config_name = args.config.split("/")[-1].replace(".yml", "")
expdir = Path(f"out/{config_name}")
if not expdir.exists():
print(f"You need to train {config_name} first!")
sys.exit(1)
submission_file_dir = expdir / "submission"
submission_file_dir.mkdir(parents=True, exist_ok=True)
logger = utils.get_logger(expdir / "tta.log")
# environment
utils.set_seed(global_params["seed"])
device = training.get_device(global_params["device"])
# data
tp, fp, train_all, test_all, train_audio, test_audio = datasets.get_metadata(
config)
submission = pd.read_csv(config["data"]["sample_submission_path"])
# labels
labels = []
duration = config["dataset"]["valid"]["params"]["duration"]
for _, sample in tp.iterrows():
t_min = sample["t_min"]
t_max = sample["t_max"]
flac_id = sample["recording_id"]
call_duration = t_max - t_min
relative_offset = (duration - call_duration) / 2
offset = min(max(0, t_min - relative_offset), 60 - duration)
tail = offset + duration
global_params = config["globals"]
# logging
config_name = args.config.split("/")[-1].replace(".yml", "")
logdir = Path(f"out/{config_name}")
logdir.mkdir(exist_ok=True, parents=True)
logger = utils.get_logger(logdir / "output.log")
# environment
utils.set_seed(global_params["seed"])
device = training.get_device(global_params["device"])
# data
tp, fp, train_all, _, train_audio, _ = datasets.get_metadata(config)
# validation
splitter = training.get_split(config)
##################################################
# Main Loop #
##################################################
for i, (trn_idx, val_idx) in enumerate(splitter.split(train_all)):
if i not in global_params["folds"]:
continue
logger.info("=" * 20)
logger.info(f"Fold {i}")
logger.info("=" * 20)
trn_df = train_all.loc[trn_idx, :].reset_index(drop=True)
val_df = train_all.loc[val_idx, :].reset_index(drop=True)
config_name = args.config.split("/")[-1].replace(".yml", "")
expdir = Path(f"out/{config_name}")
expdir.mkdir(exist_ok=True, parents=True)
logger = utils.get_logger(expdir / "ensemble.log")
oofs = []
submissions = []
names = []
for result_dict in config["results"]:
oofs.append(pd.read_csv(result_dict["oof"]))
submissions.append(pd.read_csv(result_dict["submission"]))
names.append(result_dict["name"])
tp, _, _, _, _, _ = datasets.get_metadata(config)
indices = tp[["index"]]
for i in range(len(oofs)):
oofs[i] = indices.merge(oofs[i], on="index", how="left")
labels = []
for _, sample in tp.iterrows():
t_min = sample["t_min"]
t_max = sample["t_max"]
flac_id = sample["recording_id"]
call_duration = t_max - t_min
relative_offset = (10 - call_duration) / 2
offset = min(max(0, t_min - relative_offset), 60 - 10)
tail = offset + 10