def _validate():
"""検証＆閾値決定。"""
logger = tk.log.get(__name__)
X, y = _data.load_train_data()
pred = predict_all('val', X, use_cache=True) # TODO: 仮！, use_cache=True
# 閾値の調整
pred_bin = joblib.load(CACHE_DIR / 'val' / 'bin_nas.pkl')
pred_reg = joblib.load(CACHE_DIR / 'val' / 'reg_nas.pkl')
threshold_X, threshold_y = _ath.create_data(y, pred, pred_bin, pred_reg)
split_seed = int((MODELS_DIR / 'split_seed.txt').read_text())
thresholds = np.empty((len(y), ))
for cv_index in range(CV_COUNT):
ti, vi = tk.ml.cv_indices(X,
y,
cv_count=CV_COUNT,
cv_index=cv_index,
split_seed=split_seed,
stratify=False)
ath_estimator = _ath.create_estimator(threshold_X[ti], threshold_y[ti])
thresholds[vi] = ath_estimator.predict(threshold_X[vi])
# CVで決めた閾値でevaluate
_evaluation.log_evaluation(y,
pred,
print_fn=logger.info,
threshold=thresholds)
# 全体で学習しなおして保存
ath_estimator = _ath.create_estimator(threshold_X, threshold_y)
joblib.dump(ath_estimator, MODELS_DIR / 'ath_estimator.pkl')
# 最後にインチキ閾値でevaluate
thresholds = ath_estimator.predict(threshold_X)
_evaluation.log_evaluation(y,
pred,
print_fn=logger.info,
threshold=thresholds)
def _validate():
"""検証＆閾値決定。"""
logger = tk.log.get(__name__)
X, y = _data.load_train_data()
pred = predict_all('val', X)
threshold = _evaluation.log_evaluation(y, pred, print_fn=logger.info, search_th=True)
(MODELS_DIR / 'threshold.txt').write_text(str(threshold))
def _train(args):
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()
X = _mf.get_meta_features('val', X)
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, _ = _create_network(input_dims=X.shape[-1])
gen = tk.generator.Generator()
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.Resize((101, 101), with_output=True))
model = tk.dl.models.Model(network, gen, batch_size=BATCH_SIZE)
model.compile(sgd_lr=0.01 / 128, loss=tk.dl.losses.lovasz_hinge_elup1, metrics=[tk.dl.metrics.binary_accuracy], clipnorm=10.0)
model.fit(
X_train, y_train, validation_data=(X_val, y_val),
epochs=EPOCHS,
reduce_lr_epoch_rates=(0.5, 0.75, 0.875), mixup=False, lr_warmup=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 _validate():
"""検証＆閾値決定。"""
logger = tk.log.get(__name__)
X, y = _data.load_train_data()
y = np.mean(y, axis=(1, 2, 3))
pred = predict_all('val', X)
tk.ml.print_regression_metrics(y, pred, print_fn=logger.info)
def _validate():
"""検証＆閾値決定。"""
logger = tk.log.get(__name__)
X, y = _data.load_train_data()
y = np.max(y > 0.5, axis=(1, 2, 3)).astype(np.uint8) # 0 or 1
pred = predict_all('val', X)
tk.ml.print_classification_metrics(y, pred, print_fn=logger.info)
def validate():
train_set = _data.load_train_data()
folds = tk.validation.split(train_set,
nfold,
stratify=True,
split_seed=split_seed)
model = create_model().load(models_dir)
pred = model.predict_oof(train_set, folds)
_data.save_oofp(models_dir, train_set, pred)
def train():
train_set = _data.load_train_data()
folds = tk.validation.split(train_set,
nfold,
stratify=True,
split_seed=split_seed)
model = create_model()
evals = model.cv(train_set, folds)
tk.notifications.post_evals(evals)
def _validate():
"""検証＆閾値決定。"""
logger = tk.log.get(__name__)
X, y = _data.load_train_data()
split_seed = int((MODELS_DIR / 'split_seed.txt').read_text())
ti, vi = tk.ml.cv_indices(X, y, cv_count=CV_COUNT, cv_index=0, split_seed=split_seed, stratify=False)
X_val, y_val = X[vi], y[vi]
pred_val = predict_all('val', X_val)
threshold = _evaluation.log_evaluation(y_val, pred_val, print_fn=logger.info, search_th=True)
(MODELS_DIR / 'threshold.txt').write_text(str(threshold))
def _validate():
"""検証。"""
logger = tk.log.get(__name__)
X_train, y_train = _data.load_train_data()
pred = predict_all('val', X_train)
threshold = _evaluation.log_evaluation(y_train,
pred,
print_fn=logger.info,
search_th=True)
MODELS_DIR.mkdir(parents=True, exist_ok=True)
(MODELS_DIR / 'threshold.txt').write_text(str(threshold))
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 load_train_data():
dataset = _data.load_train_data()
return dataset
def load_test_data():
mf = load_mf("test")
dataset = _data.load_train_data()
for mf_i in mf:
dataset.data = mf_i
yield dataset
def load_train_data():
dataset = _data.load_train_data()
dataset.data = load_mf("train")
return dataset
