def _predict():
"""予測。"""
logger = tk.log.get(__name__)
X_test = _data.load_test_data()
threshold = float((MODELS_DIR / 'threshold.txt').read_text())
logger.info(f'threshold = {threshold:.3f}')
pred = predict_all('test', X_test) > threshold
_data.save_submission(MODELS_DIR / 'submission.csv', pred)
def predict():
logger.info(f"source_models = {source_models}")
test_set = _data.load_test_data()
pred = np.average(
[_load_pred(n) for n in model_names],
weights=model_weights,
axis=0,
)
_data.save_prediction(models_dir, test_set, pred)
def _predict():
"""予測。"""
X_test = _data.load_test_data()
pred = predict_all('test', X_test)
_data.save_submission(MODELS_DIR / 'submission.csv', pred > 0.50)
_data.save_submission(MODELS_DIR / 'submission_0.40.csv', pred > 0.40)
_data.save_submission(MODELS_DIR / 'submission_0.45.csv', pred > 0.45)
_data.save_submission(MODELS_DIR / 'submission_0.55.csv', pred > 0.55)
_data.save_submission(MODELS_DIR / 'submission_0.60.csv', pred > 0.60)
def _train(args, fine=False):
logger = tk.log.get(__name__)
logger.info(f'args: {args}')
split_seed = int(MODEL_NAME.encode('utf-8').hex(), 16) % 10000000
MODELS_DIR.mkdir(parents=True, exist_ok=True)
(MODELS_DIR / 'split_seed.txt').write_text(str(split_seed))
X, y = _data.load_train_data()
ti, vi = tk.ml.cv_indices(X, y, cv_count=CV_COUNT, cv_index=args.cv_index, split_seed=split_seed, stratify=False)
(X_train, y_train), (X_val, y_val) = (X[ti], y[ti]), (X[vi], y[vi])
logger.info(f'cv_index={args.cv_index}: train={len(y_train)} val={len(y_val)}')
network, lr_multipliers = _create_network()
gen = tk.generator.Generator()
if fine:
pseudo_size = len(y_train) // 2
X_train = np.array(list(X_train) + [None] * pseudo_size)
y_train = np.array(list(y_train) + [None] * pseudo_size)
X_test = _data.load_test_data()
_, pi = tk.ml.cv_indices(X_test, np.zeros((len(X_test),)), cv_count=CV_COUNT, cv_index=args.cv_index, split_seed=split_seed, stratify=False)
#pred_test = predict_all('test', None, use_cache=True)[(args.cv_index + 1) % CV_COUNT] # cross-pseudo-labeling
import stack_res
pred_test = stack_res.predict_all('test', None, use_cache=True)[(args.cv_index + 1) % CV_COUNT] # cross-pseudo-labeling
gen.add(tk.generator.RandomPickData(X_test[pi], pred_test[pi]))
gen.add(tk.image.RandomFlipLR(probability=0.5, with_output=True))
gen.add(tk.image.Padding(probability=1, with_output=True))
gen.add(tk.image.RandomRotate(probability=0.25, with_output=True))
gen.add(tk.image.RandomCrop(probability=1, with_output=True))
gen.add(tk.image.RandomAugmentors([
tk.image.RandomBlur(probability=0.125),
tk.image.RandomUnsharpMask(probability=0.125),
tk.image.RandomBrightness(probability=0.25),
tk.image.RandomContrast(probability=0.25),
], probability=0.125))
gen.add(tk.image.Resize((101, 101), with_output=True))
model = tk.dl.models.Model(network, gen, batch_size=BATCH_SIZE)
if fine:
model.load_weights(MODELS_DIR / f'model.fold{args.cv_index}.h5')
model.compile(sgd_lr=0.001 / 128 if fine else 0.1 / 128, loss=tk.dl.losses.lovasz_hinge_elup1,
metrics=[tk.dl.metrics.binary_accuracy], lr_multipliers=lr_multipliers, clipnorm=10.0)
model.fit(
X_train, y_train, validation_data=(X_val, y_val),
epochs=EPOCHS // 3 if fine else EPOCHS,
cosine_annealing=True, mixup=False)
model.save(MODELS_DIR / f'model.fold{args.cv_index}.h5', include_optimizer=False)
if tk.dl.hvd.is_master():
_evaluation.log_evaluation(y_val, model.predict(X_val))
def predict_all(data_name, X, use_cache=False):
"""予測。"""
cache_path = CACHE_DIR / data_name / f'{MODEL_NAME}.pkl'
if use_cache and cache_path.is_file():
return joblib.load(cache_path)
if data_name == 'val':
X_list, vi_list = [], []
split_seed = int((MODELS_DIR / 'split_seed.txt').read_text())
for cv_index in range(CV_COUNT):
_, vi = tk.ml.cv_indices(X,
None,
cv_count=CV_COUNT,
cv_index=cv_index,
split_seed=split_seed,
stratify=False)
X_list.append(X[vi])
vi_list.append(vi)
else:
X = _data.load_test_data()
X_list = [X] * CV_COUNT
gen = tk.generator.SimpleGenerator()
model = tk.dl.models.Model.load(MODELS_DIR / f'model.fold0.h5',
gen,
batch_size=BATCH_SIZE,
multi_gpu=True)
pred_list = []
for cv_index in tk.tqdm(range(CV_COUNT), desc='predict'):
if cv_index != 0:
model.load_weights(MODELS_DIR / f'model.fold{cv_index}.h5')
X_t = X_list[cv_index]
pred = _evaluation.predict_tta(model, X_t, mode='bin')
pred_list.append(pred)
if data_name == 'val':
pred = np.empty((len(X), 1), dtype=np.float32)
for vi, p in zip(vi_list, pred_list):
pred[vi] = p
else:
pred = pred_list
cache_path.parent.mkdir(parents=True, exist_ok=True)
joblib.dump(pred, cache_path, compress=3)
return pred
def load_test_data():
dataset = _data.load_test_data()
return dataset
def _predict():
"""予測。"""
X_test = _data.load_test_data()
pred_list = predict_all('test', X_test)
pred = np.mean(pred_list, axis=0) > 0.5
_data.save_submission(MODELS_DIR / 'submission.csv', pred)
def _predict():
"""予測。"""
X_test = _data.load_test_data()
predict_all('test', X_test)
def predict():
test_set = _data.load_test_data()
model = create_model().load(models_dir)
pred_list = model.predict_all(test_set)
pred = np.mean(pred_list, axis=0)
_data.save_prediction(models_dir, test_set, pred)
