward(X,Y,True)
optimizer.zero_grad()
clip_grad_value_(model.parameters(),10)
loss.backward()
optimizer.step()
if opt.method == 'policy-MPUR':
policy_network_mpur = torch.load(model_path)['model']
policy_network_mpur.stats = stats
forward_model.policy_net = policy_network_mpur.policy_net
forward_model.policy_net.stats = stats
forward_model.policy_net.actor_critic = False
forward_model.intype('gpu')
forward_model.stats = stats
if 'ten' in opt.mfile:
forward_model.p_z = torch.load(
path.join(opt.model_dir, f'{opt.mfile}.pz'))
return forward_model, value_function, policy_network_il, policy_network_mper, stats
dataloader = DataLoader(None, opt, 'i80')
forward_model, value_function, policy_network_il, policy_network_mper, data_stats = load_models(
)
splits = torch.load(path.join(data_path, 'splits.pth'))
if opt.u_reg > 0.0:
forward_model.train()
forward_model.opt.u_hinge = opt.u_hinge
if hasattr(forward_model, 'value_function'):
forward_model.value_function.train()
planning.estimate_uncertainty_stats(forward_model,
dataloader,
n_batches=50,
npred=opt.npred)
gym.envs.registration.register(id='I-80-v1',
def train(trn_im0_path,
trn_im1_path,
n_epoch=settings.N_EPOCH,
n_batch=settings.N_BATCH,
n_height=settings.N_HEIGHT,
n_width=settings.N_WIDTH,
n_channel=settings.N_CHANNEL,
learning_rates=settings.LEARNING_RATES,
learning_bounds=settings.LEARNING_BOUNDS,
n_pyramid=settings.N_PYRAMID,
max_disparity=settings.MAX_DISPARITY,
w_ph=settings.W_PH,
w_st=settings.W_ST,
w_sm=settings.W_SM,
w_bc=settings.W_BC,
w_ar=settings.W_AR,
n_checkpoint=settings.N_CHECKPOINT,
n_summary=settings.N_SUMMARY,
checkpoint_path=settings.CHECKPOINT_PATH,
restore_path=settings.RESTORE_PATH,
n_gpu=settings.N_GPU,
n_thread=settings.N_THREAD):
event_path = os.path.join(checkpoint_path, 'event')
model_path = os.path.join(checkpoint_path, 'model.ckpt')
log_path = os.path.join(checkpoint_path, 'results.txt')
# Load image paths from paths file for training and validation
trn_im0_paths = data_utils.read_paths(trn_im0_path)
trn_im1_paths = data_utils.read_paths(trn_im1_path)
n_trn_sample = len(trn_im0_paths)
n_trn_step = n_epoch * np.ceil(n_trn_sample / n_batch).astype(np.int32)
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
# Initialize optimizer
boundaries = [np.int32(b * n_trn_step) for b in learning_bounds]
learning_rate = tf.train.piecewise_constant(global_step, boundaries,
learning_rates)
optimizer = tf.train.AdamOptimizer(learning_rate)
# Initialize dataloader
dataloader = DataLoader(shape=[n_batch, n_height, n_width, n_channel],
name='dataloader',
n_thread=n_thread,
prefetch_size=8,
normalize=True,
random_flip=True,
random_gamma=True,
gamma_range=[0.8, 1.2],
random_brightness=True,
brightness_range=[0.5, 2.0],
random_color=True,
color_range=[0.8, 1.2])
# Split data into towers for each GPU
im0_split = tf.split(dataloader.next_element[0], n_gpu, 0)
im1_split = tf.split(dataloader.next_element[1], n_gpu, 0)
# Build computation graph
tower_gradients = []
tower_losses = []
with tf.variable_scope(tf.get_variable_scope()):
for i in range(n_gpu):
with tf.device('/gpu:%d' % i):
params = []
model = MonoDispNet(im0_split[i],
im1_split[i],
n_pyramid=n_pyramid,
w_ph=w_ph,
w_st=w_st,
w_sm=w_sm,
w_bc=w_bc,
w_ar=w_ar,
max_disparity=max_disparity,
reuse_variables=tf.AUTO_REUSE,
model_index=i)
loss = model.total_loss
tower_losses.append(loss)
tower_gradients.append(optimizer.compute_gradients(loss))
# Set up gradient computations
avg_gradients = average_gradients(tower_gradients)
gradients = optimizer.apply_gradients(avg_gradients,
global_step=global_step)
total_loss = tf.reduce_mean(tower_losses)
tf.summary.scalar('learning_rate', learning_rate, ['model_0'])
tf.summary.scalar('total_loss', total_loss, ['model_0'])
trn_summary = tf.summary.merge_all('model_0')
# Count trainable parameters
n_parameter = 0
for variable in tf.trainable_variables():
n_parameter += np.array(variable.get_shape().as_list()).prod()
# Log network parameters
log('Network Parameters:', log_path)
log(
'n_batch=%d n_height=%d n_width=%d n_channel=%d ' %
(n_batch, n_height, n_width, n_channel), log_path)
log('n_pyramid=%d max_disparity=%.3f' % (n_pyramid, max_disparity),
log_path)
log(
'n_sample=%d n_epoch=%d n_step=%d n_param=%d' %
(n_trn_sample, n_epoch, n_trn_step, n_parameter), log_path)
log(
'learning_rates=[%s]' % ', '.join('{:.6f}'.format(r)
for r in learning_rates),
log_path)
log(
'boundaries=[%s]' %
', '.join('{:.2f}:{}'.format(l, b)
for l, b in zip(learning_bounds, boundaries)), log_path)
log(
'w_ph=%.3f w_st=%.3f w_sm=%.3f w_bc=%.3f w_ar=%.3f' %
(w_ph, w_st, w_sm, w_bc, w_ar), log_path)
log(
'Restoring from: %s' %
('None' if restore_path == '' else restore_path), log_path)
# Initialize Tensorflow session
config = tf.ConfigProto(allow_soft_placement=True)
config.gpu_options.allow_growth = True
session = tf.Session(config=config)
# Initialize saver for storing and restoring checkpoints
summary_writer = tf.summary.FileWriter(model_path, session.graph)
train_saver = tf.train.Saver()
# Initialize all variables
session.run(tf.global_variables_initializer())
session.run(tf.local_variables_initializer())
# If given, load the weights from the restore path
if restore_path != '':
train_saver.restore(session, restore_path)
# Begin training
log('Begin training...')
start_step = global_step.eval(session=session)
time_start = time.time()
trn_step = start_step
while trn_step < n_trn_step:
trn_im0_paths_epoch, trn_im1_paths_epoch = data_utils.shuffle_and_drop(
[trn_im0_paths, trn_im1_paths], n_batch)
dataloader.initialize(session,
im0_paths=trn_im0_paths_epoch,
im1_paths=trn_im1_paths_epoch,
augment=True)
while trn_step < n_trn_step:
try:
_, loss_value = session.run([gradients, total_loss])
if trn_step % n_summary == 0:
summary = session.run(trn_summary)
summary_writer.add_summary(summary,
global_step=trn_step)
if trn_step and trn_step % n_checkpoint == 0:
time_elapse = (time.time() -
time_start) / 3600 * trn_step / (
trn_step - start_step + 1)
time_remain = (n_trn_step / trn_step -
1.0) * time_elapse
checkpoint_log = 'batch {:>6} loss: {:.5f} time elapsed: {:.2f}h time left: {:.2f}h'
log(
checkpoint_log.format(trn_step, loss_value,
time_elapse, time_remain),
log_path)
train_saver.save(session,
model_path,
global_step=trn_step)
trn_step += 1
except tf.errors.OutOfRangeError:
break
train_saver.save(session, model_path, global_step=n_trn_step)
def train(args, net, optimizer, criterion, scheduler = None):
log_file = open(args.save_root + file_name + "_training.log", "w", 1)
log_file.write(args.exp_name+'\n')
for arg in vars(args):
print(arg, getattr(args, arg))
log_file.write(str(arg)+': '+str(getattr(args, arg))+'\n')
log_file.write(str(net))
net.train()
# loss counters
batch_time = AverageMeter()
losses = AverageMeter()
loc_losses = AverageMeter()
cls_losses = AverageMeter()
print('Loading Dataset...')
train_dataset = UCF24Detection(args.data_root, args.train_sets, SSDAugmentation(args.ssd_dim, args.means),
AnnotationTransform(), input_type=args.input_type)
val_dataset = UCF24Detection(args.data_root, 'test', BaseTransform(args.ssd_dim, args.means),
AnnotationTransform(), input_type=args.input_type,
full_test=False)
epoch_size = len(train_dataset) // args.batch_size
print ("epoch_size: ", epoch_size)
print('Training SSD on', train_dataset.name)
if args.visdom:
import visdom
viz = visdom.Visdom()
viz.port = 8097
viz.env = args.exp_name
# initialize visdom loss plot
lot = viz.line(
X=torch.zeros((1,)).cpu(),
Y=torch.zeros((1, 6)).cpu(),
opts=dict(
xlabel='Iteration',
ylabel='Loss',
title='Current SSD Training Loss',
legend=['REG', 'CLS', 'AVG', 'S-REG', ' S-CLS', ' S-AVG']
)
)
# initialize visdom meanAP and class APs plot
legends = ['meanAP']
for cls in CLASSES:
legends.append(cls)
val_lot = viz.line(
X=torch.zeros((1,)).cpu(),
Y=torch.zeros((1,args.num_classes)).cpu(),
opts=dict(
xlabel='Iteration',
ylabel='Mean AP',
title='Current SSD Validation mean AP',
legend=legends
)
)
batch_iterator = None
train_data_loader = DataLoader(train_dataset, args.batch_size, num_workers=args.num_workers,
shuffle=False, collate_fn=detection_collate, pin_memory=True)
val_data_loader = DataLoader(val_dataset, args.batch_size, num_workers=args.num_workers,
shuffle=False, collate_fn=detection_collate, pin_memory=True)
my_dict = copy.deepcopy(train_data_loader.dataset.train_vid_frame)
keys = list(my_dict.keys())
k_len = len(keys)
arr = np.arange(k_len)
xxx = copy.deepcopy(train_data_loader.dataset.ids)
itr_count = 0
torch.cuda.synchronize()
t0 = time.perf_counter()
# for iteration in range(args.max_iter + 1):
current_epoch = 0
iteration = 0
while current_epoch < (args.total_epoch + 1):
if (not batch_iterator) or (iteration % epoch_size == 0):
xxxx = copy.deepcopy(train_data_loader.dataset.ids)
np.random.shuffle(arr)
iii = 0
for arr_i in arr:
key = keys[arr_i]
rang = my_dict[key]
xxxx[iii:(iii + rang[1] - rang[0])] = xxx[rang[0]:rang[1]]
iii += rang[1] - rang[0]
train_data_loader.dataset.ids = copy.deepcopy(xxxx)
# create batch iterator
batch_iterator = iter(train_data_loader)
if scheduler is not None and iteration > 0:
scheduler.step()
current_epoch += 1
iteration += 1
# load train data
images, targets, img_indexs = next(batch_iterator)
if args.cuda:
images = Variable(images.cuda())
targets = [Variable(anno.cuda(), volatile=True) for anno in targets]
else:
images = Variable(images)
targets = [Variable(anno, volatile=True) for anno in targets]
# forward
out = net(images, img_indexs)
# backprop
optimizer.zero_grad()
loss_l, loss_c = criterion(out, targets)
loss = loss_l + loss_c
loss.backward()
torch.nn.utils.clip_grad_norm(net.parameters(), args.clip)
optimizer.step()
loc_loss = loss_l.data[0]
conf_loss = loss_c.data[0]
# print('Loss data type ',type(loc_loss))
loc_losses.update(loc_loss)
cls_losses.update(conf_loss)
losses.update((loc_loss + conf_loss)/2.0)
if iteration % args.print_step == 0 and iteration > 0:
if args.visdom:
losses_list = [loc_losses.val, cls_losses.val, losses.val, loc_losses.avg, cls_losses.avg, losses.avg]
viz.line(X=torch.ones((1, 6)).cpu() * iteration,
Y=torch.from_numpy(np.asarray(losses_list)).unsqueeze(0).cpu(),
win=lot,
update='append')
torch.cuda.synchronize()
t1 = time.perf_counter()
batch_time.update(t1 - t0)
print_line = 'Epoch {:02d}/{:02d} Iteration {:06d}/{:06d} loc-loss {:.3f}({:.3f}) cls-loss {:.3f}({:.3f}) ' \
'average-loss {:.3f}({:.3f}) Timer {:0.3f}({:0.3f}) lr {:0.5f}'.format(
current_epoch, args.total_epoch, iteration, args.max_iter, loc_losses.val, loc_losses.avg, cls_losses.val,
cls_losses.avg, losses.val, losses.avg, batch_time.val, batch_time.avg, args.lr)
torch.cuda.synchronize()
t0 = time.perf_counter()
log_file.write(print_line+'\n')
print(print_line)
# if args.visdom and args.send_images_to_visdom:
# random_batch_index = np.random.randint(images.size(0))
# viz.image(images.data[random_batch_index].cpu().numpy())
itr_count += 1
if itr_count % args.loss_reset_step == 0 and itr_count > 0:
loc_losses.reset()
cls_losses.reset()
losses.reset()
batch_time.reset()
print('Reset accumulators of ', args.exp_name,' at', itr_count*args.print_step)
itr_count = 0
if (iteration % args.eval_step == 0 or iteration == 5000) and iteration > 0:
torch.cuda.synchronize()
tvs = time.perf_counter()
print('Saving state, iter:', iteration)
torch.save(net.state_dict(), args.save_root + file_name + '_ssd300_ucf24_' +
repr(iteration) + '.pth')
net.eval() # switch net to evaluation mode
mAP, ap_all, ap_strs = validate(args, net, val_data_loader, val_dataset, iteration, iou_thresh=args.iou_thresh)
for ap_str in ap_strs:
print(ap_str)
log_file.write(ap_str+'\n')
ptr_str = '\nMEANAP:::=>'+str(mAP)+'\n'
print(ptr_str)
log_file.write(ptr_str)
if args.visdom:
aps = [mAP]
for ap in ap_all:
aps.append(ap)
viz.line(
X=torch.ones((1, args.num_classes)).cpu() * iteration,
Y=torch.from_numpy(np.asarray(aps)).unsqueeze(0).cpu(),
win=val_lot,
update='append'
)
net.train() # Switch net back to training mode
torch.cuda.synchronize()
t0 = time.perf_counter()
prt_str = '\nValidation TIME::: {:0.3f}\n\n'.format(t0-tvs)
print(prt_str)
log_file.write(ptr_str)
log_file.close()
def main():
args = parse_args()
dataset = args.dataset
model_name = args.model
args.seed = random.randint(0, 1000000)
output_folder = None
if model_name == "adv":
output_folder = conf.output_folder_bgnn_adv + "/" + str(dataset)
elif model_name == "mlp":
output_folder = conf.output_folder_bgnn_mlp + "/" + str(dataset)
if not os.path.exists(output_folder):
os.makedirs(output_folder)
seed_file = output_folder + "/random_seed.txt"
fs = open(seed_file, 'w')
wstr = "%s" % str(args.seed)
fs.write(wstr + "\n")
fs.close()
np.random.seed(args.seed)
torch.manual_seed(args.seed)
device = torch.device(
"cuda:0" if torch.cuda.is_available() and args.gpu else "cpu")
torch.autograd.set_detect_anomaly(True)
# load the bipartite graph data
dataloader = DataLoader(args.dataset, link_pred=True, rec=False)
model_config = {
"attr_one_dim": dataloader.set_one.shape[1],
"attr_two_dim": dataloader.set_two.shape[1],
"set_one_size": dataloader.set_one.shape[0],
"set_two_size": dataloader.set_two.shape[0],
"epoch": args.epochs,
"layer_depth": args.layer_depth,
"device": device,
"disc_hid_dim": args.dis_hidden,
"learning_rate": args.lr,
"weight_decay": args.weight_decay,
"dropout": args.dropout,
"batch_size": args.batch_size
}
if args.model == 'adv':
# start training
bgnn = BGNNAdversarial(dataloader, model_config)
bgnn.adversarial_learning()
elif args.model == 'mlp':
bgnn = BGNNMLP()
bgnn.relation_learning()
with open('./output_model_tuning_link_prediction.txt', 'a') as f:
f.write(
"epoch: {}, layer_depth: {}, disc_hid_dim: {}, "
"learning_rate: {}, weight_decay: {}, dropout: {}, batch_size: {}\n"
.format(model_config['epoch'], model_config['layer_depth'],
model_config['disc_hid_dim'],
model_config['learning_rate'],
model_config['weight_decay'], model_config['dropout'],
model_config['batch_size']))
f.close()
def get_name(self, name):
"""Get metrics pandas name
Args:
name: a string recording prefix of metrics name
"""
namelist = []
for frame in self._frame_list:
namelist.append(name + '_' + str(frame))
return namelist
if __name__ == '__main__':
from config import args
origin = DataLoader()
input_base = args.origin_input_base
path = os.path.join(input_base, 'test.txt')
actor = args.actor
origin.variable_operate(input_base, path, actor)
origin_x_y = origin.variable_get_data_with_window(2, each_len=2, stride=12)
pred = DataLoader()
input_base = args.result_input_base
path = os.path.join(input_base, 'data_out.txt')
pred.operate(input_base, path, actor)
pred_x_y = pred.get_data_with_window(4, 2)
pred_info = pred.get_data()
frame_list = [4, 10]
eva = Evaluation(frame_list)
opt.model_file += f'-lrtz={opt.lrt_z}'
opt.model_file += f'-updatez={opt.z_updates}'
opt.model_file += f'-inferz={opt.infer_z}'
opt.model_file += f'-learnedcost={opt.learned_cost}'
opt.model_file += f'-seed={opt.seed}'
if opt.value_model == '':
opt.model_file += '-novalue'
if opt.learned_cost == 1:
model.cost = torch.load(path.join(opt.model_dir,
opt.mfile + '.cost.model'))['model']
print(f'[will save as: {opt.model_file}]')
dataloader = DataLoader(None, opt, opt.dataset)
model.train()
model.opt.u_hinge = opt.u_hinge
planning.estimate_uncertainty_stats(model,
dataloader,
n_batches=50,
npred=opt.npred)
model.eval()
def train(nbatches, npred):
model.train()
model.policy_net.train()
total_loss_c, total_loss_u, total_loss_l, total_loss_a, n_updates, grad_norm = 0, 0, 0, 0, 0, 0
total_loss_policy = 0
for j in range(nbatches):
np.random.seed(opt.seed)
torch.manual_seed(opt.seed)
opt.save_video = (opt.save_video == 1)
opt.eval_dir = opt.model_dir + f'eval/'
print(f'[loading {opt.model_dir + opt.mfile}]')
model = torch.load(opt.model_dir + opt.mfile)
if type(model) is dict: model = model['model']
model = model.cuda()
model.eval()
# if opt.cuda == 1:
# model.intype('gpu')
dataloader = DataLoader(None, opt, opt.dataset)
# model.opt.npred = opt.npred # instruct the model about how many predictions we want it to produce
model.opt.alpha = 0
dirname = f'{opt.eval_dir}{opt.mfile}-nbatches={opt.n_batches}-npred={opt.npred}-nsample={opt.n_samples}'
if '-ten' in opt.mfile:
dirname += f'-sampling={opt.sampling}'
if opt.sampling == 'knn':
dirname += f'-density={opt.graph_density}'
elif opt.sampling == 'pdf':
dirname += f'-nmixture={opt.n_mixture}'
mfile_prior = f'{opt.model_dir}/{opt.mfile}-nfeature=128-lrt=0.0001-nmixture={opt.n_mixture}.prior'
print(f'[loading prior model: {mfile_prior}]')
model.prior = torch.load(mfile_prior).cuda()
# load z vectors. Extract them if they are not already saved.
pzfile = opt.model_dir + opt.mfile + '.pz'
captions_dict = load_captions(train_dir)
vocab = Vocabulary(captions_dict, threshold)
os.mkdir(model_dir)
print("Directory '{model_name}' created to dump vocab.pkl.".format(
model_name=model_name))
with open(os.path.join(model_dir, 'vocab.pkl'), 'wb') as f:
pickle.dump(vocab, f)
print('Dictionary Dumped !')
transform = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
dataloader = DataLoader(train_dir, vocab, transform)
data = dataloader.gen_data()
print(train_dir + ' loaded !')
vocab_size = vocab.index
cnn = get_CNN(architecture=model_name, embedding_dim=args.embedding_dim)
lstm = RNN(embedding_dim=args.embedding_dim,
hidden_dim=args.hidden_dim,
vocab_size=vocab_size)
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
cnn.to(device)
lstm.to(device)
import pandas as pd
from dataloader import DataLoader
# full_df = DataLoader().load_pickle('full')
base_path = './Data/data_1-2019/'
# df_concat.append(DataLoader().load_adv(base_path + 'carat_1_2019.csv'))
# df_concat.append(DataLoader().load_adv(base_path + 'amnet_1_2019.csv'))
# df_concat.append(DataLoader().load_adv(base_path + 'deepblue_1_2019.csv'))
# df_concat.append(DataLoader().load_adv(base_path + 'dentsux_1_2019.csv'))
# df_concat.append(DataLoader().load_adv(base_path + 'iprospect_1_2019.csv'))
# df_concat.append(DataLoader().load_adv(base_path + 'vizeum_1_2019.csv'))
carat_df = DataLoader().load_adv(base_path + 'carat_1_2019.csv')
amnet_df = DataLoader().load_adv(base_path + 'amnet_1_2019.csv')
deepblue_df = DataLoader().load_adv(base_path + 'deepblue_1_2019.csv')
dentsux_df = DataLoader().load_adv(base_path + 'dentsux_1_2019.csv')
iprospect_df = DataLoader().load_adv(base_path + 'iprospect_1_2019.csv')
vizeum_df = DataLoader().load_adv(base_path + 'vizeum_1_2019.csv')
#Add columns CARAT
carat_df.insert(0, 'Macro DB', 'Advertmind')
carat_df.insert(1, 'DB', 'CARAT')
carat_df.insert(2, 'Macro Agencia', 'CARAT')
carat_df.insert(3, 'Agencia', 'CARAT')
#Add columns DEEPBLUE
deepblue_df.insert(0, 'Macro DB', 'Advertmind')
deepblue_df.insert(1, 'DB', 'DEEPBLUE')
deepblue_df.insert(2, 'Macro Agencia', 'CARAT')
### TORCH SETUP ###
print('-> Using PyTorch', th.__version__)
th.manual_seed(args.seed)
if args.gpu: th.cuda.manual_seed(args.seed)
X_TENSOR = ''
if not args.gpu:
X_TENSOR = 'torch.DoubleTensor' if args.double else 'torch.FloatTensor'
else:
X_TENSOR = 'torch.cuda.DoubleTensor' if args.double else 'torch.cuda.FloatTensor'
#th.set_default_tensor_type('torch.cuda.HalfTensor') # Bad Stability
th.set_default_tensor_type(X_TENSOR)
th.backends.cudnn.benchmark=True # enable cuDNN auto-tuner
### DataLoader ###
dataloader = DataLoader()
#dataloader.satellite(64, 'trainval', 100)
### Model ###
class Model(th.nn.Module):
''' Reference: caffe/examples/cifar10 # 70%
https://github.com/tensorflow/models/blob/master/tutorials/image/cifar10/cifar10.py
'''
def __init__(self):
super(Model, self).__init__()
self.SEQ1 = th.nn.Sequential(OrderedDict([
('conv1', th.nn.Conv2d(3, 32, 5, stride=1, padding=2)),
('bn1', th.nn.BatchNorm2d(32)),
('relu1', th.nn.ReLU()),
('pool1', th.nn.MaxPool2d(3, stride=2, padding=1)),
CLASS = 10 W_LEN = 200 CUDA = 1 CRADLE_SIZE = 50 REPRO_SIZE = 20 UNION = 10 weights = np.load(save_npy_name) weights = torch.from_numpy(weights).cuda()[:W_LEN] weights2 = np.load(save_npy_name2) weights2 = torch.from_numpy(weights2).cuda()[:W_LEN] weights3 = np.load(save_npy_name3) weights3 = torch.from_numpy(weights3).cuda()[:W_LEN] dl = DataLoader(True, CUDA) images, labels = dl.get_all() dl_test = DataLoader(False, CUDA) images_t, labels_t = dl_test.get_all() images = get_images_output(weights, images) images_t = get_images_output(weights, images_t) images = get_images_output(weights2, images) images_t = get_images_output(weights2, images_t) images = get_images_output(weights3, images) images_t = get_images_output(weights3, images_t) images[images == 0] = -1 images_t[images_t == 0] = -1 images = images.repeat(REPRO_SIZE, 1, 1)
def __init__(self):
DataLoader.__init__(self)
self.read_train()
self.read_test()
def main():
# Step 0: preparation
writer = LogWriter(logdir="./log/scalar")
place = paddle.fluid.CUDAPlace(0)
with fluid.dygraph.guard(place):
# Step 1: Define training dataloader
image_folder = ""
image_list_file = "dummy_data/fabric_list.txt"
transform = Transform() #Normalize2() # [0,255]-->[0,1]
x_data = DataLoader(image_folder, image_list_file, transform=transform)
x_dataloader = fluid.io.DataLoader.from_generator(capacity=2,
return_list=True)
x_dataloader.set_sample_generator(x_data, args.batch_size)
total_batch = len(x_data) // args.batch_size
# Step 2: Create model
if args.net == "basic":
D = Discriminator()
G = Generator()
E = Invertor()
else:
raise NotImplementedError(
f"args.net: {args.net} is not Supported!")
# Step 3: Define criterion and optimizer
criterion = Basic_Loss
D_optim = AdamOptimizer(learning_rate=args.lr,
parameter_list=D.parameters())
G_optim = AdamOptimizer(learning_rate=args.lr,
parameter_list=G.parameters())
E_optim = AdamOptimizer(learning_rate=args.lr,
parameter_list=E.parameters())
G_loss_meter = AverageMeter()
D_loss_meter = AverageMeter()
E_loss_meter = AverageMeter()
D.train()
G.train()
E.train()
# Step 4: Slight Training
iteration = -1
is_slight_Train = True
for epoch in range(1, args.epoch_num + 1):
#optim Discriminator
for (x, x_labels) in x_dataloader():
n = x.shape[0]
if is_slight_Train:
iteration += 1
x = fluid.layers.cast(x, dtype="float32")
x = fluid.layers.transpose(x, perm=[0, 3, 1, 2])
preds_x = D(x)
preds_x_array = preds_x.numpy()
#print("D(x),1",preds_array.shape, np.mean(preds_array))
writer.add_scalar(tag="D(x)=1",
step=iteration,
value=np.mean(preds_x_array))
if np.mean(preds_x_array) >= 0.98:
is_slight_Train = False
z = np.random.rand(n, 64)
zeros = np.zeros((n, 1))
z = to_variable(z)
zeros = to_variable(zeros)
z = fluid.layers.cast(z, dtype="float32")
zeros = fluid.layers.cast(zeros, dtype="int64")
preds_fx = D(G(z))
preds_fx_array = preds_fx.numpy()
writer.add_scalar(tag="D(G(z))=0",
step=iteration,
value=np.mean(preds_fx_array))
D_loss = criterion(preds_x, x_labels) + criterion(
preds_fx, zeros)
D_loss.backward()
D_optim.minimize(D_loss)
D.clear_gradients()
D_loss_meter.update(D_loss.numpy()[0], n)
writer.add_scalar(tag="D_loss",
step=iteration,
value=D_loss_meter.avg)
print(f"EPOCH[{epoch:03d}/{args.epoch_num:03d}], " +
f"STEP{iteration}, " +
f"Average D Loss: {D_loss_meter.avg:4f}, ")
z = np.random.rand(n, 64)
ones = np.ones((n, 1))
z = to_variable(z)
ones = to_variable(ones)
z = fluid.layers.cast(z, dtype="float32")
ones = fluid.layers.cast(ones, dtype="int64")
preds = D(G(z))
preds_array = preds.numpy()
writer.add_scalar(tag="D(G(z))=1",
step=iteration,
value=np.mean(preds_array))
G_loss = criterion(preds, ones)
G_loss.backward()
G_optim.minimize(G_loss)
G.clear_gradients()
G_loss_meter.update(G_loss.numpy()[0], n)
writer.add_scalar(tag="G_loss",
step=iteration,
value=G_loss_meter.avg)
print(f"EPOCH[{epoch:03d}/{args.epoch_num:03d}], " +
f"STEP{iteration}, " +
f"Average G Loss: {G_loss_meter.avg:4f}")
if epoch % args.save_freq == 0 or epoch == args.epoch_num or not is_slight_Train:
D_model_path = os.path.join(args.checkpoint_folder,
f"D_{args.net}-Epoch-{epoch}")
G_model_path = os.path.join(args.checkpoint_folder,
f"G_{args.net}-Epoch-{epoch}")
# save model and optmizer states
model_dict = D.state_dict()
fluid.save_dygraph(model_dict, D_model_path)
optim_dict = D_optim.state_dict()
fluid.save_dygraph(optim_dict, D_model_path)
model_dict = G.state_dict()
fluid.save_dygraph(model_dict, G_model_path)
optim_dict = G_optim.state_dict()
fluid.save_dygraph(optim_dict, G_model_path)
print(
f'----- Save model: {D_model_path}.pdparams, {G_model_path}.pdparams'
)
if not is_slight_Train:
break
# Step 5: full training for Generator and Discriminator
D_optim = AdamOptimizer(learning_rate=args.lr * 10,
parameter_list=D.parameters())
G_optim = AdamOptimizer(learning_rate=args.lr * 10,
parameter_list=G.parameters())
G_loss_meter = AverageMeter()
D_loss_meter = AverageMeter()
for epoch in range(1, args.epoch_num + 1):
for (x, x_labels) in x_dataloader():
n = x.shape[0]
iteration += 1
x = fluid.layers.cast(x, dtype="float32")
x = fluid.layers.transpose(x, perm=[0, 3, 1, 2])
preds1 = D(x)
preds_array = preds1.numpy()
writer.add_scalar(tag="D(x)=1",
step=iteration,
value=np.mean(preds_array))
z = np.random.rand(n, 64)
zeros = np.zeros((n, 1))
z = to_variable(z)
zeros = to_variable(zeros)
z = fluid.layers.cast(z, dtype="float32")
zeros = fluid.layers.cast(zeros, dtype="int64")
preds2 = D(G(z))
preds_array = preds2.numpy()
#print("DG(z),0:",preds_array.shape, np.mean(preds_array))
writer.add_scalar(tag="D(G(z))=0",
step=iteration,
value=np.mean(preds_array))
D_loss = criterion(preds1, x_labels) + criterion(preds2, zeros)
D_loss.backward()
D_optim.minimize(D_loss)
D.clear_gradients()
D_loss_meter.update(D_loss.numpy()[0], n)
writer.add_scalar(tag="D_loss",
step=iteration,
value=D_loss_meter.avg)
print(f"EPOCH[{epoch:03d}/{args.epoch_num:03d}], " +
f"STEP{iteration}, " +
f"Average D Loss: {D_loss_meter.avg:4f} ")
z = np.random.rand(n, 64)
ones = np.ones((n, 1))
z = to_variable(z)
ones = to_variable(ones)
z = fluid.layers.cast(z, dtype="float32")
ones = fluid.layers.cast(ones, dtype="int64")
preds = D(G(z))
preds_array = preds.numpy()
#print("DG(z),1:",preds_array.shape, np.mean(preds_array))
writer.add_scalar(tag="D(G(z))=1",
step=iteration,
value=np.mean(preds_array))
G_loss = criterion(preds, ones)
G_loss.backward()
G_optim.minimize(G_loss)
G.clear_gradients()
G_loss_meter.update(G_loss.numpy()[0], n)
writer.add_scalar(tag="G_loss",
step=iteration,
value=G_loss_meter.avg)
print(f"EPOCH[{epoch:03d}/{args.epoch_num:03d}], " +
f"STEP{iteration}, " +
f"Average G Loss: {G_loss_meter.avg:4f}")
if epoch % args.save_freq == 0 or epoch == args.epoch_num:
D_model_path = os.path.join(args.checkpoint_folder,
f"D_{args.net}-Epoch-{epoch}")
G_model_path = os.path.join(args.checkpoint_folder,
f"G_{args.net}-Epoch-{epoch}")
# save model and optmizer states
model_dict = D.state_dict()
fluid.save_dygraph(model_dict, D_model_path)
optim_dict = D_optim.state_dict()
fluid.save_dygraph(optim_dict, D_model_path)
model_dict = G.state_dict()
fluid.save_dygraph(model_dict, G_model_path)
optim_dict = G_optim.state_dict()
fluid.save_dygraph(optim_dict, G_model_path)
print(
f'----- Save model: {D_model_path}.pdparams, {G_model_path}.pdparams'
)
# Step 6: full training for Inverter
E_optim = AdamOptimizer(learning_rate=args.lr * 10,
parameter_list=E.parameters())
E_loss_meter = AverageMeter()
for epoch in range(1, args.epoch_num + 1):
for (x, x_labels) in x_dataloader():
n = x.shape[0]
iteration += 1
x = fluid.layers.cast(x, dtype="float32")
image = x.numpy()[0] * 255
writer.add_image(tag="x", step=iteration, img=image)
x = fluid.layers.transpose(x, perm=[0, 3, 1, 2])
invert_x = G(E(x))
invert_image = fluid.layers.transpose(invert_x,
perm=[0, 2, 3, 1])
invert_image = invert_image.numpy()[0] * 255
#print("D(x),1",preds_array.shape, np.mean(preds_array))
writer.add_image(tag="invert_x",
step=iteration,
img=invert_image)
print(np.max(invert_image), np.min(invert_image))
E_loss = fluid.layers.mse_loss(invert_x, x)
print("E_loss shape:", E_loss.numpy().shape)
E_loss.backward()
E_optim.minimize(E_loss)
E.clear_gradients()
E_loss_meter.update(E_loss.numpy()[0], n)
writer.add_scalar(tag="E_loss",
step=iteration,
value=E_loss_meter.avg)
print(f"EPOCH[{epoch:03d}/{args.epoch_num:03d}], " +
f"STEP{iteration}, " +
f"Average E Loss: {E_loss_meter.avg:4f}, ")
if epoch % args.save_freq == 0 or epoch == args.epoch_num:
E_model_path = os.path.join(args.checkpoint_folder,
f"E_{args.net}-Epoch-{epoch}")
# save model and optmizer states
model_dict = E.state_dict()
fluid.save_dygraph(model_dict, E_model_path)
optim_dict = E_optim.state_dict()
fluid.save_dygraph(optim_dict, E_model_path)
print(
f'----- Save model: {E_model_path}.pdparams, {E_model_path}.pdparams'
)