def analyse_moving():
global image_file
sentence = e.get()
path = './results/moving.png'
save_result(qq_msgs='./inputs/msgs.txt', path=path)
image_file = tk.PhotoImage(file=path)
canvas.itemconfig(image, image=image_file)
def color_adjust(x, args):
if args.iter % args.save_intervel == 0:
save_result(x, args)
args.iter += 1
# Input for VGG
x_vgg = np.asarray(np.reshape(x, args.shape), dtype=np.float32)
x_vgg_var = Variable(chainer.dataset.concat_examples([x_vgg], args.gpu))
# Poisson loss
poisson_loss = F.sum(
F.square(args.content_laplace -
F.convolution_2d(x_vgg_var, W=args.W_laplace, pad=1)) *
args.inverse_border_mask_var)
# tv loss
tv_loss = total_variation(x_vgg_var)
# Content loss
content_loss = F.sum(
F.square((args.bg_var - x_vgg_var) * args.inverse_mask_var))
loss = args.poisson_weight * poisson_loss + args.tv_weight * tv_loss + args.content_weight * content_loss
# Backward
loss.backward()
# Transfer loss & diff from GPU to CPU
loss = cuda.to_cpu(loss.data)
dx = np.squeeze(cuda.to_cpu(x_vgg_var.grad))
return loss, np.asarray(dx.flatten(), dtype=np.float64)
def simple_regression(x_train, x_test, y_train, y_test):
# 线性回归
# model = LinearRegression()
# 岭回归
# model = Ridge(alpha=0.1)
# lasso 回归
model = Lasso(alpha=0.1)
# kernal ridge
# model = KernelRidge(kernel='rbf', alpha=0.1)
# model = KernelRidge(kernel='linear', alpha=0.1)
# model = KernelRidge(kernel='sigmoid', alpha=0.1)
# model = KernelRidge(kernel='poly', alpha=0.1)
# model = KernelRidge(kernel='laplacian', alpha=0.1)
# model = KernelRidge(kernel='cosine', alpha=0.1)
# 不知道为什么用不了
# model = KernelRidge(kernel='chi2', alpha=0.1)
# model = KernelRidge(kernel='additive_chi2', alpha=0.1)
model.fit(x_train, y_train)
score = model.score(x_test, y_test)
print("score", score)
y_pred = model.predict(x_test)
print("mean_squared_error", mean_squared_error(y_test, y_pred))
plot_learning_curve(model, "rate", x_train, y_train)
result = model.predict(test_df)
save_result(result)
print(result)
def analyse_shuoshuo():
global image_file
sentence = e.get()
path = "results/my_shuoshuo_sentiment.png"
save_result(qq_msgs="inputs/msgs.txt", path=path)
image_file = tk.PhotoImage(file=path)
canvas.itemconfig(image, image=image_file)
def val_epoch(model,
ratings,
negs,
K,
use_cuda=True,
output=None,
epoch=None,
processes=1):
if epoch is None:
print("Initial evaluation")
else:
print("Epoch {} evaluation".format(epoch))
mlperf_log.ncf_print(key=mlperf_log.EVAL_START, value=epoch)
start = datetime.now()
model.eval()
if processes > 1:
context = mp.get_context('spawn')
_eval_one = partial(eval_one, model=model, K=K, use_cuda=use_cuda)
with context.Pool(processes=processes) as workers:
hits_ndcg_numpred = workers.starmap(_eval_one, zip(ratings, negs))
hits, ndcgs, num_preds = zip(*hits_ndcg_numpred)
else:
hits, ndcgs, num_preds = [], [], []
for rating, items in zip(ratings, negs):
hit, ndcg, num_pred = eval_one(rating,
items,
model,
K,
use_cuda=use_cuda)
hits.append(hit)
ndcgs.append(ndcg)
num_preds.append(num_pred)
hits = np.array(hits, dtype=np.float32)
ndcgs = np.array(ndcgs, dtype=np.float32)
assert len(set(num_preds)) == 1
num_neg = num_preds[0] - 1 # one true positive, many negatives
mlperf_log.ncf_print(key=mlperf_log.EVAL_SIZE,
value={
"epoch": epoch,
"value": len(hits) * (1 + num_neg)
})
mlperf_log.ncf_print(key=mlperf_log.EVAL_HP_NUM_USERS, value=len(hits))
mlperf_log.ncf_print(key=mlperf_log.EVAL_HP_NUM_NEG, value=num_neg)
end = datetime.now()
if output is not None:
result = OrderedDict()
result['timestamp'] = datetime.now()
result['duration'] = end - start
result['epoch'] = epoch
result['K'] = K
result['hit_rate'] = np.mean(hits)
result['NDCG'] = np.mean(ndcgs)
utils.save_result(result, output)
return hits, ndcgs
def predict_result(train_data, test_data): X_train = prepare_X_train(train_data) X_test = prepare_X_test(test_data) y_train = train_data['Survived'] clf = RandomForestClassifier(n_estimators=200, random_state=241) clf.fit(X_train, y_train) predicted = clf.predict(X_test) save_result(predicted)
def val_epoch(model, x, y, dup_mask, real_indices, K, samples_per_user, num_user, output=None,
epoch=None, loss=None, use_cuda=False):
start = datetime.now()
log_2 = math.log(2)
model.eval()
hits = torch.tensor(0.)
ndcg = torch.tensor(0.)
if use_cuda:
hits = torch.tensor(0., device='cuda')
ndcg = torch.tensor(0., device='cuda')
else:
hits = torch.tensor(0.)
ndcg = torch.tensor(0.)
with torch.no_grad():
for i, (u,n) in enumerate(zip(x,y)):
if use_cuda: res = model(u.cuda().view(-1), n.cuda().view(-1), sigmoid=True).detach().view(-1,samples_per_user)
else: res = model(u.cpu().view(-1), n.cpu().view(-1), sigmoid=True).detach().view(-1,samples_per_user)
# set duplicate results for the same item to -1 before topk
res[dup_mask[i]] = -1
out = torch.topk(res,K)[1]
# topk in pytorch is stable(if not sort)
# key(item):value(predicetion) pairs are ordered as original key(item) order
# so we need the first position of real item(stored in real_indices) to check if it is in topk
if use_cuda: ifzero = (out == real_indices[i].cuda().view(-1,1))
else: ifzero = (out == real_indices[i].cpu().view(-1,1))
hits += ifzero.sum()
ndcg += (log_2 / (torch.nonzero(ifzero)[:,1].view(-1).to(torch.float)+2).log_()).sum()
mlperf_log.ncf_print(key=mlperf_log.EVAL_SIZE, value={"epoch": epoch, "value": num_user * samples_per_user})
mlperf_log.ncf_print(key=mlperf_log.EVAL_HP_NUM_USERS, value=num_user)
mlperf_log.ncf_print(key=mlperf_log.EVAL_HP_NUM_NEG, value=samples_per_user - 1)
end = datetime.now()
hits = hits.item()
ndcg = ndcg.item()
if output is not None:
result = OrderedDict()
result['timestamp'] = datetime.now()
result['duration'] = end - start
result['epoch'] = epoch
result['K'] = K
result['hit_rate'] = hits/num_user
result['NDCG'] = ndcg/num_user
result['loss'] = loss
utils.save_result(result, output)
return hits/num_user, ndcg/num_user
def val_epoch(model,
ratings,
negs,
K,
use_cuda=True,
output=None,
epoch=None,
processes=1,
num_users=-1):
if epoch is None:
msglogger.info("Initial evaluation")
else:
msglogger.info("Epoch {} evaluation".format(epoch))
start = datetime.now()
model.eval()
if num_users > 0:
ratings = ratings[:num_users]
negs = negs[:num_users]
if processes > 1:
context = mp.get_context('spawn')
_eval_one = partial(eval_one, model=model, K=K, use_cuda=use_cuda)
with context.Pool(processes=processes) as workers:
hits_and_ndcg = workers.starmap(_eval_one, zip(ratings, negs))
hits, ndcgs = zip(*hits_and_ndcg)
else:
hits, ndcgs = [], []
with tqdm.tqdm(zip(ratings, negs), total=len(ratings)) as t:
for rating, items in t:
hit, ndcg = eval_one(rating,
items,
model,
K,
use_cuda=use_cuda)
hits.append(hit)
ndcgs.append(ndcg)
steps_completed = len(hits) + 1
if steps_completed % 100 == 0:
t.set_description('HR@10 = {0:.4f}, NDCG = {1:.4f}'.format(
np.mean(hits), np.mean(ndcgs)))
hits = np.array(hits, dtype=np.float32)
ndcgs = np.array(ndcgs, dtype=np.float32)
end = datetime.now()
if output is not None:
result = OrderedDict()
result['timestamp'] = datetime.now()
result['duration'] = end - start
result['epoch'] = epoch
result['K'] = K
result['hit_rate'] = np.mean(hits)
result['NDCG'] = np.mean(ndcgs)
utils.save_result(result, output)
return hits, ndcgs
def predict_result(train_data, test_data): scaler = StandardScaler() X_train = prepare_X_train(train_data, scaler) X_test = prepare_X_test(test_data, scaler) y_train = train_data['Survived'] clf = LogisticRegression(C=1000.0, random_state=241) clf.fit(X_train, y_train) predicted = clf.predict(X_test) save_result(predicted, X_test.index)
def run_evaluation(args, model, data_loaders, model_description, n_choices,
layers_types, downsample_layers):
start = time.time()
num_samples = utils.get_number_of_samples(args.dataset)
all_values = {}
device = 'cuda'
#setting up random seeds
utils.setup_torch(args.seed)
#creating model skeleton based on description
propagate_weights = []
for layer in model_description:
cur_weights = [0 for i in range(n_choices)]
cur_weights[layers_types.index(layer)] = 1
propagate_weights.append(cur_weights)
model.propagate = propagate_weights
#Create the computationally identical model but without multiple choice blocks (just a single path net)
#This is needed to correctly measure MACs
pruned_model = models.SinglePathSupernet(
num_classes=utils.get_number_of_classes(args.dataset),
propagate=propagate_weights,
put_downsampling=downsample_layers) #.to(device)
pruned_model.propagate = propagate_weights
inputs = torch.randn((1, 3, 32, 32))
total_ops, total_params = profile(pruned_model, (inputs, ), verbose=True)
all_values['MMACs'] = np.round(total_ops / (1000.0**2), 2)
all_values['Params'] = int(total_params)
del pruned_model
del inputs
################################################
criterion = torch.nn.CrossEntropyLoss()
#Initialize batch normalization parameters
utils.bn_update(device, data_loaders['train_for_bn_recalc'], model)
val_res = utils.evaluate(device, data_loaders['val'], model, criterion,
num_samples['val'])
test_res = utils.evaluate(device, data_loaders['test'], model, criterion,
num_samples['test'])
all_values['val_loss'] = np.round(val_res['loss'], 3)
all_values['val_acc'] = np.round(val_res['accuracy'], 3)
all_values['test_loss'] = np.round(test_res['loss'], 3)
all_values['test_acc'] = np.round(test_res['accuracy'], 3)
print(all_values, 'time taken: %.2f sec.' % (time.time() - start))
utils.save_result(all_values, args.dir, model_description)
def run(ckpt_fpath):
checkpoint = torch.load(ckpt_fpath)
""" Load Config """
config = dict_to_cls(checkpoint['config'])
""" Build Data Loader """
if config.corpus == "MSVD":
corpus = MSVD(config)
elif config.corpus == "MSR-VTT":
corpus = MSRVTT(config)
train_iter, val_iter, test_iter, vocab = \
corpus.train_data_loader, corpus.val_data_loader, corpus.test_data_loader, corpus.vocab
print(
'#vocabs: {} ({}), #words: {} ({}). Trim words which appear less than {} times.'
.format(vocab.n_vocabs, vocab.n_vocabs_untrimmed, vocab.n_words,
vocab.n_words_untrimmed, config.loader.min_count))
""" Build Models """
decoder = Decoder(rnn_type=config.decoder.rnn_type,
num_layers=config.decoder.rnn_num_layers,
num_directions=config.decoder.rnn_num_directions,
feat_size=config.feat.size,
feat_len=config.loader.frame_sample_len,
embedding_size=config.vocab.embedding_size,
hidden_size=config.decoder.rnn_hidden_size,
attn_size=config.decoder.rnn_attn_size,
output_size=vocab.n_vocabs,
rnn_dropout=config.decoder.rnn_dropout)
decoder.load_state_dict(checkpoint['decoder'])
model = CaptionGenerator(decoder, config.loader.max_caption_len, vocab)
model = model.cuda()
""" Train Set """
"""
train_vid2pred = get_predicted_captions(train_iter, model, model.vocab, beam_width=5, beam_alpha=0.)
train_vid2GTs = get_groundtruth_captions(train_iter, model.vocab)
train_scores = score(train_vid2pred, train_vid2GTs)
print("[TRAIN] {}".format(train_scores))
"""
""" Validation Set """
"""
val_vid2pred = get_predicted_captions(val_iter, model, model.vocab, beam_width=5, beam_alpha=0.)
val_vid2GTs = get_groundtruth_captions(val_iter, model.vocab)
val_scores = score(val_vid2pred, val_vid2GTs)
print("[VAL] scores: {}".format(val_scores))
"""
""" Test Set """
test_vid2pred = get_predicted_captions(test_iter,
model,
model.vocab,
beam_width=5,
beam_alpha=0.)
test_vid2GTs = get_groundtruth_captions(test_iter, model.vocab)
test_scores = score(test_vid2pred, test_vid2GTs)
print("[TEST] {}".format(test_scores))
test_save_fpath = os.path.join(C.result_dpath,
"{}_{}.csv".format(config.corpus, 'test'))
save_result(test_vid2pred, test_vid2GTs, test_save_fpath)
def test_model(model, dataset, path_to_results):
running_loss = 0.0
for image_number in tqdm(range(len(dataset))):
path_to_image = dataset.iloc[image_number]['image']
image = Image.open(path_to_image).convert('L') # PIL Image grayscale
np_image = np.array(image)[..., np.newaxis] # numpy array from PIL Image
path_to_save = os.path.join(path_to_results, os.path.basename(path_to_image))
save_result(model, np_image, FRAME_SIZE, OVERLAY_SIZE, path_to_save, figsize=(16, 9))
def val_epoch(model, x, y, dup_mask, real_indices, K, samples_per_user, num_user, output=None,
epoch=None, distributed=False):
start = datetime.now()
log_2 = math.log(2)
model.eval()
with torch.no_grad():
p = []
for u,n in zip(x,y):
p.append(model(u, n, sigmoid=True).detach())
del x
del y
temp = torch.cat(p).view(-1,samples_per_user)
del p
# set duplicate results for the same item to -1 before topk
temp[dup_mask] = -1
out = torch.topk(temp,K)[1]
# topk in pytorch is stable(if not sort)
# key(item):value(predicetion) pairs are ordered as original key(item) order
# so we need the first position of real item(stored in real_indices) to check if it is in topk
ifzero = (out == real_indices.view(-1,1))
hits = ifzero.sum()
ndcg = (log_2 / (torch.nonzero(ifzero)[:,1].view(-1).to(torch.float)+2).log_()).sum()
LOGGER.log(key=tags.EVAL_SIZE, value={"epoch": epoch, "value": num_user * samples_per_user})
LOGGER.log(key=tags.EVAL_HP_NUM_USERS, value=num_user)
LOGGER.log(key=tags.EVAL_HP_NUM_NEG, value=samples_per_user - 1)
end = datetime.now()
if distributed:
torch.distributed.all_reduce(hits, op=torch.distributed.reduce_op.SUM)
torch.distributed.all_reduce(ndcg, op=torch.distributed.reduce_op.SUM)
hits = hits.item()
ndcg = ndcg.item()
if output is not None:
result = OrderedDict()
result['timestamp'] = datetime.now()
result['duration'] = end - start
result['epoch'] = epoch
result['K'] = K
result['hit_rate'] = hits/num_user
result['NDCG'] = ndcg/num_user
utils.save_result(result, output)
model.train()
return hits/num_user, ndcg/num_user
def train(target_distribution: Distribution) -> None:
"""Training Normalizing Flow"""
target_distribution.save_distribution()
normalizing_flow = NormalizingFlow(K=NORMALIZING_FLOW_LAYER_NUM)
z_0, log_q_0 = normalizing_flow.get_placeholder()
z_k, log_q_k = normalizing_flow.forward(z_0, log_q_0)
loss = normalizing_flow.calc_loss(z_k, log_q_k, target_distribution)
trainer = normalizing_flow.get_trainer(loss)
logger.info('Calculation graph constructed')
loss_values = []
with tf.Session() as sess:
logger.info('Session Start')
sess.run(tf.global_variables_initializer())
logger.info('All variables initialized')
logger.info(f'Training Start (number of iterations: {ITERATION})')
for iteration in range(ITERATION + 1):
z_0_batch = NormalDistribution.sample(BATCH_SIZE)
log_q_0_batch = np.log(NormalDistribution.calc_prob(z_0_batch))
_, loss_value = sess.run([trainer, loss], {
z_0: z_0_batch,
log_q_0: log_q_0_batch
})
loss_values.append(loss_value)
if iteration % 100 == 0:
iteration_digits = len(str(ITERATION))
logger.info(
f'Iteration: {iteration:<{iteration_digits}} Loss: {loss_value}'
)
if iteration % SAVE_FIGURE_INTERVAL == 0:
z_0_batch_for_visualize = NormalDistribution.sample(
NUMBER_OF_SAMPLES_FOR_VISUALIZE)
log_q_0_batch_for_visualize = np.log(
NormalDistribution.calc_prob(z_0_batch_for_visualize))
z_k_value = sess.run(
z_k, {
z_0: z_0_batch_for_visualize,
log_q_0: log_q_0_batch_for_visualize
})
save_result(z_k_value, iteration, target_distribution.__name__)
save_loss_values(loss_values, target_distribution.__name__)
logger.info('Training Finished')
logger.info('Session Closed')
def train_model(cfg, model, Xvalid, Yvalid, writer, save_dir):
mll_max = -np.inf
best_iter = 0
accuracy_list = []
mll_list_que = deque([mll_max, mll_max, mll_max])
best_model_que = deque([model, model, model])
for i in tdqm(range(cfg.iterations), ascii=" .oO0", bar_format="{total_time}: {percentage:.0f}%|{bar}{r_bar}"):
if i == 0:
for j in range(cfg.sghmc_step):
model.sghmc_step()
summary = model.train_hypers(tb=True)
print("Iteration", i, end='\r')
mll, sum_mll = model.print_sample_performance(tb=True)
print(f"MLL: {mll}")
# mll = struct.unpack('fff', sum_mll)[2]
# set_trace()
writer.add_summary(summary, global_step=i)
writer.add_summary(sum_mll, global_step=i)
else:
for j in range(cfg.sghmc_step):
model.sghmc_step()
model.train_hypers()
print("Iteration", i, end='\r')
mll, _ = model.print_sample_performance(tb=True)
print(f"MLL: {mll}")
if np.round(mll - mll_max, decimals=5) > 0:
# accuracy = measure_accuracy(model)
mll_max = mll
print('MLL increased ({:.7f} --> {:.7f}). Updating values ....'.format(mll_list_que[-1], mll_max))
mll_list_que.append(mll)
best_model_que.append(model) # append best model so far
best_model_que.popleft() # remove worst model so far
mll_list_que.popleft()
best_iter = i
# save best model
print(f'################## save best model at {save_dir} ##################')
model_name = f'{best_iter}_bestmodel_{mll_list_que[-1]}'
best_model_que[-1].save(save_dir, name=model_name)
accuracy = measure_accuracy(best_model_que[-1], Xvalid, Yvalid)
accuracy_list.append(accuracy)
print("Best Model Test accuracy:", accuracy)
acc_mll_df = pd.DataFrame(accuracy_list, columns=['accuracy'])
acc_mll_df['mll'] = mll_list_que[-1]
save_result(acc_mll_df, save_dir, name='_mll_accuracy')
def test(model, result_file):
print('Begin testing')
model.eval()
dataset = Text2EmojiTestset()
testloader = DataLoader(dataset, batch_size=10000,
shuffle=False, num_workers=4)
pred = []
with torch.no_grad():
for data in testloader:
pred += list(predict(model(data)))
save_result(result_file, pred)
model.train()
print('Testing complete')
def home():
if request.method=="GET":
return render_template("form.html")
else:
button=request.form['button']
if button=='Cancel':
return redirect(url_for('home'))
elif button=='Results':
return redirect(url_for('results'))
name=request.form['username']
rating=request.form['rating']
assert name != ""
utils.save_result(name,rating)
flash("Name:%s Rating:%s"%(name,rating))
return redirect(url_for('home'))
def train(model,
dataloader,
epochs,
optimizer,
criterion,
save_output_every=10,
save_model_every=50,
only_show=False):
device = utils.get_device()
for epoch in tqdm(range(1, epochs + 1)):
losses_per_epoch = []
accuracies_per_epoch = []
# go over all batches
for step, (input, target) in enumerate(dataloader):
model.train()
input, target = input.to(device), target.to(device)
target = utils.preprocess_target(target)
pred = model(input)
optimizer.zero_grad()
loss = criterion(pred, target)
loss.backward()
optimizer.step()
acc = utils.get_accuracy(pred, target)
losses_per_epoch.append(loss.item())
accuracies_per_epoch.append(acc)
mean_loss = np.mean(losses_per_epoch)
mean_acc = np.mean(accuracies_per_epoch)
print(f'{"-"*30} Epoch {epoch} {"-"*30}')
print('Loss: %.3f Accuracy: %.3f' % (mean_loss, mean_acc))
if epoch % save_output_every == 0:
utils.save_result(epoch,
input,
pred,
target,
name='epoch',
only_show=only_show)
if epoch % save_model_every == 0:
utils.save_model(epoch, model)
def val_epoch(model,
ratings,
negs,
K,
ctx,
output=None,
epoch=None,
processes=1):
thistime = time.time()
if epoch is None:
print("Initial evaluation")
else:
print("Epoch {} evaluation".format(epoch))
start = datetime.now()
# model.eval()
if processes > 1:
context = mp.get_context('spawn')
_eval_one = partial(eval_one, model=model, K=K, ctx=ctx)
with context.Pool(processes=processes) as workers:
zip(ratings, negs)
hits_and_ndcg = workers.starmap(_eval_one, zip(ratings, negs))
hits, ndcgs = zip(*hits_and_ndcg)
else:
hits, ndcgs = [], []
for rating, items in zip(ratings, negs):
hit, ndcg = eval_one(rating, items, model, K, ctx=ctx)
hits.append(hit)
ndcgs.append(ndcg)
hits = np.array(hits, dtype=np.float32)
ndcgs = np.array(ndcgs, dtype=np.float32)
end = datetime.now()
if output is not None:
result = OrderedDict()
result['timestamp'] = datetime.now()
result['duration'] = end - start
result['epoch'] = epoch
result['K'] = K
result['hit_rate'] = np.mean(hits)
result['NDCG'] = np.mean(ndcgs)
utils.save_result(result, output)
print("epoch time:")
print(time.time() - thistime)
return hits, ndcgs
def main_experiment(env_parameter, agent_parameter, experiment_parameter):
print "EXPERIMENT START"
numRuns = experiment_parameter[0]
numEpisodes = experiment_parameter[1]
max_steps = experiment_parameter[2]
average_steps = [0.0 for i in range(numEpisodes)]
average_acc_reward = [0.0 for i in range(numEpisodes)]
global numSteps, accumulated_reward
for i in range(numRuns):
print "Runs", i, "start..."
task_env, task_agent = RL_init(env_parameter, agent_parameter)
for j in range(numEpisodes):
RL_epispdes(task_env, task_agent, max_steps)
average_steps[j] += numSteps / (numRuns + 1.0)
average_acc_reward[j] += accumulated_reward / (numRuns + 1.0)
if i == numRuns - 1:
print "test mode"
test(task_env, task_agent, max_steps)
utils.save_result(average_steps, "avg_steps.txt")
utils.save_result(average_acc_reward, "avg_acc_reward.txt")
np.save("Q.npy", task_agent.Q_value)
plot_steps = plt.figure(1)
plt.plot(average_steps)
plt.ylabel('Steps') #'# of steps'
plt.xlabel('Episodes') #'# of episodes'
#plot_steps.show()
plot_reward = plt.figure(2)
plt.plot(average_acc_reward)
plt.ylabel('Accumulated Rewards') #'# of steps'
plt.xlabel('Episodes') #'# of episodes'
#plot_reward.show()
plt.show()
print "EXPERIMENT END"
pass
def train_model_sghmc(cfg, model, Xvalid, Yvalid, Xtest, Ytest, writer,
save_dir):
mll_max = -np.inf
best_iter = 0
best_model = model
for i in tdqm(range(cfg.train.iterations),
ascii=" .oO0",
bar_format="{total_time}: {percentage:.0f}%|{bar}{r_bar}"):
model.sghmc_step()
if i % cfg.train.train_hyper == 0:
summary = model.train_hypers()
writer.add_summary(summary, global_step=i)
print("Iteration", i, end='\r')
mll, sum_mll = model.print_sample_performance()
print(f"MLL: {mll}")
# set_trace()
writer.add_summary(sum_mll, global_step=i)
if np.round(mll - mll_max, decimals=5) > 0:
print('MLL increased ({:.7f} --> {:.7f}). Updating values ....'.
format(mll_max, mll))
mll_max = mll
best_model = model
best_iter = i
if cfg.save_model:
# save best model
print(
f'################## save best model at {save_dir} ##################'
)
model_name = f'{best_iter}_bestmodel_{mll_max}'
best_model.save(save_dir, name=model_name)
accuracy = measure_accuracy(best_model, Xvalid, Yvalid, collect=True)
acc_mll_df = pd.DataFrame([accuracy], columns=['valid_accuracy'])
acc_mll_df['mll'] = mll_max
accuracy = measure_accuracy(best_model, Xtest, Ytest)
acc_mll_df['test_accuracy'] = accuracy
print(f"Best Model Test accuracy: {accuracy} with MLL {mll_max}")
# export accuracy
save_result(acc_mll_df, save_dir, name='_mll_accuracy')
def color_adjust(x, args):
if args.iter % args.save_intervel == 0:
save_result(x, args)
args.iter += 1
# Input for VGG
x_vgg = np.asarray(np.reshape(x, args.shape), dtype=np.float32)
x_vgg_var = Variable(chainer.dataset.concat_examples([x_vgg], args.gpu))
# Poisson loss
poisson_loss = F.mean_squared_error(
(args.content_laplace + args.border_sum) * args.mask_var,
F.convolution_2d(x_vgg_var * args.mask_var, W=args.W_laplace, pad=1) *
args.mask_var)
poisson_loss *= np.prod(x_vgg_var.shape)
# tv loss
tv_loss = total_variation(x_vgg_var)
# Concent loss
content_loss = 0
x_features = extract({'data': x_vgg_var}, args.vgg, args.content_layers)
x_features = {key: value[0] for key, value in x_features.items()}
for layer in args.content_layers:
content_loss += F.mean_squared_error(args.content_features[layer],
x_features[layer])
# Realism loss
y = args.realism_cnn(x_vgg_var, dropout=False)
b, _, w, h = y.shape
xp = cuda.get_array_module(x_vgg_var.data)
realism_loss = F.sum(y[:, 0, :, :])
loss = args.poisson_weight * poisson_loss + args.realism_weight * realism_loss + args.tv_weight * tv_loss + args.content_weight * content_loss
# Backward
loss.backward()
# Transfer loss & diff from GPU to CPU
loss = cuda.to_cpu(loss.data)
dx = np.squeeze(cuda.to_cpu(x_vgg_var.grad))
return loss, np.asarray(dx.flatten(), dtype=np.float64)
def val_epoch(model, ratings, negs, K, use_cuda=True, output=None, epoch=None):
start = datetime.now()
hits, ndcgs = [], []
for rating, items in zip(ratings, negs):
hit, ndcg = eval_one(rating, items, model, K, use_cuda=use_cuda)
hits.append(hit)
ndcgs.append(ndcg)
hits = np.array(hits, dtype=np.float32)
ndcgs = np.array(ndcgs, dtype=np.float32)
end = datetime.now()
if output is not None:
result = OrderedDict()
result['timestamp'] = datetime.now()
result['duration'] = end - start
result['epoch'] = epoch
result['K'] = K
result['hit_rate'] = np.mean(hits)
result['NDCG'] = np.mean(ndcgs)
utils.save_result(result, output)
return hits, ndcgs
np.reshape(Theta[:hidden_layer_size * (input_layer_size + 1)], (hidden_layer_size, input_layer_size + 1), order='F'))
Theta2 = np.matrix(
np.reshape(Theta[hidden_layer_size * (input_layer_size + 1):], (num_labels, hidden_layer_size + 1), order='F'))
p = fnx.predict(Theta1, Theta2, X)
precision = 0
for i in range(len(y)):
if y[i] == p[i]:
precision += 1
print('Training Set Accuracy:', (1.0 * precision) / len(y))
return Theta1, Theta2
if __name__ == '__main__':
cuisine_list, ingredients_list, X, y = utl.load_train('number')
ingredients_count = len(ingredients_list)
cuisines_count = len(cuisine_list)
Theta1, Theta2 = train_nn(ingredients_count, ingredients_count//16, cuisines_count, X, y)
T, ids = utl.load_test(ingredients_list)
p = fnx.predict(Theta1, Theta2, T)
utl.save_result('nn', cuisine_list, p, ids)
def main(para):
#data = DH.load_data()
data = para['data']
MODEL = para['do_what'] #hold_out submission cv tune
CALIB = para['callib']
BAG = para['bag']
xgb_para = {'base_score': 0.5,
'colsample_bylevel': 1,
'colsample_bytree': 0.3,
'gamma': 0,
'learning_rate': 0.04,
'max_delta_step': 0,
'max_depth': 8,
'min_child_weight': 5,
'missing': None,
'n_estimators': 2,
'nthread': 4,
'objective': 'binary:logistic',
'reg_alpha': 0,
'reg_lambda': 700,
'scale_pos_weight': 15,
'seed': 0,
'silent': True,
'subsample': 0.7}
xgb_para['n_estimators'] = 1500
clf = xgb.XGBClassifier(**xgb_para)
if MODEL == 'tune':
def objective(args):
print args
rf_para = {'bootstrap': True,
'class_weight': None,
'criterion': 'gini',
'max_depth': None,
'max_features': 'auto',
'max_leaf_nodes': None,
'min_samples_leaf': 1,
'min_samples_split': 2,
'min_weight_fraction_leaf': 0.0,
'n_estimators': 10,
'n_jobs': 1,
'oob_score': False,
'random_state': None,
'verbose': 0,
'warm_start': False}
n_estimators,max_depth, max_features= args
rf_para['n_estimators'] = int(n_estimators) + 1
rf_para['max_depth'] = int(max_depth)
rf_para['max_features'] = max_features
rf_para['random_state'] = 4
rf_para['n_jobs'] = 4
rf_para['class_weight'] = 'balanced'
clf = RandomForestClassifier(**rf_para)
clf.fit(data['val_train_x'],data['val_train_y'])
pred = clf.predict_proba(data['val_test_x'])
score = roc_auc_score(data['val_test_y'],pred[:,1])*-1.
rsv = {
'loss': score,
'status': STATUS_OK,
# -- store other results like this
'n_estimators': n_estimators,
'max_depth':max_depth,
}
return rsv
space = (
hp.quniform('n_estimators',100,1000,100),
hp.quniform('max_depth', 4, 10, 1),
hp.quniform('max_features', 0.1,1., 0.2)
)
trials = Trials()
best_sln = fmin(objective, space, algo=tpe.suggest, max_evals=40,trials=trials)
rinfo = trials.results
df = pd.DataFrame(rinfo)
df.to_csv('./tune.csv',index=False)
print best_sln
if MODEL == 'submission':
print 'working'
if CALIB:
tp = {}
tp['base_estimator'] = clf
tp['method'] = 'sigmoid'
tp['cv'] = 5
clf = CalibratedClassifierCV(**tp)
if BAG:
bagp = {'base_estimator': None,
'bootstrap': True,
'bootstrap_features': False,
'max_features': 1.0,
'max_samples': 1.0,
'n_estimators': 10,
'n_jobs': 1,
'oob_score': False,
'random_state': None,
'verbose': 0,
'warm_start': False}
bagp['random_state'] = 4
bagp['base_estimator'] = clf
bagp['n_jobs'] = 4
clf = BaggingClassifier(**bagp)
print 'fiting'
clf.fit(data['train_x'],data['train_y'])
test_x = data['test'].drop('id',axis=1)
p = clf.predict_proba(test_x)[:,0]
test = data['test']
test['score'] = p
utils.save_result('xgb_pre.csv', test[['id','score']])
logger.info('submission done!')
if MODEL == 'hold_out':
if CALIB:
tp = {}
tp['base_estimator'] = clf
tp['method'] = 'sigmoid'
tp['cv'] = 5
tp['cv'] = StratifiedKFold(data['val_train_y'], n_folds=5,shuffle=True,\
random_state=4)
clf = CalibratedClassifierCV(**tp)
if BAG:
bagp = {'base_estimator': None,
'bootstrap': True,
'bootstrap_features': False,
'max_features': 1.0,
'max_samples': 1.0,
'n_estimators': 10,
'n_jobs': 1,
'oob_score': False,
'random_state': None,
'verbose': 0,
'warm_start': False}
bagp['random_state'] = 4
bagp['base_estimator'] = clf
bagp['n_jobs'] = 4
clf = BaggingClassifier(**bagp)
if not CALIB:
clf.fit(data['val_train_x'],data['val_train_y'],verbose=True,
eval_set=[(data['val_test_x'],data['val_test_y'])],eval_metric='auc',early_stopping_rounds=2100)
print clf.best_score, clf.best_iteration
rs = 'best_score %f, best_iteration %d ' %(clf.best_score, clf.best_iteration)
logger.info(rs)
logger.info('score %f, model: %s'%(clf.best_score, clf))
else:
clf.fit(data['val_train_x'],data['val_train_y'])
p1 = clf.predict_proba(data['val_test_x'])[:,1]
score = roc_auc_score(data['val_test_y'],p1)
logger.info('score %f, model: %s'%(score,clf))
def main():
parser = argparse.ArgumentParser(
description='Modified Poisson image editing')
parser.add_argument('--poisson_weight',
type=float,
default=1,
help='Weight for poisson loss')
parser.add_argument('--content_weight',
type=float,
default=5e-4,
help='Weight for content loss')
parser.add_argument('--tv_weight',
type=float,
default=1e-3,
help='Weight for tv loss')
parser.add_argument('--n_iteration',
type=int,
default=3500,
help='# of iterations')
parser.add_argument('--save_intervel',
type=int,
default=100,
help='save result every # of iterations')
parser.add_argument('--rand_init',
type=lambda x: x == 'True',
default=True,
help='Random init input if True')
parser.add_argument('--gpu',
type=int,
default=0,
help='GPU ID (negative value indicates CPU)')
parser.add_argument(
'--data_root',
default='/data1/wuhuikai/benchmark/TransientAttributes/imageCropped',
help='Root folder for cropped transient attributes dataset')
parser.add_argument('--load_size',
type=int,
default=224,
help='Scale image to load_size')
parser.add_argument('--total_instance',
type=int,
default=175,
help='# of instance to run test')
parser.add_argument('--seed', type=int, default=5, help='Random seed')
parser.add_argument('--min_ratio',
type=float,
default=0.2,
help='Min ratio for size of random rect')
parser.add_argument('--max_ratio',
type=float,
default=0.6,
help='Max ratio for size of random rect')
parser.add_argument('--result_folder',
default='transient_attributes_result/modified_result',
help='Name for folder storing results')
parser.add_argument('--result_name',
default='loss.txt',
help='Name for file saving loss change')
args = parser.parse_args()
print('Input arguments:')
for key, value in vars(args).items():
print('\t{}: {}'.format(key, value))
print('')
args.prefix_name = '_'.join(
sorted([
'{}({})'.format(key, value) for key, value in vars(args).items()
if key in set([
'poisson_weight', 'realism_weight', 'content_weight',
'tv_weight', 'n_iteration', 'rand_init'
])
]))
# Init image list
print('Load images from {} ...'.format(args.data_root))
folders = [
folder for folder in os.listdir(args.data_root)
if os.path.isdir(os.path.join(args.data_root, folder))
]
imgs_each_folder = {
folder: glob.glob(os.path.join(args.data_root, folder, '*'))
for folder in folders
}
print('\t {} images in {} folders in total ...\n'.format(
np.sum([len(v) for k, v in imgs_each_folder.items()]), len(folders)))
# Init result folder
if not os.path.isdir(args.result_folder):
os.makedirs(args.result_folder)
print('Result will save to {} ...\n'.format(args.result_folder))
# Init Constant Variable
args.W_laplace = Variable(make_kernel(
3, 3,
np.asarray([[-1, -1, -1], [-1, 8, -1], [-1, -1, -1]],
dtype=np.float32)),
volatile='auto')
args.W_laplace.to_gpu()
loss_change = []
np.random.seed(args.seed)
for i in range(args.total_instance):
folder = np.random.choice(folders)
print('Processing {}/{}, select folder {} ...'.format(
i + 1, args.total_instance, folder))
obj_path, bg_path = np.random.choice(imgs_each_folder[folder],
2,
replace=False)
print('\tObj: {}, Bg: {} ...'.format(os.path.basename(obj_path),
os.path.basename(bg_path)))
obj_img = imread(obj_path)
args.orig_size = obj_img.shape[:2]
obj_vgg = im_preprocess_vgg(obj_img, args.load_size, dtype=np.float32)
bg_vgg = im_preprocess_vgg(imread(bg_path),
args.load_size,
dtype=np.float32)
args.shape = obj_vgg.shape
# random rect
w, h = np.asarray(
np.random.uniform(args.min_ratio, args.max_ratio, 2) *
args.load_size,
dtype=np.uint32)
sx, sy = np.random.randint(0, args.load_size - w), np.random.randint(
0, args.load_size - h)
expand_mask = np.zeros((1, args.load_size, args.load_size),
dtype=np.float32)
expand_mask[:, sx:sx + w, sy:sy + h] = 1
mask = np.zeros_like(expand_mask)
mask[:, sx + 1:sx + w - 1, sy + 1:sy + h - 1] = 1
inverse_mask = 1 - mask
## vars
obj_var = Variable(chainer.dataset.concat_examples([obj_vgg],
args.gpu),
volatile='on')
bg_var = Variable(chainer.dataset.concat_examples([bg_vgg], args.gpu),
volatile='auto')
mask_var = F.broadcast_to(
Variable(chainer.dataset.concat_examples([mask], args.gpu),
volatile='auto'), obj_var.shape)
inverse_mask_var = F.broadcast_to(
Variable(chainer.dataset.concat_examples([inverse_mask], args.gpu),
volatile='auto'), obj_var.shape)
inverse_border_mask_var = F.broadcast_to(
Variable(chainer.dataset.concat_examples([1 - expand_mask + mask],
args.gpu),
volatile='auto'), obj_var.shape)
## Laplace
content_laplace = F.convolution_2d(
obj_var, W=args.W_laplace, pad=1) * mask_var + F.convolution_2d(
bg_var, W=args.W_laplace, pad=1) * inverse_mask_var
content_laplace.volatile = 'off'
copy_paste_vgg = obj_vgg * mask + bg_vgg * inverse_mask
## args
args.content_laplace = content_laplace
args.mask = mask
args.inverse_mask = inverse_mask
args.inverse_mask_var = inverse_mask_var
args.inverse_border_mask_var = inverse_border_mask_var
args.bg_vgg = bg_vgg
args.bg_var = bg_var
args.copy_paste_vgg = copy_paste_vgg
args.im_name = 'folder_{}_obj_{}_bg_{}'.format(
folder,
os.path.splitext(os.path.basename(obj_path))[0],
os.path.splitext(os.path.basename(bg_path))[0])
args.iter = 0
x_init = np.asarray(
np.random.randn(*args.shape) * 0.001,
dtype=np.float32) if args.rand_init else np.copy(copy_paste_vgg)
print('\tOptimize start ...')
res = minimize(color_adjust,
x_init,
args=(args),
method='L-BFGS-B',
jac=True,
options={
'maxiter': args.n_iteration,
'disp': True
})
# Cut and paste loss
args.iter = -1
f0, _ = color_adjust(copy_paste_vgg, args)
print('\tOptimize done, loss = {} from {}\n'.format(res.fun, f0))
loss_change.append((args.im_name, f0, res.fun))
args.iter = ''
save_result(res.x, args)
with open(os.path.join(args.result_folder, args.result_name), 'w') as f:
for name, f0, fb in loss_change:
f.write('{} {} {}\n'.format(name, f0, fb))
print_freq=10,
verbose=True)
# sa.find(stop_fun=lambda fitness: fitness == fp.max_possible_fit)
result_record.append(sa.find())
result = nn.evaluate_on_all_datasets(sa)
result_train.append(result[0])
result_val.append(result[1])
result_test.append(result[2])
print("takes %d seconds" % (datetime.now() - start_time).seconds)
print("Averaged train result: %f" % np.mean(result_train))
print("Averaged val result: %f" % np.mean(result_val))
print("Averaged test result: %f" % np.mean(result_test))
plt.plot(range(10, 50001, 10), np.mean(result_record, axis=0))
plt.savefig("fig/nn_sa_9995")
plt.show()
save_result(result_record, "nn_sa_9995")
# sa = SimulatedAnnealing(params, neighbor_nn, nn, max_try_per_step=10000, max_iter=4000, temperature_decay=0.997)
# sa.find()
# nn.evaluate_on_all_datasets(sa)
# def crossover(params):
# first, second = params
# offspring1 = {}
# offspring2 = {}
# for weight_name in first:
# ran = np.random.random(size=first[weight_name].shape)
# offspring1[weight_name] = np.where(ran < 0.5, first[weight_name], second[weight_name])
# offspring2[weight_name] = np.where(ran > 0.5, first[weight_name], second[weight_name])
# return offspring1, offspring2
#
y = tf.nn.softmax(tf.matmul(x, W) + b)
y_ = tf.placeholder(tf.float32, [None, cuisine_count])
t = tf.placeholder(tf.float32, [None, ingredients_count])
p = tf.nn.softmax(tf.matmul(t, W) + b)
cross_entropy = -tf.reduce_sum(y_*tf.log(y))
train_step = tf.train.GradientDescentOptimizer(0.001).minimize(cross_entropy)
# train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
init = tf.initialize_all_variables()
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
sess.run(init)
for d in ['/cpu:0', '/gpu:0']:
with tf.device(d):
for i in range(100):
batch_xs, batch_ys = utl.next_batch(xs, ys, 12000)
sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print(sess.run(accuracy, feed_dict={x: xs, y_: ys}))
cls = sess.run(p, feed_dict={t: ts})
utl.save_result('tf_gd', cuisine_list, np.argmax(cls, axis=1).tolist(), ids, 'number')
y_ = tf.placeholder(tf.float32, [None, cuisine_count])
t = tf.placeholder(tf.float32, [None, ingredients_count])
p = tf.nn.softmax(tf.matmul(t, W) + b)
cross_entropy = -tf.reduce_sum(y_ * tf.log(y))
train_step = tf.train.GradientDescentOptimizer(0.001).minimize(cross_entropy)
# train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
init = tf.initialize_all_variables()
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
sess.run(init)
for d in ['/cpu:0', '/gpu:0']:
with tf.device(d):
for i in range(100):
batch_xs, batch_ys = utl.next_batch(xs, ys, 12000)
sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
print(sess.run(accuracy, feed_dict={x: xs, y_: ys}))
cls = sess.run(p, feed_dict={t: ts})
utl.save_result('tf_gd', cuisine_list,
np.argmax(cls, axis=1).tolist(), ids, 'number')
def main():
parser = argparse.ArgumentParser(
description='Poisson image editing using RealismCNN')
parser.add_argument('--poisson_weight',
type=float,
default=1,
help='Weight for poisson loss')
parser.add_argument('--realism_weight',
type=float,
default=1e4,
help='Weight for realism loss')
parser.add_argument('--content_weight',
type=float,
default=1,
help='Weight for content loss')
parser.add_argument('--tv_weight',
type=float,
default=1e-1,
help='Weight for tv loss')
parser.add_argument('--n_iteration',
type=int,
default=1000,
help='# of iterations')
parser.add_argument('--save_intervel',
type=int,
default=100,
help='save result every # of iterations')
parser.add_argument('--rand_init',
type=lambda x: x == 'True',
default=True,
help='Random init input if True')
parser.add_argument('--content_layers',
type=str2list,
default='conv4_1',
help='Layers for content_loss, sperated by ;')
parser.add_argument('--gpu',
type=int,
default=0,
help='GPU ID (negative value indicates CPU)')
parser.add_argument('--realism_model_path',
default='model/realismCNN_all_iter3.npz',
help='Path for pretrained Realism model')
parser.add_argument('--content_model_path',
default='model/VGG_ILSVRC_19_layers.pkl',
help='Path for pretrained VGG model')
parser.add_argument(
'--data_root',
default='/data1/wuhuikai/benchmark/Realistic/color_adjustment',
help='Root folder for color adjustment dataset')
parser.add_argument('--img_folder',
default='pngimages',
help='Folder for stroing images')
parser.add_argument('--list_name',
default='list.txt',
help='Name for file storing image list')
parser.add_argument('--load_size',
type=int,
default=224,
help='Scale image to load_size')
parser.add_argument('--result_folder',
default='image_editing_result',
help='Name for folder storing results')
parser.add_argument('--result_name',
default='loss.txt',
help='Name for file saving loss change')
args = parser.parse_args()
args.content_layers = set(args.content_layers)
print('Input arguments:')
for key, value in vars(args).items():
print('\t{}: {}'.format(key, value))
print('')
args.prefix_name = '_'.join(
sorted([
'{}({})'.format(key, value) for key, value in vars(args).items()
if key not in set([
'realism_model_path', 'content_model_path', 'data_root',
'img_folder', 'list_name', 'result_folder', 'result_name'
])
]))
# Init CNN model
realism_cnn = RealismCNN()
print('Load pretrained Realism model from {} ...'.format(
args.realism_model_path))
serializers.load_npz(args.realism_model_path, realism_cnn)
print('Load pretrained VGG model from {} ...\n'.format(
args.content_model_path))
with open(args.content_model_path, 'rb') as f:
vgg = pickle.load(f)
if args.gpu >= 0:
chainer.cuda.get_device(args.gpu).use() # Make a specified GPU current
realism_cnn.to_gpu() # Copy the model to the GPU
vgg.to_gpu()
# Init image list
im_root = os.path.join(args.data_root, args.img_folder)
print('Load images from {} according to list {} ...'.format(
im_root, args.list_name))
with open(os.path.join(args.data_root, args.list_name)) as f:
im_list = f.read().strip().split('\n')
total = len(im_list)
print('{} images loaded done!\n'.format(total))
# Init result folder
if not os.path.isdir(args.result_folder):
os.makedirs(args.result_folder)
print('Result will save to {} ...\n'.format(args.result_folder))
# Init Constant Variable
W_laplace = Variable(make_kernel(
3, 3,
np.asarray([[0, -1, 0], [-1, 4, -1], [0, -1, 0]], dtype=np.float32)),
volatile='auto')
W_laplace.to_gpu()
args.W_laplace = W_laplace
W_sum = Variable(make_kernel(
3, 3, np.asarray([[0, 1, 0], [1, 0, 1], [0, 1, 0]], dtype=np.float32)),
volatile='auto')
W_sum.to_gpu()
loss_change = []
for idx, im_name in enumerate(im_list):
print('Processing {}/{}, name = {} ...'.format(idx + 1, total,
im_name))
obj_vgg = im_preprocess_vgg(imread(
os.path.join(im_root, '{}_obj.png'.format(im_name))),
args.load_size,
dtype=np.float32)
bg_vgg = im_preprocess_vgg(imread(
os.path.join(im_root, '{}_bg.png'.format(im_name))),
args.load_size,
dtype=np.float32)
expand_mask = im_preprocess_vgg(imread(
os.path.join(im_root, '{}_softmask.png'.format(im_name))),
args.load_size,
sub_mean=False,
dtype=np.uint8,
preserve_range=False)
args.orig_size = (args.load_size, args.load_size)
args.shape = bg_vgg.shape
## mask
mask = erosion(np.squeeze(expand_mask), np.ones((3, 3),
dtype=np.uint8))
mask = np.asarray(mask[np.newaxis, :, :], dtype=np.float32)
expand_mask = np.asarray(expand_mask, dtype=np.float32)
inverse_mask = 1 - mask
## vars
obj_var = Variable(chainer.dataset.concat_examples([obj_vgg],
args.gpu),
volatile='on')
mask_var = F.broadcast_to(
Variable(chainer.dataset.concat_examples([mask], args.gpu)),
obj_var.shape)
## Laplace
content_laplace = F.convolution_2d(obj_var, W=W_laplace, pad=1)
content_laplace.volatile = 'off'
# prefilled
border = bg_vgg * expand_mask * inverse_mask
border_var = Variable(chainer.dataset.concat_examples([border],
args.gpu),
volatile='on')
border_sum = F.convolution_2d(border_var, W=W_sum, pad=1)
border_sum.volatile = 'off'
print('\tExtracting content image features ...')
copy_paste_vgg = obj_vgg * mask + bg_vgg * inverse_mask
copy_paste_var = Variable(chainer.dataset.concat_examples(
[copy_paste_vgg], args.gpu),
volatile='on')
content_features = extract({'data': copy_paste_var}, vgg,
args.content_layers)
content_features = {
key: value[0]
for key, value in content_features.items()
}
for _, value in content_features.items():
value.volatile = 'off'
## args
args.vgg = vgg
args.realism_cnn = realism_cnn
args.border_sum = border_sum
args.content_laplace = content_laplace
args.content_features = content_features
args.mask = mask
args.mask_var = mask_var
args.inverse_mask = inverse_mask
args.bg_vgg = bg_vgg
args.copy_paste_vgg = copy_paste_vgg
args.im_name = im_name
args.iter = 0
x_init = np.asarray(
np.random.randn(*args.shape) * 0.001,
dtype=np.float32) if args.rand_init else np.copy(copy_paste_vgg)
print('\tOptimize start ...')
res = minimize(color_adjust,
x_init,
args=(args),
method='L-BFGS-B',
jac=True,
options={
'maxiter': args.n_iteration,
'disp': False
})
# Cut and paste loss
args.iter = -1
f0, _ = color_adjust(copy_paste_vgg, args)
print('\tOptimize done, loss = {} from {}\n'.format(res.fun, f0))
loss_change.append((im_name, f0, res.fun))
args.iter = ''
save_result(res.x, args)
with open(os.path.join(args.result_folder, args.result_name), 'w') as f:
for name, f0, fb in loss_change:
f.write('{} {} {}\n'.format(name, f0, fb))
def test_save_result(self, makedirs_mock, savefig_mock):
makedirs_mock.return_value = None
savefig_mock.return_value = None
save_result(self.z_k_value, self.iteration, 'distribution_name')
from sklearn.multiclass import OneVsRestClassifier
from sklearn.linear_model import LogisticRegression
import utils as utl
cuisine_list, ingredients_list, x, y = utl.load_train('number')
classifier = OneVsRestClassifier(LogisticRegression(C=1e6)).fit(x, y)
p = classifier.predict(x)
precision = 0
for i in range(len(y)):
if y[i] == p[i]:
precision += 1
accuracy = (1.0 * precision) / len(y)
print('Training Set Accuracy:', accuracy)
t, ids = utl.load_test(ingredients_list)
p = classifier.predict(t)
utl.save_result('sk_lr', cuisine_list, p, ids, 'number')
project_name = "itext"
elif repo_id == "6":
project_name = "spring"
elif repo_id == "7":
project_name = "lucene"
else:
print "repository id not exist"
try:
db = Database(user, passwd, database)
cnn = db.connect()
cursor = cnn.cursor()
if no_parse != True:
change_info = get_change_line(cursor, db, start_datetime, end_datetime, repo_id, change_type)
save_file = change_file_path + change_type + "_" + project_name + ".csv"
save_result(change_info, save_file)
change_metric = get_change_metric(cursor, db, previous_datetime, start_datetime, repo_id, change_type)
changemetric_file = changemetric_file_path + change_type + "_" + project_name + ".csv"
save_result(change_metric, changemetric_file)
cnn.close()
if extensions != None:
emg = _get_extensions_manager(extensions)
change_file = change_file_path + change_type + "_" + project_name + ".csv"
changemetric_file = changemetric_file_path + change_type + "_" + project_name + ".csv"
emg.run_extensions(
change_file, understand_file, stat_file, rank_file, changemetric_file, predict_file, change_type
)
# print "Finish. Repository: "+ repo_id +", time: " + start_time+"-" +end_time+", bug: "+str(buggy) + ", Num_hunks: "+str(total_num_hunk)
except MySQLdb.Error, e:
print "Error %d: %s" % (e.args[0], e.args[1])
torch.mean(torch.FloatTensor(G_losses))))
########## test ##########
G.eval() # stop train and start test
z = torch.randn((5*5, 100)).view(-1, 100, 1, 1)
if GPU_MODE:
z = Variable(z.cuda(), volatile=True)
else:
z = Variable(z, volatile=True)
random_image = G(z)
fixed_image = G(fixed_z)
G.train() # stop test and start train
p = DIR + '/Random_results/MNIST_GAN_' + str(epoch + 1) + '.png'
fixed_p = DIR + '/Fixed_results/MNIST_GAN_' + str(epoch + 1) + '.png'
utils.save_result(random_image, (epoch+1), save=True, path=p)
utils.save_result(fixed_image, (epoch+1), save=True, path=fixed_p)
train_hist['D_losses'].append(torch.mean(torch.FloatTensor(D_losses)))
train_hist['G_losses'].append(torch.mean(torch.FloatTensor(G_losses)))
train_hist['per_epoch_times'].append(per_epoch_time)
end_time = time()
total_time = end_time - end_time
print("Avg per epoch time: %.2f, total %d epochs time: %.2f" % (torch.mean(torch.FloatTensor(train_hist['per_epoch_times'])), EPOCH, total_time))
print("Training finish!!!...")
# save parameters
torch.save(G.state_dict(), DIR + "/generator_param.pkl")
torch.save(D.state_dict(), DIR + "/discriminator_param.pkl")
# save history
from sklearn.multiclass import OneVsRestClassifier
from sklearn.linear_model import LogisticRegression
import utils as utl
cuisine_list, ingredients_list, x, y = utl.load_train('number')
classifier = OneVsRestClassifier(LogisticRegression(C=1e6)).fit(x, y)
p = classifier.predict(x)
precision = 0
for i in range(len(y)):
if y[i] == p[i]:
precision += 1
accuracy = (1.0 * precision) / len(y)
print('Training Set Accuracy:', accuracy)
t, ids = utl.load_test(ingredients_list)
p = classifier.predict(t)
utl.save_result('sk_lr', cuisine_list, p, ids, 'number')
shuffle=True,
random_state=4891)
for i in range(1, MAX_POWER + 1):
clf = KNeighborsClassifier(n_neighbors=2**i, n_jobs=1)
clf = CalibratedClassifierCV(clf, method='isotonic', cv=kf)
utils.VJUH(X_train,
X_test,
y_train,
clf,
MODEL_NAME + str(i),
'ensemble',
stratify_labels=utils.load_stratify_labels())
train = []
test = []
for i in range(1, MAX_POWER + 1):
tr, ts, _ = utils.load_data(data_name=MODEL_NAME + str(i),
data_folder='ensemble')
train.append(tr)
test.append(ts)
result = pd.concat(train, axis=1, ignore_index=True)
utils.save_result(result.values,
MODEL_NAME[:-1],
data_folder='ensemble',
test=False)
result = pd.concat(test, axis=1, ignore_index=True)
utils.save_result(result.values,
MODEL_NAME[:-1],
data_folder='ensemble',
test=True)
