def main():
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
DATASET_DIRECTORY = '../data_part1'
X, y, X_hidden = dataset_manip.load_dataset(DATASET_DIRECTORY)
num_classes = len(set(y))
print('X.shape = ' + str(X.shape))
print('X_hidden.shape = ' + str(X_hidden.shape))
ens = Ensemble(input_shape=(77, 71, 1),
num_classes=10,
num_models=11,
batch_size=512,
path='./ensemble_files',
load=False)
ens.train(X=X, y=y, epochs_per_model=300, split_rate=0.9)
print(ens.measure_accuracy(X, y))
return
X_train, X_validation, y_train, y_validation = dataset_manip.split_dataset(
X, y, rate=0.5)
model = Model(image_shape=X.shape[1:],
num_classes=num_classes,
model_path='./model_files/model',
batch_size=512,
first_run=True) # 1250
model.train(X_train, y_train, X_validation, y_validation, 500)
model.train_unsupervised(X_hidden, X_validation, y_validation, 200)
print('Final Accuracy: {}'.format(
model.measure_accuracy(X_validation, y_validation)))
def __init__(self, sourceFile, targetFile):
self.SWindow = []
self.TWindow = []
self.TPredictWindow = []
self.SDataBuffer = [] #Queue
self.TDataBuffer = [] #Queue
self.SInitialDataBuffer = []
self.TInitialDataBuffer = []
self.changeDetector = ChangeDetection(Properties.GAMMA,
Properties.SENSITIVITY,
Properties.MAX_WINDOW_SIZE)
self.ensemble = Ensemble(Properties.ENSEMBLE_SIZE)
classNameList = []
self.source = Stream(sourceFile, classNameList,
Properties.INITIAL_DATA_SIZE)
self.target = Stream(targetFile, classNameList,
Properties.INITIAL_DATA_SIZE)
Properties.MAXVAR = self.source.MAXVAR
self.gateway = JavaGateway(
start_callback_server=True,
gateway_parameters=GatewayParameters(port=Properties.PY4JPORT),
callback_server_parameters=CallbackServerParameters(
port=Properties.PY4JPORT + 1))
self.app = self.gateway.entry_point
def process_all_images(config):
filenames = sorted(config.files)
tables = {}
segmentations = {}
ensembles = {}
for segmentation in config.segmentations:
segmentations[segmentation] = {}
ensembles['walker_binary'] = {}
ensembles['opt'] = {}
erosions = {}
methods = {'unet': {}, 'walker_binary': {}, 'opt': {}}
for key in ['jac', 'af1', 'merge_rate', 'split_rate']:
tables[key] = pd.DataFrame(columns=list(methods.keys()), copy=True)
os.makedirs(config.output, exist_ok=True)
root_dir = os.path.join(config.output, config.filename)
if os.path.exists(root_dir):
shutil.rmtree(root_dir)
os.makedirs(root_dir, exist_ok=True)
counter = 0
for file in filenames:
if counter % 50 == 0:
print(counter)
if counter == config.counter:
break
annot_path = os.path.join(config.annot, file.strip())
annot = skimage.io.imread(annot_path, as_gray=True)
for segmentation in segmentations.keys():
path = os.path.join(config.root, segmentation, file.strip())
segmentations[segmentation]['orig'] = skimage.io.imread(
path, as_gray=True)
segmentations[segmentation][
'results'] = comp.get_per_image_metrics(
annot, segmentations[segmentation]['orig'], False)
segmentations[segmentation]['mask'] = np.where(
segmentations[segmentation]['orig'] > 0, 255, 0)
for ensemble in ensembles.keys():
ensembles[ensemble]['orig'] = Ensemble(
segmentations, config.erosions, config.beta).ensemble(ensemble)
ensembles[ensemble]['results'] = comp.get_per_image_metrics(
annot, ensembles[ensemble]['orig'], False)
ensembles[ensemble]['mask'] = np.where(
ensembles[ensemble]['orig'] > 0, 255, 0)
for key in tables.keys():
results = {}
# for segmentation in segmentations.keys():
# results[segmentation] = segmentations[segmentation]['results'][key]
avg = statistics.mean([
segmentations[segmentation]['results'][key]
for segmentation in segmentations
])
results['unet'] = avg
for ensemble in ensembles.keys():
if ensemble != 'union':
results[ensemble] = ensembles[ensemble]['results'][key]
tables[key] = tables[key].append(results, ignore_index=True)
counter += 1
os.makedirs(os.path.join(root_dir, 'stats'), exist_ok=True)
output_charts(tables, list(methods.keys()),
os.path.join(root_dir, 'stats'), config)
def main():
print(args)
print("=> creating model '{}'".format(args.arch))
model = Ensemble()
model = torch.nn.DataParallel(model).cuda()
print(model)
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
test_data = datautil.SceneDataset(args.data,img_transform=
transforms.Compose([
transforms.Resize((args.img_size,args.img_size)),
transforms.ToTensor(),
normalize]))
test_loader = torch.utils.data.DataLoader(test_data,batch_size=args.batch_size,shuffle=False,num_workers=4,pin_memory=True)
checkpoint = torch.load(args.test_model)
model.load_state_dict(checkpoint['state_dict'])
#model.load_state_dict(checkpoint)
if os.path.isdir(args.data):
ret = test(test_loader,model)
imgs = [i[:-4] for i in os.listdir(args.data)]
with open('result3_.csv', 'w') as f:
'''
f.write(','.join(['FILE_ID','CATEGORY_ID'])+'\n')
f.write('\n'.join([','.join([str(a),str(b)]) for a,b in zip(imgs,ret)]))
'''
#FILE_ID,CATEGORY_ID0,CATEGORY_ID1,CATEGORY_ID2
f.write(','.join(['FILE_ID','CATEGORY_ID0','CATEGORY_ID1','CATEGORY_ID2'])+'\n')
f.write('\n'.join([','.join([str(a)]+[str(int(i)) for i in b]) for a,b in zip(imgs,ret)]))
else:
test_labeled(test_loader,model)
async def main():
this_dir = os.path.dirname(os.path.abspath(__file__))
input_path = os.path.join(this_dir, "input.txt")
with open(input_path) as f:
raw_code = f.readline()
e = Ensemble(raw_code)
await e.run()
def create_song(graph_attributes={
'graph_type': 'Small World',
'average_degree': 4,
'rewiring_prob': 0.3
},
number_players=20,
number_time_steps=300,
tempo=108,
player_attributes=None):
"""
arguments:
graph_type : 'Small World', 'Random', 'Configuration', 'Structured'
average_degree
number_of_players
rewiring_prob
number_time_steps
tempo
player_attributes: {
duration: (min_duration, max_duration)
note_change_choices: 'All', 'Neighbors of Neighbors'
harmonicity threshold: 'Fixed' or 'Moving Average'
fixed threshold
moving average threshold
susceptibility to influence
}
"""
graph_type = graph_attributes['graph_type']
#create player graph
if graph_type == 'Small World':
#assert rewiring_prob
G = nx.watts_strogatz_graph(number_players,
graph_attributes['average_degree'],
graph_attributes['rewiring_prob'])
elif graph_type == 'Random':
pass
elif graph_type == 'Structured':
pass
#add starting pitch to node
starting_pitches = {i: 'random' for i in range(len(G))}
nx.set_node_attributes(G, starting_pitches, 'starting_pitch')
#create ensembel object5
ensemble = Ensemble(G, player_attributes)
#evolve ensemble
ensemble.evolve(number_time_steps)
#show pitch history
pitch_history_data = ensemble.get_pitch_history_data()
harmonicity_data = ensemble.get_harmonicity_data()
#create file
filename = create_midi_file(ensemble, tempo)
create_data_file(filename.replace('.mid', '.txt'), pitch_history_data,
harmonicity_data)
return filename, pitch_history_data, harmonicity_data
def __init__(self, sourceFile, targetFile):
self.SDataBufferArr = None #2D array representation of self.SDataBuffer
self.SDataLabels = None
self.TDataBufferArr = None #2D array representation of self.TDataBuffer
self.TDataLabels = None
self.useKliepCVSigma = Properties.useKliepCVSigma
self.kliep = None
self.useSvmCVParams = Properties.useSvmCVParams
self.ensemble = Ensemble(Properties.ENSEMBLE_SIZE)
self.initialWindowSize = int(Properties.INITIAL_DATA_SIZE)
self.maxWindowSize = int(Properties.MAX_WINDOW_SIZE)
self.enableForceUpdate = int(Properties.enableForceUpdate)
self.forceUpdatePeriod = int(Properties.forceUpdatePeriod)
"""
- simulate source and target streams from corresponding files.
"""
print("Reading the Source Dataset")
self.source = Stream(sourceFile, Properties.INITIAL_DATA_SIZE)
print("Reading the Target Dataset")
self.target = Stream(targetFile, Properties.INITIAL_DATA_SIZE)
print("Finished Reading the Target Dataset")
Properties.MAXVAR = self.source.initialData.shape[0]
def worker(fold, n_users, n_items, dataset_dir):
traFilePath = dataset_dir + 'ratings__' + str(fold + 1) + '_tra.txt'
trasR = lil_matrix(
matBinarize(loadSparseR(n_users, n_items, traFilePath),
binarize_threshold))
print(
dataset_dir.split('/')[-2] + '@%d:' % (fold + 1), trasR.shape,
trasR.nnz, '%.2f' % (trasR.nnz / float(trasR.shape[0])))
tstFilePath = dataset_dir + 'ratings__' + str(fold + 1) + '_tst.txt'
tstsR = lil_matrix(
matBinarize(loadSparseR(n_users, n_items, tstFilePath),
binarize_threshold))
sampler = Sampler(trasR=trasR, batch_size=batch_size)
en = Ensemble(n_users, n_items, kensemble, topN, split_method,
eval_metrics, reg, n_factors, batch_size)
scores = en.train(fold + 1, trasR, tstsR, sampler)
print(
dataset_dir.split('/')[-2] + '@%d:' % (fold + 1),
','.join(['%s' % eval_metric for eval_metric in eval_metrics]) +
'@%d=' % (topN) + ','.join(['%.6f' % (score) for score in scores]))
en.close()
return scores
def reward_func(sigma_index_lst=[1, 2, 3],
default_n=20,
epoch_num=4,
epoch_min=100,
epoch_step=50):
'''
input
sigma_lst - The component index from the ssa gene for example the gen [0, 1, 0] -> sigma_lst=[1] #the index where gen=1
default_n - the window length for ssa - <= N /2 where N is the length of the time series - default 20
epoch_num - The number of submodel used
epoch_min - Min epoch of submodel
epoch_step - number of epoch difference bw 2 submodels
output
a tuple contain 2 value (nse_q, nse_h)
'''
K.clear_session()
with open('./settings/model/config.yaml', 'r') as f:
config = yaml.load(f, Loader=yaml.FullLoader)
# train
model = Ensemble(mode='train',
model_kind='rnn_cnn',
sigma_lst=sigma_index_lst,
default_n=default_n,
epoch_num=epoch_num,
epoch_min=epoch_min,
epoch_step=epoch_step,
**config)
model.train_model_outer()
# test
model = Ensemble(mode='test',
model_kind='rnn_cnn',
sigma_lst=sigma_index_lst,
default_n=default_n,
epoch_num=epoch_num,
epoch_min=epoch_min,
epoch_step=epoch_step,
**config)
model.train_model_outer()
model.retransform_prediction(mode='roll')
return model.evaluate_model(mode='roll')
def get_unique_model():
xg = xgb.XGBRegressor(n_estimators=200,
learning_rate=0.02,
gamma=0,
subsample=0.75,
colsample_bytree=1,
max_depth=6)
en = ElasticNet(l1_ratio=0.95, alpha=0.15, max_iter=50000)
ada = AdaBoostRegressor(learning_rate=0.01,
loss='square',
n_estimators=100)
lr = Ilbeom_Linear()
lst = [xg, en, ada, lr]
return Ensemble(lst)
def __init__(self, saved_model: str = None):
"""Create a new object.
Args:
- saved_model (str optional): load a pre-treined model if `saved_name` is not None
"""
super().__init__()
# Creating a XGBoost model for stacking
xgb_params = {}
xgb_params['learning_rate'] = 0.01
xgb_params['n_estimators'] = 750
xgb_params['max_depth'] = 6
xgb_params['colsample_bytree'] = 0.6
xgb_params['min_child_weight'] = 0.6
xgb_model = XGBClassifier(**xgb_params)
# Creating a random forest model for stacking
rf_params = {}
rf_params['n_estimators'] = 200
rf_params['max_depth'] = 6
rf_params['min_samples_split'] = 70
rf_params['min_samples_leaf'] = 30
rf_model = RandomForestClassifier(**rf_params)
# Creating a Logist Regression model to act as a stacker of other base models
log_model = LogisticRegression()
# Creating the stack
stack = Ensemble(n_splits=3,
stacker=log_model,
base_models=(rf_model, xgb_model))
# To use as a prefix of model and processed dataset
self.datetime_prefix = datetime.datetime.now().replace(
microsecond=0).isoformat().replace(':', '-')
# Loads a saved model or create a new one
if saved_model:
self.model_name = saved_model
else:
self.model_name = self.datetime_prefix + '_fraud_ensemble.bin'
# The final model
self.model = stack
print('Model: {}'.format(self.model_name))
def main():
#print(sys.argv)
test_set_path = sys.argv[1]
output_file_path = sys.argv[2]
X_test = dataset_manip.load_images(load_directory(test_set_path)) / 255
#model = Model(image_shape = (77, 71, 1), num_classes = 10, model_path = './model_files/model', batch_size = 512, first_run = False)
#dataset_manip.store_predictions(dataset_manip.get_filenames(test_set_path), model.predict(X_test), output_file_path)
ens = Ensemble(input_shape=(77, 71, 1),
num_classes=10,
num_models=11,
batch_size=512,
path='./ensemble_files',
load=True)
dataset_manip.store_predictions(dataset_manip.get_filenames(test_set_path),
ens.predict(X_test), output_file_path)
def main():
# Dataset path
dataset_name = ['credit_card_clients_balanced', 'credit_card_clients']
for data_name in dataset_name:
dataset_path = os.getcwd() + "\\dataset\\" + data_name + ".csv"
dataset = pd.read_csv(dataset_path, encoding='utf-8')
# Datasets columns
data_x = dataset[[
'X1', 'X2', 'X3', 'X4', 'X5', 'X6', 'X7', 'X8', 'X9', 'X10', 'X11',
'X12', 'X13', 'X14', 'X15', 'X16', 'X17', 'X18', 'X19', 'X20',
'X21', 'X22', 'X23'
]]
data_y = dataset['Y']
# Preprocessing data
min_max_scaler = preprocessing.MinMaxScaler()
X_normalized = min_max_scaler.fit_transform(data_x)
acc_rate = []
reject_rate = []
# Runs to test the model
for i in range(20):
print('---------------- Ensemble -----------------')
print('--- MLP - SVM - KNN - GMM - Naive Bayes ---')
print(i + 1, 'of 20 iterations')
X_train, X_test, y_train, y_test = train_test_split(X_normalized,
data_y,
test_size=0.2)
y_train = np.array(y_train)
y_test = np.array(y_test)
model = Ensemble()
model.train(X_train, y_train, gridSearch=False)
y_hat = model.predict(X_test)
error, reject = model.evaluate(y_hat, y_test)
acc_rate.append(1 - error)
reject_rate.append(reject)
graphics(acc_rate, reject_rate, data_name)
def __init__(self,
selected_algorithms='all',
selected_hyperparameters='default',
ensemble_size=3,
ensemble_method='Logit',
error_matrix_values='default',
verbose=True):
"""instantiates an AutoLearner object """
self.error_matrix = ErrorMatrix(selected_algorithms,
selected_hyperparameters,
ensemble_size, error_matrix_values,
verbose)
"""error matrix defined for specific dataset"""
self.ensemble = Ensemble(ensemble_size=ensemble_size,
ensemble_method=ensemble_method,
verbose=verbose)
"""instantiate empty ensemble object"""
def run(ncyc,
N=1,
lim=(20, 20),
T=300,
ensemble=None,
animation=False,
dframe=0.001):
#initialize system
time_total = 0 #total "time" of the system
time_pulse = 0 #time of the pulse
if ensemble == None:
ensemble = Ensemble(N, lim, T)
ensemble.Plot("Initial Configuration")
else:
ensemble.Plot("Initial Configuration")
#start simulation
y = [ensemble.Energy_Total()]
start_time = time.time()
#for i in trange(ncyc):
for i in range(ncyc):
if (i in list(range(0, ncyc, int(ncyc / 20)))):
print("{0} cycles: {1} s".format(i, time.time() - start_time))
ensemble.Cycle(time_pulse=time_pulse, time_total=time_total)
y.append(ensemble.Energy_Total())
time_total += dt
if time_pulse + dt > 120:
time_pulse += dt - 120
else:
time_pulse += dt
print("Elapsed time:", time.time() - start_time, "(s)")
print("Initial energy:", y[0], "(J)")
print("Final energy:", y[-1], "(J)")
print("Average energy:", ensemble.Average(), "(J)")
ensemble.Plot("Final Configuration")
x = range(0, len(y))
fig, ax = plt.subplots(figsize=(20, 10))
plt.plot(x, y)
plt.xlim([0, ncyc])
plt.ylim([min(y), max(y)])
plt.title("Total Energy vs. Cycle")
return ensemble
def main():
ATOM_NUM = 1
CYCLE_NUM = 50
paricles = [Particle() for i in range(ATOM_NUM)]
ensemble = Ensemble(paricles)
ensemble.positions += 2
print("初期位置\n", ensemble.positions)
print("初速度\n", ensemble.velocities)
myfield = Field(ensemble, dt=0.01)
for i in range(CYCLE_NUM):
myfield.update()
print("t:", myfield.dt * (i + 1))
print("x:", myfield.ensemble.positions)
print("v:", myfield.ensemble.velocities)
print()
def ensemble_test():
ATOM_NUM = 2
paricles = [Particle() for i in range(ATOM_NUM)]
ensemble = Ensemble(paricles)
print("位置の配列\n", ensemble.positions)
ensemble.positions += 1 #全体に+1
print("+1\n", ensemble.positions)
ensemble.positions *= 2 #全体に*2
print("*2\n", ensemble.positions)
ensemble.positions *= np.array([1, 2, 3]) #x*1, y*2, z*3
print("x*1, y*2, z*3\n", ensemble.positions)
ensemble.positions = np.ones((ensemble.N, 3)) * 100 #100にセット
print("=100\n", ensemble.positions)
print("速度の配列\n", ensemble.velocities)
def test(texts, classes, models, nn_params, folds=4):
'''
Check the performance on an SVM implementation,
given a list of texts and their classes (negative/neutral/positive)
Uses k-fold cross-validation (keeping in mind to divide the data
appropriately, depending on the class)
'''
classes = np.array(classes)
texts = np.array(texts)
wrongs = []
auc_sum = 0
for train, test in cross_validation.StratifiedKFold(classes, folds):
texts_train = texts[train]
classes_train = classes[train]
texts_test = texts[test]
classes_test = classes[test]
n = Ensemble(texts_train, classes_train, nn_params, models)
predictions = n.classify(texts_test)
predictions[predictions<0] = 0
auc = calculate_auc(classes_test, predictions)
print auc
auc_sum += auc
for i in range(len(texts_test)):
if abs(classes_test[i] - predictions[i]) > 0.5:
wrongs.append((classes_test[i], predictions[i], texts_test[i]))
'''
import csv
writer = open('wrongs.csv', 'w')
for w in wrongs:
writer.write('%s,%s,%s\n' % w)
writer.close()
'''
return auc_sum / folds
print('\n___PARTITIONS 2___')
partitions2 = np.transpose(partitions)
ensemble2 = Ensemble(partitions=partitions2, n_cluster=3, partitions_format='PE')
e2, ts2, pr2 = ensemble2.mcla(times=True, partial_results=True)
for t in ts2:
print(f'{t[0]}: {t[1]}s')
for r in pr2:
print(r[0])
print(r[1])
"""
print('\n___PARTITIONS 3___')
partitions3 = np.random.randint(8, size=(8, 100000))
ensemble3 = Ensemble(partitions=partitions3,
n_cluster=8,
partitions_format='PE')
e3, ts3, _pr3 = ensemble3.mcla(times=True)
for t in ts3:
print(f'{t[0]}: {t[1]}s')
"""
hypergraph4 = np.array([
[1, 1, 1, 0, 0, 0, 0],
[0, 0, 0, 1, 1, 0, 0],
[0, 0, 0, 0, 0, 1, 1],
[0, 0, 0, 0, 0, 1, 1],
[1, 1, 1, 0, 0, 0, 0],
[0, 0, 0, 1, 1, 0, 0],
[1, 1, 0, 0, 0, 0, 0],
[0, 0, 1, 1, 0, 0, 0],
[0, 0, 0, 0, 1, 1, 1],
def main():
input_dir = "/amit/kaggle/tgs"
output_dir = "/artifacts"
image_size_target = 128
batch_size = 32
epochs_to_train = 300
bce_loss_weight_gamma = 0.98
sgdr_min_lr = 0.0001 # 0.0001, 0.001
sgdr_max_lr = 0.001 # 0.001, 0.03
sgdr_cycle_epochs = 20
sgdr_cycle_epoch_prolongation = 3
sgdr_cycle_end_patience = 3
train_abort_epochs_without_improval = 30
ensemble_model_count = 3
swa_epoch_to_start = 30
model_dir = sys.argv[1] if len(sys.argv) > 1 else None
train_data = TrainData(input_dir)
train_set = TrainDataset(train_data.train_set_df, image_size_target, augment=True)
train_set_data_loader = DataLoader(train_set, batch_size=batch_size, shuffle=True, num_workers=8)
val_set = TrainDataset(train_data.val_set_df, image_size_target, augment=False)
val_set_data_loader = DataLoader(val_set, batch_size=batch_size, shuffle=False, num_workers=2)
if model_dir:
model = create_model(pretrained=False).to(device)
model.load_state_dict(torch.load("{}/model.pth".format(model_dir), map_location=device))
else:
model = create_model(pretrained=True).to(device)
torch.save(model.state_dict(), "{}/model.pth".format(output_dir))
swa_model = create_model(pretrained=False).to(device)
print("train_set_samples: %d, val_set_samples: %d" % (len(train_set), len(val_set)))
global_val_precision_best_avg = float("-inf")
global_swa_val_precision_best_avg = float("-inf")
sgdr_cycle_val_precision_best_avg = float("-inf")
epoch_iterations = len(train_set) // batch_size
# optimizer = optim.SGD(model.parameters(), lr=sgdr_max_lr, weight_decay=0, momentum=0.9, nesterov=True)
optimizer = optim.Adam(model.parameters(), lr=sgdr_max_lr)
lr_scheduler = CosineAnnealingLR(optimizer, T_max=sgdr_cycle_epochs, eta_min=sgdr_min_lr)
optim_summary_writer = SummaryWriter(log_dir="{}/logs/optim".format(output_dir))
train_summary_writer = SummaryWriter(log_dir="{}/logs/train".format(output_dir))
val_summary_writer = SummaryWriter(log_dir="{}/logs/val".format(output_dir))
swa_val_summary_writer = SummaryWriter(log_dir="{}/logs/swa_val".format(output_dir))
sgdr_iterations = 0
sgdr_reset_count = 0
batch_count = 0
epoch_of_last_improval = 0
sgdr_next_cycle_end_epoch = sgdr_cycle_epochs + sgdr_cycle_epoch_prolongation
swa_update_count = 0
ensemble_model_index = 0
for model_file_path in glob.glob("{}/model-*.pth".format(output_dir)):
model_file_name = os.path.basename(model_file_path)
model_index = int(model_file_name.replace("model-", "").replace(".pth", ""))
ensemble_model_index = max(ensemble_model_index, model_index + 1)
print('{"chart": "best_val_precision", "axis": "epoch"}')
print('{"chart": "val_precision", "axis": "epoch"}')
print('{"chart": "val_loss", "axis": "epoch"}')
print('{"chart": "sgdr_reset", "axis": "epoch"}')
print('{"chart": "precision", "axis": "epoch"}')
print('{"chart": "loss", "axis": "epoch"}')
print('{"chart": "swa_val_precision", "axis": "epoch"}')
print('{"chart": "swa_val_loss", "axis": "epoch"}')
train_start_time = time.time()
criterion = nn.BCEWithLogitsLoss()
for epoch in range(epochs_to_train):
epoch_start_time = time.time()
model.train()
train_loss_sum = 0.0
train_precision_sum = 0.0
train_step_count = 0
for batch in train_set_data_loader:
images, masks, mask_weights = \
batch[0].to(device, non_blocking=True), \
batch[1].to(device, non_blocking=True), \
batch[2].to(device, non_blocking=True)
lr_scheduler.step(epoch=min(sgdr_cycle_epochs, sgdr_iterations / epoch_iterations))
optimizer.zero_grad()
prediction_logits = model(images)
predictions = torch.sigmoid(prediction_logits)
criterion.weight = mask_weights
loss = criterion(prediction_logits, masks)
loss.backward()
optimizer.step()
train_loss_sum += loss.item()
train_precision_sum += np.mean(precision_batch(predictions, masks))
sgdr_iterations += 1
train_step_count += 1
batch_count += 1
optim_summary_writer.add_scalar("lr", get_learning_rate(optimizer), batch_count + 1)
train_loss_avg = train_loss_sum / train_step_count
train_precision_avg = train_precision_sum / train_step_count
val_loss_avg, val_precision_avg = evaluate(model, val_set_data_loader, criterion)
model_improved_within_sgdr_cycle = val_precision_avg > sgdr_cycle_val_precision_best_avg
if model_improved_within_sgdr_cycle:
torch.save(model.state_dict(), "{}/model-{}.pth".format(output_dir, ensemble_model_index))
sgdr_cycle_val_precision_best_avg = val_precision_avg
model_improved = val_precision_avg > global_val_precision_best_avg
ckpt_saved = False
if model_improved:
torch.save(model.state_dict(), "{}/model.pth".format(output_dir))
global_val_precision_best_avg = val_precision_avg
ckpt_saved = True
swa_model_improved = False
if epoch + 1 >= swa_epoch_to_start:
if model_improved_within_sgdr_cycle:
swa_update_count += 1
moving_average(swa_model, model, 1.0 / swa_update_count)
bn_update(train_set_data_loader, swa_model)
swa_model_improved = val_precision_avg > global_swa_val_precision_best_avg
if swa_model_improved:
torch.save(swa_model.state_dict(), "{}/swa_model.pth".format(output_dir))
global_swa_val_precision_best_avg = val_precision_avg
if model_improved or swa_model_improved:
epoch_of_last_improval = epoch
sgdr_reset = False
if (epoch + 1 >= sgdr_next_cycle_end_epoch) and (epoch - epoch_of_last_improval >= sgdr_cycle_end_patience):
sgdr_iterations = 0
sgdr_next_cycle_end_epoch = epoch + 1 + sgdr_cycle_epochs + sgdr_cycle_epoch_prolongation
ensemble_model_index += 1
sgdr_cycle_val_precision_best_avg = float("-inf")
sgdr_reset_count += 1
sgdr_reset = True
swa_val_loss_avg, swa_val_precision_avg = evaluate(swa_model, val_set_data_loader, criterion)
optim_summary_writer.add_scalar("sgdr_reset", sgdr_reset_count, epoch + 1)
train_summary_writer.add_scalar("loss", train_loss_avg, epoch + 1)
train_summary_writer.add_scalar("precision", train_precision_avg, epoch + 1)
val_summary_writer.add_scalar("loss", val_loss_avg, epoch + 1)
val_summary_writer.add_scalar("precision", val_precision_avg, epoch + 1)
swa_val_summary_writer.add_scalar("loss", swa_val_loss_avg, epoch + 1)
swa_val_summary_writer.add_scalar("precision", swa_val_precision_avg, epoch + 1)
epoch_end_time = time.time()
epoch_duration_time = epoch_end_time - epoch_start_time
print(
"[%03d/%03d] %ds, lr: %.6f, loss: %.3f, val_loss: %.3f|%.3f, prec: %.3f, val_prec: %.3f|%.3f, ckpt: %d, rst: %d" % (
epoch + 1,
epochs_to_train,
epoch_duration_time,
get_learning_rate(optimizer),
train_loss_avg,
val_loss_avg,
swa_val_loss_avg,
train_precision_avg,
val_precision_avg,
swa_val_precision_avg,
int(ckpt_saved),
int(sgdr_reset)),
flush=True)
print('{"chart": "best_val_precision", "x": %d, "y": %.3f}' % (epoch + 1, global_val_precision_best_avg))
print('{"chart": "val_precision", "x": %d, "y": %.3f}' % (epoch + 1, val_precision_avg))
print('{"chart": "val_loss", "x": %d, "y": %.3f}' % (epoch + 1, val_loss_avg))
print('{"chart": "sgdr_reset", "x": %d, "y": %.3f}' % (epoch + 1, sgdr_reset_count))
print('{"chart": "precision", "x": %d, "y": %.3f}' % (epoch + 1, train_precision_avg))
print('{"chart": "loss", "x": %d, "y": %.3f}' % (epoch + 1, train_loss_avg))
print('{"chart": "swa_val_precision", "x": %d, "y": %.3f}' % (epoch + 1, swa_val_precision_avg))
print('{"chart": "swa_val_loss", "x": %d, "y": %.3f}' % (epoch + 1, swa_val_loss_avg))
if sgdr_reset and sgdr_reset_count >= ensemble_model_count and epoch - epoch_of_last_improval >= train_abort_epochs_without_improval:
print("early abort")
break
optim_summary_writer.close()
train_summary_writer.close()
val_summary_writer.close()
train_end_time = time.time()
print()
print("Train time: %s" % str(datetime.timedelta(seconds=train_end_time - train_start_time)))
eval_start_time = time.time()
print()
print("evaluation of the training model")
model.load_state_dict(torch.load("{}/model.pth".format(output_dir), map_location=device))
analyze(Ensemble([model]), train_data.val_set_df, use_tta=False)
analyze(Ensemble([model]), train_data.val_set_df, use_tta=True)
score_to_model = {}
ensemble_model_candidates = glob.glob("{}/model-*.pth".format(output_dir))
ensemble_model_candidates.append("{}/swa_model.pth".format(output_dir))
for model_file_path in ensemble_model_candidates:
model_file_name = os.path.basename(model_file_path)
m = create_model(pretrained=False).to(device)
m.load_state_dict(torch.load(model_file_path, map_location=device))
val_loss_avg, val_precision_avg = evaluate(m, val_set_data_loader, criterion)
print("ensemble '%s': val_loss=%.3f, val_precision=%.3f" % (model_file_name, val_loss_avg, val_precision_avg))
if len(score_to_model) < ensemble_model_count or min(score_to_model.keys()) < val_precision_avg:
del score_to_model[min(score_to_model.keys())]
score_to_model[val_precision_avg] = m
ensemble_models = list(score_to_model.values())
for ensemble_model in ensemble_models:
val_loss_avg, val_precision_avg = evaluate(ensemble_model, val_set_data_loader, criterion)
print("ensemble: val_loss=%.3f, val_precision=%.3f" % (val_loss_avg, val_precision_avg))
model = Ensemble(ensemble_models)
mask_threshold_global, mask_threshold_per_cc = analyze(model, train_data.val_set_df, use_tta=True)
eval_end_time = time.time()
print()
print("Eval time: %s" % str(datetime.timedelta(seconds=eval_end_time - eval_start_time)))
print()
print("submission preparation")
submission_start_time = time.time()
test_data = TestData(input_dir)
calculate_predictions(test_data.df, model, use_tta=True)
calculate_prediction_masks(test_data.df, mask_threshold_global)
print()
print(test_data.df.groupby("predictions_cc").agg({"predictions_cc": "count"}))
write_submission(test_data.df, "prediction_masks", "{}/{}".format(output_dir, "submission.csv"))
write_submission(test_data.df, "prediction_masks_best", "{}/{}".format(output_dir, "submission_best.csv"))
submission_end_time = time.time()
print()
print("Submission time: %s" % str(datetime.timedelta(seconds=submission_end_time - submission_start_time)))
@child('e3')
def i2(x):
return x**2
@child('e3')
def i3(x, y):
return x**3 + y
if __name__ == '__main__':
# create our first ensemble and give it a name
e1 = Ensemble('e1')
# create a second ensemble
e2 = Ensemble('e2')
# you may use the ensembles as long as you specify which model you use
print(e1(child='f', x=2))
print(e1(child='g', y=3))
print(e2(child='f', x=2))
# try to use model `g` but it's not in ensemble `e2`
try:
print(e2(child='g', y=3))
except ValueError:
pass
# try to use model `h` but it's not decorated with @model
model = resnet101()
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, 1)
if latest_model_path != "":
model.load_state_dict(torch.load(latest_model_path))
model.cuda()
# Set parameters for model
criterion = Loss(Wt1, Wt0)
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer,
mode='min',
patience=1,
verbose=True)
# Train model
model = train_model(model, criterion, optimizer, dataloaders,
scheduler, dataset_sizes, epochs - current_epoch,
costs, accs, num_ID, model_type)
# For testing ensemble model
else:
model = Ensemble("models/best_model_dense_4.pth",
"models/best_model_res_12.pth",
"models/best_model_vgg_19.pth")
model.cuda()
criterion = Loss(Wt1, Wt0)
test_acc, test_loss = test_ensemble_mean(model, criterion, dataloaders,
dataset_sizes)
# classes.append(int(row[0]))
# #results = n.classify(texts)
# #results[results<0] = 0
# #print calculate_auc(classes, results)
# r1 = np.array(m1.classify(texts))
# print calculate_auc(classes, r1)
# r2 = np.array(m2.classify(texts))
# print calculate_auc(classes, r2)
# r = (1.2*r1 + 0.8*r2) / 2
# r[r>1] = 1
# r[r<0] = 0
# print calculate_auc(classes, r)
#print TestSVM.test_model(texts, classes, models[-1])
#print TestSVM.test(texts, classes, models, nn_params)
n = Ensemble(texts, classes, nn_params, models)
end = time.time()
# print "training time="
# print end-start
start = time.time()
# evaluate the classfier on verification dataset
texts = []
inp = raw_input()
while inp:
texts.append(inp.decode('utf8'))
inp = raw_input()
results = n.classify(texts)
positives = sum(1 for label in labels if label)
predicted_positives = sum(1 for pred in preds if pred)
true_positives = sum(1 for label, pred in zip(labels, preds)
if label and pred)
return 100.0 * true_positives / predicted_positives, 100.0 * true_positives / positives
def evaluate(model):
def wrapper(dataset):
preds = [model(x=x) for x, _ in dataset]
precision, recall = get_results(dataset, preds)
return {
'precision': f'{precision:.1f}%',
'recall': f'{recall:.1f}%',
}
return wrapper
if __name__ == '__main__':
e = Ensemble('ensemble', children=[model1, model2], mode='all')
results = Ensemble('results', children=[model1, model2, e])
results.decorate_children(evaluate)
print(results)
pprint(results(dataset=get_dataset()))
"""
{'ensemble': {'precision': '100.0%', 'recall': '100.0%'},
'model1': {'precision': '18.2%', 'recall': '100.0%'},
'model2': {'precision': '30.0%', 'recall': '100.0%'}}
"""
print(
"[epoch: {:d}] avg train_loss: {:.3f} eval ll: {:.3f} ({:.1f}s)"
.format(epoch,
sum(losses) / len(losses), eval_ll,
time.time() - tic))
print("running test code")
name = "sample_" + str(epoch) + ".txt"
test_code(model, name=name)
print("ran test code")
if __name__ == '__main__':
import args
from model import Model
from ensemble import Ensemble
if method == 'ensemble':
model = Ensemble(vectors).to(device)
else:
model = Model(vectors).to(device)
train(model)
# import dill
# with open('model.p', 'rb') as h:
# model = dill.load(h)
# visualize_attn(model)
experiment_path = "./experiments/ensemble/"
data_path = "./datasets/paper_ultimate_conv/"
cf_data = imp.load_source("cf_data",
data_path + "cfg_testing_data_creation.py")
nets = []
net_paths = glob.glob(experiment_path + "*/")
# Load the networks
for net_path in net_paths:
net = NetworkConvDropout()
net.init(29, 29, 13, 134, 135)
net.load_parameters(open_h5file(net_path + "net.net"))
n_out = net.n_out
nets.append(net)
ensemble_net = Ensemble(nets)
# Load the scaler
scaler = pickle.load(open(net_paths[0] + "s.scaler", "rb"))
# Files on which to evaluate the network
file_list = list_miccai_files(**{
"mode": "folder",
"path": "./datasets/miccai/2/"
})
n_files = len(file_list)
# Options for the generation of the dataset
# The generation/evaluation of the dataset has to be split into batches as a whole brain does not fit into memory
batch_size = 100000
select_region = SelectWholeBrain()
def __init__(self):
self.ensemble = Ensemble(instruments)
from saveobject import save_obj
N = 100
steps = 1000
repeat = 30
res = 0.01
b1 = 0
b2 = 3
B = np.arange(b1, b2, res)
B = B[B != 0]
B = 1 / B
M = np.array([0])
ensemble = Ensemble(N, B, M, steps, repeat, False)
ensemble.getStats()
beta = ensemble.beta
mu = ensemble.mu
stats = ensemble.stats
save_obj(stats, "stats8")
# keys = ["energy","magnetization","population","entropy"]
# def calcStats(size,beta,mu,steps,times):
# global keys
# stats = {}
# arr = {}
# for key in keys:
'max_depth': 6,
'n_estimators': 1000,
'learning_rate': 0.025,
'subsample': 0.9
}
models = {
"LGB-1": LGBMClassifier(**lgb_params),
"XGB-1": XGBClassifier(**xgb_params),
"LGB-2": LGBMClassifier(**lgb_params2),
#"LGB-3": LGBMClassifier(**lgb_params3),
"XGB-2": XGBClassifier(**xgb_params2),
#"CAT": CatBoostClassifier(**cat_params),
#"GBM": GradientBoostingClassifier(**gb_params),
#"RF": RandomForestClassifier(**rf_params),
#"ET": ExtraTreesClassifier(**et_params),
#"ABC": AdaBoostClassifier(n_estimators=100),
}
start = time.time()
stack = Ensemble(4,
models.values(),
stacker=SGDClassifier(loss="log", max_iter=1000))
y_pred = stack.fit_predict(X, y, X_test)
print("Finished ensembling in %.2f seconds" % (time.time() - start))
sub = pd.DataFrame()
sub['id'] = id_test
sub['target'] = y_pred
sub.to_csv("%s.csv" % ("-".join(models.keys())), index=False)
def main():
system = Ensemble(3, 3, lim=(30, 30), T=300)
run(5000, ensemble=system)
