nshtrainer 1.0.0b43__tar.gz → 1.0.0b44__tar.gz

{nshtrainer-1.0.0b43 → nshtrainer-1.0.0b44}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: nshtrainer
-Version: 1.0.0b43
+Version: 1.0.0b44
 Summary: 
 Author: Nima Shoghi
 Author-email: nimashoghi@gmail.com

{nshtrainer-1.0.0b43 → nshtrainer-1.0.0b44}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "nshtrainer"
-version = "1.0.0-beta43"
+version = "1.0.0-beta44"
 description = ""
 authors = ["Nima Shoghi <nimashoghi@gmail.com>"]
 readme = "README.md"

{nshtrainer-1.0.0b43 → nshtrainer-1.0.0b44}/src/nshtrainer/callbacks/__init__.py

@@ -40,6 +40,10 @@ from .log_epoch import LogEpochCallback as LogEpochCallback
 from .log_epoch import LogEpochCallbackConfig as LogEpochCallbackConfig
 from .lr_monitor import LearningRateMonitor as LearningRateMonitor
 from .lr_monitor import LearningRateMonitorConfig as LearningRateMonitorConfig
+from .metric_validation import MetricValidationCallback as MetricValidationCallback
+from .metric_validation import (
+    MetricValidationCallbackConfig as MetricValidationCallbackConfig,
+)
 from .norm_logging import NormLoggingCallback as NormLoggingCallback
 from .norm_logging import NormLoggingCallbackConfig as NormLoggingCallbackConfig
 from .print_table import PrintTableMetricsCallback as PrintTableMetricsCallback

nshtrainer-1.0.0b44/src/nshtrainer/callbacks/metric_validation.py

@@ -0,0 +1,75 @@
+from __future__ import annotations
+
+import logging
+from typing import Literal
+
+from lightning.pytorch.utilities.exceptions import MisconfigurationException
+from typing_extensions import final, override, assert_never
+
+from .._callback import NTCallbackBase
+from ..metrics import MetricConfig
+from .base import CallbackConfigBase, callback_registry
+
+log = logging.getLogger(__name__)
+
+
+@final
+@callback_registry.register
+class MetricValidationCallbackConfig(CallbackConfigBase):
+    name: Literal["metric_validation"] = "metric_validation"
+
+    error_behavior: Literal["raise", "warn"] = "raise"
+    """
+    Behavior when an error occurs during validation:
+    - "raise": Raise an error and stop the training.
+    - "warn": Log a warning and continue the training.
+    """
+
+    validate_default_metric: bool = True
+    """Whether to validate the default metric from the root config."""
+
+    metrics: list[MetricConfig] = []
+    """List of metrics to validate."""
+
+    @override
+    def create_callbacks(self, trainer_config):
+        metrics = self.metrics.copy()
+        if (
+            self.validate_default_metric
+            and (default_metric := trainer_config.primary_metric) is not None
+        ):
+            metrics.append(default_metric)
+
+        yield MetricValidationCallback(self, metrics)
+
+
+class MetricValidationCallback(NTCallbackBase):
+    def __init__(
+        self, config: MetricValidationCallbackConfig, metrics: list[MetricConfig]
+    ):
+        super().__init__()
+
+        self.config = config
+        self.metrics = metrics
+
+    @override
+    def on_sanity_check_end(self, trainer, pl_module):
+        super().on_sanity_check_end(trainer, pl_module)
+
+        log.debug("Validating metrics...")
+        logged_metrics = set(trainer.logged_metrics.keys())
+        for metric in self.metrics:
+            if metric.validation_monitor in logged_metrics:
+                continue
+
+            match self.config.error_behavior:
+                case "raise":
+                    raise MisconfigurationException(
+                        f"Metric '{metric.validation_monitor}' not found in logged metrics."
+                    )
+                case "warn":
+                    log.warning(
+                        f"Metric '{metric.validation_monitor}' not found in logged metrics."
+                    )
+                case _:
+                    assert_never(self.config.error_behavior)

{nshtrainer-1.0.0b43 → nshtrainer-1.0.0b44}/src/nshtrainer/configs/__init__.py

@@ -39,6 +39,9 @@ from nshtrainer.callbacks import (
 )
 from nshtrainer.callbacks import LearningRateMonitorConfig as LearningRateMonitorConfig
 from nshtrainer.callbacks import LogEpochCallbackConfig as LogEpochCallbackConfig
+from nshtrainer.callbacks import (
+    MetricValidationCallbackConfig as MetricValidationCallbackConfig,
+)
 from nshtrainer.callbacks import NormLoggingCallbackConfig as NormLoggingCallbackConfig
 from nshtrainer.callbacks import (
     OnExceptionCheckpointCallbackConfig as OnExceptionCheckpointCallbackConfig,
@@ -287,6 +290,7 @@ __all__ = [
     "MPIEnvironmentPlugin",
     "MPSAcceleratorConfig",
     "MetricConfig",
+    "MetricValidationCallbackConfig",
     "MishNonlinearityConfig",
     "MixedPrecisionPluginConfig",
     "NonlinearityConfig",

{nshtrainer-1.0.0b43 → nshtrainer-1.0.0b44}/src/nshtrainer/configs/callbacks/__init__.py

@@ -28,6 +28,9 @@ from nshtrainer.callbacks import (
 )
 from nshtrainer.callbacks import LearningRateMonitorConfig as LearningRateMonitorConfig
 from nshtrainer.callbacks import LogEpochCallbackConfig as LogEpochCallbackConfig
+from nshtrainer.callbacks import (
+    MetricValidationCallbackConfig as MetricValidationCallbackConfig,
+)
 from nshtrainer.callbacks import NormLoggingCallbackConfig as NormLoggingCallbackConfig
 from nshtrainer.callbacks import (
     OnExceptionCheckpointCallbackConfig as OnExceptionCheckpointCallbackConfig,
@@ -65,6 +68,7 @@ from . import finite_checks as finite_checks
 from . import gradient_skipping as gradient_skipping
 from . import log_epoch as log_epoch
 from . import lr_monitor as lr_monitor
+from . import metric_validation as metric_validation
 from . import norm_logging as norm_logging
 from . import print_table as print_table
 from . import rlp_sanity_checks as rlp_sanity_checks
@@ -91,6 +95,7 @@ __all__ = [
     "LearningRateMonitorConfig",
     "LogEpochCallbackConfig",
     "MetricConfig",
+    "MetricValidationCallbackConfig",
     "NormLoggingCallbackConfig",
     "OnExceptionCheckpointCallbackConfig",
     "PrintTableMetricsCallbackConfig",
@@ -110,6 +115,7 @@ __all__ = [
     "gradient_skipping",
     "log_epoch",
     "lr_monitor",
+    "metric_validation",
     "norm_logging",
     "print_table",
     "rlp_sanity_checks",

nshtrainer-1.0.0b44/src/nshtrainer/configs/callbacks/metric_validation/__init__.py

@@ -0,0 +1,21 @@
+from __future__ import annotations
+
+__codegen__ = True
+
+from nshtrainer.callbacks.metric_validation import (
+    CallbackConfigBase as CallbackConfigBase,
+)
+from nshtrainer.callbacks.metric_validation import MetricConfig as MetricConfig
+from nshtrainer.callbacks.metric_validation import (
+    MetricValidationCallbackConfig as MetricValidationCallbackConfig,
+)
+from nshtrainer.callbacks.metric_validation import (
+    callback_registry as callback_registry,
+)
+
+__all__ = [
+    "CallbackConfigBase",
+    "MetricConfig",
+    "MetricValidationCallbackConfig",
+    "callback_registry",
+]

{nshtrainer-1.0.0b43 → nshtrainer-1.0.0b44}/src/nshtrainer/configs/trainer/__init__.py

@@ -38,6 +38,9 @@ from nshtrainer.trainer._config import LogEpochCallbackConfig as LogEpochCallbac
 from nshtrainer.trainer._config import LoggerConfig as LoggerConfig
 from nshtrainer.trainer._config import LoggerConfigBase as LoggerConfigBase
 from nshtrainer.trainer._config import MetricConfig as MetricConfig
+from nshtrainer.trainer._config import (
+    MetricValidationCallbackConfig as MetricValidationCallbackConfig,
+)
 from nshtrainer.trainer._config import (
     NormLoggingCallbackConfig as NormLoggingCallbackConfig,
 )
@@ -164,6 +167,7 @@ __all__ = [
     "MPIEnvironmentPlugin",
     "MPSAcceleratorConfig",
     "MetricConfig",
+    "MetricValidationCallbackConfig",
     "MixedPrecisionPluginConfig",
     "NormLoggingCallbackConfig",
     "OnExceptionCheckpointCallbackConfig",

{nshtrainer-1.0.0b43 → nshtrainer-1.0.0b44}/src/nshtrainer/configs/trainer/_config/__init__.py

@@ -34,6 +34,9 @@ from nshtrainer.trainer._config import LogEpochCallbackConfig as LogEpochCallbac
 from nshtrainer.trainer._config import LoggerConfig as LoggerConfig
 from nshtrainer.trainer._config import LoggerConfigBase as LoggerConfigBase
 from nshtrainer.trainer._config import MetricConfig as MetricConfig
+from nshtrainer.trainer._config import (
+    MetricValidationCallbackConfig as MetricValidationCallbackConfig,
+)
 from nshtrainer.trainer._config import (
     NormLoggingCallbackConfig as NormLoggingCallbackConfig,
 )
@@ -77,6 +80,7 @@ __all__ = [
     "LoggerConfig",
     "LoggerConfigBase",
     "MetricConfig",
+    "MetricValidationCallbackConfig",
     "NormLoggingCallbackConfig",
     "OnExceptionCheckpointCallbackConfig",
     "PluginConfig",

{nshtrainer-1.0.0b43 → nshtrainer-1.0.0b44}/src/nshtrainer/trainer/_config.py

@@ -40,6 +40,7 @@ from ..callbacks.base import CallbackConfigBase
 from ..callbacks.debug_flag import DebugFlagCallbackConfig
 from ..callbacks.log_epoch import LogEpochCallbackConfig
 from ..callbacks.lr_monitor import LearningRateMonitorConfig
+from ..callbacks.metric_validation import MetricValidationCallbackConfig
 from ..callbacks.rlp_sanity_checks import RLPSanityChecksCallbackConfig
 from ..callbacks.shared_parameters import SharedParametersCallbackConfig
 from ..loggers import (
@@ -697,6 +698,10 @@ class TrainerConfig(C.Config):
     - The trainer is running in fast_dev_run mode.
     - The trainer is running a sanity check (which happens before starting the training routine).
     """
+    auto_validate_metrics: MetricValidationCallbackConfig | None = (
+        MetricValidationCallbackConfig()
+    )
+    """If enabled, will automatically validate the metrics before starting the training routine."""
 
     lightning_kwargs: LightningTrainerKwargs = LightningTrainerKwargs()
     """
@@ -768,6 +773,7 @@ class TrainerConfig(C.Config):
         yield self.shared_parameters
         yield self.reduce_lr_on_plateau_sanity_checking
         yield self.auto_set_debug_flag
+        yield self.auto_validate_metrics
         yield from self.callbacks
 
     def _nshtrainer_all_logger_configs(self) -> Iterable[LoggerConfigBase | None]: