fix(ckpt): correct safetensors path for decimal-suffix stems; return actual save path

CHANGELOG.rst

@@ -72,6 +72,17 @@ Fixed
 - Fix bugs in utility function `torch_ecg.utils.make_serializable`: the previous implementation
   does not drop some types of unserializable items correctly. Two additional parameters
   `drop_unserializable` and `drop_paths` are added.
+- Fix ``CkptMixin.save()`` silently truncating checkpoint filenames that contain
+  decimal values (e.g. ``…metric_0.91``): ``pathlib.Path.with_suffix(".safetensors")``
+  treated ``.91`` as the existing suffix and replaced it, producing
+  ``…metric_0.safetensors`` instead of the correct ``…metric_0.91.safetensors``.
+  The method now appends ``.safetensors`` for paths without a recognised
+  extension, and returns the final ``Path`` used so callers can track it.
+- Fix ``BaseTrainer`` checkpoint cleanup permanently failing: ``saved_models``
+  stored the raw stem path while the actual file on disk had a ``.safetensors``
+  suffix, causing every ``os.remove()`` call to raise ``FileNotFoundError``.
+  The trainer now stores the path returned by ``save_checkpoint()``, and handles
+  both single-file (``os.remove``) and directory (``shutil.rmtree``) checkpoints.
 
 Security
 ~~~~~~~~

benchmarks/train_crnn_cinc2020/scoring_metrics.py

@@ -2,19 +2,10 @@
 metrics from the official scoring repository
 """
 
-from numbers import Real
 from typing import List, Sequence, Tuple
 
 import numpy as np
 
-try:
-    import torch_ecg  # noqa: F401
-except ModuleNotFoundError:
-    import sys
-    from pathlib import Path
-
-    sys.path.insert(0, str(Path(__file__).absolute().parents[2]))
-
 from torch_ecg.databases.aux_data.cinc2020_aux_data import load_weights
 
 __all__ = [
@@ -168,7 +159,7 @@ def compute_f_measure(labels: np.ndarray, outputs: np.ndarray) -> float:
 
 
 # Compute F-beta and G-beta measures from the unofficial phase of the Challenge.
-def compute_beta_measures(labels: np.ndarray, outputs: np.ndarray, beta: Real) -> Tuple[float, float]:
+def compute_beta_measures(labels: np.ndarray, outputs: np.ndarray, beta: float) -> Tuple[float, float]:
     """checked,"""
     num_recordings, num_classes = np.shape(labels)
 

benchmarks/train_crnn_cinc2021/scoring_metrics.py

@@ -2,19 +2,10 @@
 metrics from the official scoring repository
 """
 
-from numbers import Real
 from typing import List, Sequence, Tuple, Union
 
 import numpy as np
 
-try:
-    import torch_ecg  # noqa: F401
-except ModuleNotFoundError:
-    import sys
-    from pathlib import Path
-
-    sys.path.insert(0, str(Path(__file__).absolute().parents[2]))
-
 from torch_ecg.databases.aux_data.cinc2021_aux_data import load_weights
 
 __all__ = [
@@ -233,7 +224,7 @@ def compute_f_measure(labels: np.ndarray, outputs: np.ndarray) -> Tuple[float, n
 
 
 # Compute F-beta and G-beta measures from the unofficial phase of the Challenge.
-def compute_beta_measures(labels: np.ndarray, outputs: np.ndarray, beta: Real) -> Tuple[float, float]:
+def compute_beta_measures(labels: np.ndarray, outputs: np.ndarray, beta: float) -> Tuple[float, float]:
     """checked,"""
     num_recordings, num_classes = np.shape(labels)
 

benchmarks/train_crnn_cinc2021/special_detectors.py

@@ -17,7 +17,6 @@
 """
 
 from itertools import repeat
-from numbers import Real
 from typing import Any, Optional, Sequence
 
 import numpy as np
@@ -54,7 +53,7 @@
 
 def special_detectors(
     raw_sig: np.ndarray,
-    fs: Real,
+    fs: int,
     sig_fmt: str = "channel_first",
     leads: Sequence[str] = Standard12Leads,
     verbose: int = 0,
@@ -66,7 +65,7 @@ def special_detectors(
     ----------
     raw_sig: ndarray,
         the raw multi-lead ecg signal, with units in mV
-    fs: real number,
+    fs: int,
         sampling frequency of `sig`
     sig_fmt: str, default "channel_first",
         format of the multi-lead ecg signal,
@@ -133,19 +132,19 @@ def special_detectors(
 
 def pacing_rhythm_detector(
     raw_sig: np.ndarray,
-    fs: Real,
+    fs: int,
     sig_fmt: str = "channel_first",
     leads: Sequence[str] = Standard12Leads,
     ret_prob: bool = True,
     verbose: int = 0,
-) -> Real:
+) -> float:
     """to be improved (fine-tuning hyper-parameters in cfg.py),
 
     Parameters
     ----------
     raw_sig: ndarray,
         the raw multi-lead ecg signal, with units in mV
-    fs: real number,
+    fs: int,
         sampling frequency of `sig`
     sig_fmt: str, default "channel_first",
         format of the multi-lead ecg signal,
@@ -238,7 +237,7 @@ def pacing_rhythm_detector(
 def electrical_axis_detector(
     filtered_sig: np.ndarray,
     rpeaks: np.ndarray,
-    fs: Real,
+    fs: int,
     sig_fmt: str = "channel_first",
     leads: Sequence[str] = Standard12Leads,
     method: Optional[str] = None,
@@ -255,7 +254,7 @@ def electrical_axis_detector(
         the filtered multi-lead ecg signal, with units in mV
     rpeaks: ndarray,
         array of indices of the R peaks
-    fs: real number,
+    fs: int,
         sampling frequency of `sig`
     sig_fmt: str, default "channel_first",
         format of the multi-lead ecg signal,
@@ -375,8 +374,8 @@ def electrical_axis_detector(
 
 def brady_tachy_detector(
     rpeaks: np.ndarray,
-    fs: Real,
-    normal_rr_range: Optional[Sequence[Real]] = None,
+    fs: int,
+    normal_rr_range: Optional[Sequence[int]] = None,
     verbose: int = 0,
 ) -> str:
     """to be improved (fine-tuning hyper-parameters in cfg.py),
@@ -391,7 +390,7 @@ def brady_tachy_detector(
     ----------
     rpeaks: ndarray,
         array of indices of the R peaks
-    fs: real number,
+    fs: int,
         sampling frequency of the ecg signal
     normal_rr_range: sequence of int, optional,
         the range of normal rr interval, with units in ms;
@@ -438,7 +437,7 @@ def brady_tachy_detector(
 def LQRSV_detector(
     filtered_sig: np.ndarray,
     rpeaks: np.ndarray,
-    fs: Real,
+    fs: int,
     sig_fmt: str = "channel_first",
     leads: Sequence[str] = Standard12Leads,
     verbose: int = 0,
@@ -451,7 +450,7 @@ def LQRSV_detector(
         the filtered multi-lead ecg signal, with units in mV
     rpeaks: ndarray,
         array of indices of the R peaks
-    fs: real number,
+    fs: int,
         sampling frequency of the ecg signal
     sig_fmt: str, default "channel_first",
         format of the 12 lead ecg signal,
@@ -511,7 +510,7 @@ def LQRSV_detector(
 def LQRSV_detector_backup(
     filtered_sig: np.ndarray,
     rpeaks: np.ndarray,
-    fs: Real,
+    fs: int,
     sig_fmt: str = "channel_first",
     leads: Sequence[str] = Standard12Leads,
     verbose: int = 0,
@@ -524,7 +523,7 @@ def LQRSV_detector_backup(
         the filtered 12-lead ecg signal, with units in mV
     rpeaks: ndarray,
         array of indices of the R peaks
-    fs: real number,
+    fs: int,
         sampling frequency of the ecg signal
     sig_fmt: str, default "channel_first",
         format of the 12 lead ecg signal,
@@ -608,7 +607,7 @@ def LQRSV_detector_backup(
 def PRWP_detector(
     filtered_sig: np.ndarray,
     rpeaks: np.ndarray,
-    fs: Real,
+    fs: int,
     sig_fmt: str = "channel_first",
     leads: Sequence[str] = Standard12Leads,
     verbose: int = 0,
@@ -621,7 +620,7 @@ def PRWP_detector(
         the filtered multi-lead ecg signal, with units in mV
     rpeaks: ndarray,
         array of indices of the R peaks
-    fs: real number,
+    fs: int,
         sampling frequency of the ecg signal
     sig_fmt: str, default "channel_first",
         format of the 12 lead ecg signal,

benchmarks/train_crnn_cinc2023/data_reader.py

@@ -4,7 +4,6 @@
 import re
 import warnings
 from ast import literal_eval
-from numbers import Real
 from pathlib import Path
 from typing import Any, Dict, List, Literal, Optional, Sequence, Tuple, Union
 
@@ -599,9 +598,9 @@ def load_data(
         return_channels: bool = False,
     ) -> Union[
         np.ndarray,
-        Tuple[np.ndarray, Real],
+        Tuple[np.ndarray, int],
         Tuple[np.ndarray, List[str]],
-        Tuple[np.ndarray, Real, List[str]],
+        Tuple[np.ndarray, int, List[str]],
     ]:
         """Load EEG data from the record.
 
@@ -640,7 +639,7 @@ def load_data(
         -------
         data : numpy.ndarray
             The loaded EEG data.
-        data_fs : numbers.Real, optional
+        data_fs : int, optional
             Sampling frequency of the output signal.
             Returned if `return_fs` is True.
         data_channels : list of str, optional
@@ -721,7 +720,7 @@ def load_bipolar_data(
         units: Literal["mV", "uV", "muV", "μV", None] = "uV",
         fs: Optional[int] = None,
         return_fs: bool = False,
-    ) -> Union[np.ndarray, Tuple[np.ndarray, Real]]:
+    ) -> Union[np.ndarray, Tuple[np.ndarray, int]]:
         """Load bipolar EEG data from the record.
 
         Bipolar EEG is the difference between two channels.
@@ -754,7 +753,7 @@ def load_bipolar_data(
         -------
         data : numpy.ndarray
             The loaded EEG data.
-        data_fs : numbers.Real, optional
+        data_fs : int, optional
             Sampling frequency of the output signal.
 
         """

benchmarks/train_hybrid_cpsc2020/metrics.py

@@ -1,18 +1,8 @@
 """ """
 
-from numbers import Real
 from typing import List, Sequence, Tuple, Union
 
 import numpy as np
-
-try:
-    import torch_ecg  # noqa: F401
-except ModuleNotFoundError:
-    import sys
-    from pathlib import Path
-
-    sys.path.insert(0, str(Path(__file__).absolute().parents[2]))
-
 from cfg import BaseCfg
 
 from torch_ecg.cfg import CFG
@@ -323,7 +313,7 @@ def compute_f_measure(labels: np.ndarray, outputs: np.ndarray) -> float:
 
 
 # Compute F-beta and G-beta measures from the unofficial phase of the Challenge.
-def compute_beta_measures(labels: np.ndarray, outputs: np.ndarray, beta: Real) -> Tuple[float, float]:
+def compute_beta_measures(labels: np.ndarray, outputs: np.ndarray, beta: float) -> Tuple[float, float]:
     """checked,"""
     num_recordings, num_classes = np.shape(labels)
 

benchmarks/train_hybrid_cpsc2020/signal_processing/ecg_denoise.py

@@ -15,19 +15,10 @@
 to add
 """
 
-from numbers import Real
 from typing import List
 
 import numpy as np
 
-try:
-    import torch_ecg  # noqa: F401
-except ModuleNotFoundError:
-    import sys
-    from pathlib import Path
-
-    sys.path.insert(0, str(Path(__file__).absolute().parents[3]))
-
 from torch_ecg.cfg import CFG
 from torch_ecg.utils.utils_data import mask_to_intervals
 
@@ -36,7 +27,7 @@
 ]
 
 
-def ecg_denoise(filtered_sig: np.ndarray, fs: Real, config: CFG) -> List[List[int]]:
+def ecg_denoise(filtered_sig: np.ndarray, fs: int, config: CFG) -> List[List[int]]:
     """
 
     a naive function removing non-ECG segments (flat and motion artefact)
@@ -45,7 +36,7 @@ def ecg_denoise(filtered_sig: np.ndarray, fs: Real, config: CFG) -> List[List[in
     ----------
     filtered_sig: ndarray,
         1d filtered (typically bandpassed) ECG signal,
-    fs: real number,
+    fs: int,
         sampling frequency of `filtered_sig`
     config: dict,
         configs of relavant parameters, like window, step, etc.

benchmarks/train_hybrid_cpsc2020/signal_processing/ecg_preproc.py

@@ -19,7 +19,6 @@
 import os
 import time
 from copy import deepcopy
-from numbers import Real
 from typing import Dict, Optional
 
 import numpy as np
@@ -64,14 +63,14 @@
 ]
 
 
-def preprocess_signal(raw_sig: np.ndarray, fs: Real, config: Optional[CFG] = None) -> Dict[str, np.ndarray]:
+def preprocess_signal(raw_sig: np.ndarray, fs: int, config: Optional[CFG] = None) -> Dict[str, np.ndarray]:
     """
 
     Parameters
     ----------
     raw_sig: ndarray,
         the raw ecg signal
-    fs: real number,
+    fs: int,
         sampling frequency of `raw_sig`
     config: dict, optional,
         extra process configuration,
@@ -135,7 +134,7 @@ def preprocess_signal(raw_sig: np.ndarray, fs: Real, config: Optional[CFG] = Non
 
 def parallel_preprocess_signal(
     raw_sig: np.ndarray,
-    fs: Real,
+    fs: int,
     config: Optional[CFG] = None,
     save_dir: Optional[str] = None,
     save_fmt: str = "npy",
@@ -147,7 +146,7 @@ def parallel_preprocess_signal(
     ----------
     raw_sig: ndarray,
         the raw ecg signal
-    fs: real number,
+    fs: int,
         sampling frequency of `raw_sig`
     config: dict, optional,
         extra process configuration,