nshtrainer 1.0.0b47__tar.gz → 1.0.0b50__tar.gz

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: nshtrainer
-Version: 1.0.0b47
+Version: 1.0.0b50
 Summary: 
 Author: Nima Shoghi
 Author-email: nimashoghi@gmail.com

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "nshtrainer"
-version = "1.0.0-beta47"
+version = "1.0.0-beta50"
 description = ""
 authors = ["Nima Shoghi <nimashoghi@gmail.com>"]
 readme = "README.md"

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/src/nshtrainer/callbacks/checkpoint/best_checkpoint.py

@@ -51,7 +51,7 @@ class BestCheckpointCallbackConfig(BaseCheckpointCallbackConfig):
 class BestCheckpointCallback(CheckpointBase[BestCheckpointCallbackConfig]):
     @property
     def _metric_name_normalized(self):
-        return self.metric.name.replace("/", "_").replace(" ", "_").replace(".", "_")
+        return self.metric.monitor.replace("/", "_").replace(" ", "_").replace(".", "_")
 
     @override
     def __init__(
@@ -69,12 +69,12 @@ class BestCheckpointCallback(CheckpointBase[BestCheckpointCallbackConfig]):
 
     @override
     def default_filename(self):
-        return f"epoch{{epoch}}-step{{step}}-{self._metric_name_normalized}{{{self.metric.validation_monitor}}}"
+        return f"epoch{{epoch}}-step{{step}}-{self._metric_name_normalized}{{{self.metric.monitor}}}"
 
     @override
     def topk_sort_key(self, metadata: CheckpointMetadata):
         return metadata.metrics.get(
-            self.metric.validation_monitor,
+            self.metric.monitor,
             float("-inf" if self.metric.mode == "max" else "inf"),
         )
 

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/src/nshtrainer/callbacks/early_stopping.py

@@ -68,7 +68,7 @@ class EarlyStoppingCallback(_EarlyStopping):
         del config, metric
 
         super().__init__(
-            monitor=self.metric.validation_monitor,
+            monitor=self.metric.monitor,
             mode=self.metric.mode,
             patience=self.config.patience,
             min_delta=self.config.min_delta,

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/src/nshtrainer/callbacks/metric_validation.py

@@ -55,14 +55,14 @@ class MetricValidationCallback(Callback):
         self.metrics = metrics
 
     def _check_metrics(self, trainer: Trainer):
-        metric_names = ", ".join(metric.validation_monitor for metric in self.metrics)
+        metric_names = ", ".join(metric.monitor for metric in self.metrics)
         log.info(f"Validating metrics: {metric_names}...")
         logged_metrics = set(trainer.logged_metrics.keys())
 
         invalid_metrics: list[str] = []
         for metric in self.metrics:
-            if metric.validation_monitor not in logged_metrics:
-                invalid_metrics.append(metric.validation_monitor)
+            if metric.monitor not in logged_metrics:
+                invalid_metrics.append(metric.monitor)
 
         if invalid_metrics:
             msg = (

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/src/nshtrainer/callbacks/rlp_sanity_checks.py

@@ -171,7 +171,7 @@ class CustomRLPImplementation(Protocol):
     __reduce_lr_on_plateau__: bool
 
 
-class _RLPSanityCheckModuleMixin(LightningModule):
+class RLPSanityCheckModuleMixin(LightningModule):
     def reduce_lr_on_plateau_config(
         self,
         lr_scheduler: LRSchedulerTypeUnion | LRSchedulerConfigType,

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/src/nshtrainer/data/datamodule.py

@@ -11,13 +11,13 @@ from lightning.pytorch import LightningDataModule
 from typing_extensions import Never, TypeVar, deprecated, override
 
 from ..model.mixins.callback import CallbackRegistrarModuleMixin
-from ..model.mixins.debug import _DebugModuleMixin
+from ..model.mixins.debug import DebugModuleMixin
 
 THparams = TypeVar("THparams", bound=C.Config, infer_variance=True)
 
 
 class LightningDataModuleBase(
-    _DebugModuleMixin,
+    DebugModuleMixin,
     CallbackRegistrarModuleMixin,
     LightningDataModule,
     ABC,

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/src/nshtrainer/loggers/__init__.py

@@ -2,7 +2,6 @@ from __future__ import annotations
 
 from typing import Annotated
 
-import nshconfig as C
 from typing_extensions import TypeAliasType
 
 from .actsave import ActSaveLoggerConfig as ActSaveLoggerConfig

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/src/nshtrainer/lr_scheduler/reduce_lr_on_plateau.py

@@ -7,6 +7,7 @@ from torch.optim.lr_scheduler import ReduceLROnPlateau
 from typing_extensions import final, override
 
 from ..metrics._config import MetricConfig
+from ..util.config import EpochsConfig
 from .base import LRSchedulerConfigBase, LRSchedulerMetadata, lr_scheduler_registry
 
 
@@ -21,13 +22,13 @@ class ReduceLROnPlateauConfig(LRSchedulerConfigBase):
     """Metric to monitor.
     If not provided, the primary metric of the runner will be used."""
 
-    patience: int
+    patience: int | EpochsConfig
     r"""Number of epochs with no improvement after which learning rate will be reduced."""
 
     factor: float
     r"""Factor by which the learning rate will be reduced. new_lr = lr * factor."""
 
-    cooldown: int = 0
+    cooldown: int | EpochsConfig = 0
     r"""Number of epochs to wait before resuming normal operation after lr has been reduced."""
 
     min_lr: float | list[float] = 0.0
@@ -49,28 +50,34 @@ class ReduceLROnPlateauConfig(LRSchedulerConfigBase):
         if (metric := self.metric) is None:
             from ..trainer import Trainer
 
-            assert isinstance(
-                trainer := lightning_module.trainer, Trainer
-            ), "The trainer must be a `nshtrainer.Trainer` instance."
+            assert isinstance(trainer := lightning_module.trainer, Trainer), (
+                "The trainer must be a `nshtrainer.Trainer` instance."
+            )
 
-            assert (
-                metric := trainer.hparams.primary_metric
-            ) is not None, "Primary metric must be provided if metric is not specified."
+            assert (metric := trainer.hparams.primary_metric) is not None, (
+                "Primary metric must be provided if metric is not specified."
+            )
+
+        if isinstance(patience := self.patience, EpochsConfig):
+            patience = int(patience.value)
+
+        if isinstance(cooldown := self.cooldown, EpochsConfig):
+            cooldown = int(cooldown.value)
 
         lr_scheduler = ReduceLROnPlateau(
             optimizer,
             mode=metric.mode,
             factor=self.factor,
-            patience=self.patience,
+            patience=patience,
             threshold=self.threshold,
             threshold_mode=self.threshold_mode,
-            cooldown=self.cooldown,
+            cooldown=cooldown,
             min_lr=self.min_lr,
             eps=self.eps,
         )
         return {
             "scheduler": lr_scheduler,
-            "monitor": metric.validation_monitor,
+            "monitor": metric.monitor,
         }
 
     @override

nshtrainer-1.0.0b50/src/nshtrainer/metrics/_config.py

@@ -0,0 +1,25 @@
+from __future__ import annotations
+
+import builtins
+from typing import Any, Literal
+
+import nshconfig as C
+
+
+class MetricConfig(C.Config):
+    monitor: str
+    """The name of the metric to monitor."""
+
+    mode: Literal["min", "max"]
+    """
+    The mode of the primary metric:
+    - "min" for metrics that should be minimized (e.g., loss)
+    - "max" for metrics that should be maximized (e.g., accuracy)
+    """
+
+    @property
+    def best(self):
+        return builtins.min if self.mode == "min" else builtins.max
+
+    def is_better(self, a: Any, b: Any):
+        return self.best(a, b) == a

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/src/nshtrainer/model/base.py

@@ -15,9 +15,9 @@ from lightning.pytorch.utilities.model_helpers import is_overridden
 from lightning.pytorch.utilities.rank_zero import rank_zero_warn
 from typing_extensions import Never, TypeVar, deprecated, override
 
-from ..callbacks.rlp_sanity_checks import _RLPSanityCheckModuleMixin
+from ..callbacks.rlp_sanity_checks import RLPSanityCheckModuleMixin
 from .mixins.callback import CallbackModuleMixin
-from .mixins.debug import _DebugModuleMixin
+from .mixins.debug import DebugModuleMixin
 from .mixins.logger import LoggerLightningModuleMixin
 
 log = logging.getLogger(__name__)
@@ -54,8 +54,8 @@ VALID_REDUCE_OPS = (
 
 
 class LightningModuleBase(
-    _DebugModuleMixin,
-    _RLPSanityCheckModuleMixin,
+    DebugModuleMixin,
+    RLPSanityCheckModuleMixin,
     LoggerLightningModuleMixin,
     CallbackModuleMixin,
     LightningModule,

{nshtrainer-1.0.0b47 → nshtrainer-1.0.0b50}/src/nshtrainer/model/mixins/debug.py

@@ -28,7 +28,7 @@ def _trainer(module: Any):
     return trainer
 
 
-class _DebugModuleMixin:
+class DebugModuleMixin:
     @property
     def nshtrainer_or_none(self):
         return _trainer(self)

nshtrainer-1.0.0b50/src/nshtrainer/model/mixins/logger.py

@@ -0,0 +1,275 @@
+from __future__ import annotations
+
+import dataclasses
+from collections import deque
+from collections.abc import Callable, Generator, Mapping
+from contextlib import contextmanager
+from typing import Any, ClassVar
+
+import torchmetrics
+from lightning.pytorch import LightningModule
+from lightning.pytorch.utilities.types import _METRIC
+from lightning_utilities.core.rank_zero import rank_zero_warn
+from typing_extensions import override
+
+from ...util.typing_utils import mixin_base_type
+
+
+@dataclasses.dataclass(frozen=True, kw_only=True)
+class _LogContextKwargs:
+    __ignore_fields__: ClassVar[set[str]] = {"prefix", "disabled"}
+
+    prefix: str | None = None
+    disabled: bool | None = None
+    prog_bar: bool | None = None
+    logger: bool | None = None
+    on_step: bool | None = None
+    on_epoch: bool | None = None
+    reduce_fx: str | Callable | None = None
+    enable_graph: bool | None = None
+    sync_dist: bool | None = None
+    sync_dist_group: Any | None = None
+    add_dataloader_idx: bool | None = None
+    batch_size: int | None = None
+    rank_zero_only: bool | None = None
+
+    def to_dict(self):
+        d = dataclasses.asdict(self)
+        for field in self.__ignore_fields__:
+            d.pop(field, None)
+
+        # Pop all None values
+        for k in list(d.keys()):
+            if d[k] is None:
+                d.pop(k)
+
+        return d
+
+
+class LoggerLightningModuleMixin(mixin_base_type(LightningModule)):
+    @override
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        self._logger_prefix_stack = deque[_LogContextKwargs]()
+
+    @property
+    def logging_enabled(self) -> bool:
+        # Logging is disabled in barebones mode.
+        if (trainer := self._trainer) is not None and trainer.barebones:
+            # Warn the user once that logging is disabled in barebones mode.
+            if not hasattr(self, "_barebones_logging_warned"):
+                rank_zero_warn(
+                    "Logging is disabled in barebones mode. "
+                    "This is to reduce the overhead of logging in barebones mode. "
+                    "If you want to enable logging, set `barebones=False` in the Trainer.",
+                )
+                self._barebones_logging_warned = True
+            return False
+
+        # If no loggers are registered, then logging is disabled.
+        if not self.logger:
+            return False
+
+        # Check if the topmost non-null context is disabled
+        for context in reversed(self._logger_prefix_stack):
+            if context.disabled is not None:
+                return not context.disabled
+
+        # Otherwise, logging is enabled.
+        return True
+
+    @contextmanager
+    def log_context(
+        self,
+        prefix: str | None = None,
+        disabled: bool | None = None,
+        prog_bar: bool | None = None,
+        logger: bool | None = None,
+        on_step: bool | None = None,
+        on_epoch: bool | None = None,
+        reduce_fx: str | Callable | None = None,
+        enable_graph: bool | None = None,
+        sync_dist: bool | None = None,
+        sync_dist_group: Any | None = None,
+        add_dataloader_idx: bool | None = None,
+        batch_size: int | None = None,
+        rank_zero_only: bool | None = None,
+    ) -> Generator[None, None, None]:
+        self._logger_prefix_stack.append(
+            _LogContextKwargs(
+                prefix=prefix,
+                disabled=disabled,
+                prog_bar=prog_bar,
+                logger=logger,
+                on_step=on_step,
+                on_epoch=on_epoch,
+                reduce_fx=reduce_fx,
+                enable_graph=enable_graph,
+                sync_dist=sync_dist,
+                sync_dist_group=sync_dist_group,
+                add_dataloader_idx=add_dataloader_idx,
+                batch_size=batch_size,
+                rank_zero_only=rank_zero_only,
+            )
+        )
+        try:
+            yield
+        finally:
+            _ = self._logger_prefix_stack.pop()
+
+    def _make_prefix_and_kwargs_dict(self, kwargs: _LogContextKwargs):
+        prefix = "".join(c.prefix for c in self._logger_prefix_stack if c.prefix)
+
+        fn_kwargs: dict[str, Any] = {}
+        for c in self._logger_prefix_stack:
+            fn_kwargs.update(c.to_dict())
+
+        fn_kwargs.update(kwargs.to_dict())
+        return prefix, fn_kwargs
+
+    @override
+    def log(
+        self,
+        name: str,
+        value: _METRIC,
+        prog_bar: bool | None = None,
+        logger: bool | None = None,
+        on_step: bool | None = None,
+        on_epoch: bool | None = None,
+        reduce_fx: str | Callable | None = None,
+        enable_graph: bool | None = None,
+        sync_dist: bool | None = None,
+        sync_dist_group: Any | None = None,
+        add_dataloader_idx: bool | None = None,
+        batch_size: int | None = None,
+        metric_attribute: str | None = None,
+        rank_zero_only: bool | None = None,
+    ) -> None:
+        """Log a key, value pair.
+
+        Example::
+
+            self.log('train_loss', loss)
+
+        The default behavior per hook is documented here: :ref:`extensions/logging:Automatic Logging`.
+
+        Args:
+            name: key to log. Must be identical across all processes if using DDP or any other distributed strategy.
+            value: value to log. Can be a ``float``, ``Tensor``, or a ``Metric``.
+            prog_bar: if ``True`` logs to the progress bar.
+            logger: if ``True`` logs to the logger.
+            on_step: if ``True`` logs at this step. The default value is determined by the hook.
+                See :ref:`extensions/logging:Automatic Logging` for details.
+            on_epoch: if ``True`` logs epoch accumulated metrics. The default value is determined by the hook.
+                See :ref:`extensions/logging:Automatic Logging` for details.
+            reduce_fx: reduction function over step values for end of epoch. :meth:`torch.mean` by default.
+            enable_graph: if ``True``, will not auto detach the graph.
+            sync_dist: if ``True``, reduces the metric across devices. Use with care as this may lead to a significant
+                communication overhead.
+            sync_dist_group: the DDP group to sync across.
+            add_dataloader_idx: if ``True``, appends the index of the current dataloader to
+                the name (when using multiple dataloaders). If False, user needs to give unique names for
+                each dataloader to not mix the values.
+            batch_size: Current batch_size. This will be directly inferred from the loaded batch,
+                but for some data structures you might need to explicitly provide it.
+            metric_attribute: To restore the metric state, Lightning requires the reference of the
+                :class:`torchmetrics.Metric` in your model. This is found automatically if it is a model attribute.
+            rank_zero_only: Tells Lightning if you are calling ``self.log`` from every process (default) or only from
+                rank 0. If ``True``, you won't be able to use this metric as a monitor in callbacks
+                (e.g., early stopping). Warning: Improper use can lead to deadlocks! See
+                :ref:`Advanced Logging <visualize/logging_advanced:rank_zero_only>` for more details.
+
+        """
+        # If logging is disabled, then do nothing.
+        if not self.logging_enabled:
+            return
+
+        prefix, fn_kwargs = self._make_prefix_and_kwargs_dict(
+            _LogContextKwargs(
+                prog_bar=prog_bar,
+                logger=logger,
+                on_step=on_step,
+                on_epoch=on_epoch,
+                reduce_fx=reduce_fx,
+                enable_graph=enable_graph,
+                sync_dist=sync_dist,
+                sync_dist_group=sync_dist_group,
+                add_dataloader_idx=add_dataloader_idx,
+                batch_size=batch_size,
+                rank_zero_only=rank_zero_only,
+            )
+        )
+        name = f"{prefix}{name}"
+        return super().log(name, value, metric_attribute=metric_attribute, **fn_kwargs)
+
+    def log_dict(
+        self,
+        dictionary: Mapping[str, _METRIC] | torchmetrics.MetricCollection,
+        prog_bar: bool | None = None,
+        logger: bool | None = None,
+        on_step: bool | None = None,
+        on_epoch: bool | None = None,
+        reduce_fx: str | Callable | None = None,
+        enable_graph: bool | None = None,
+        sync_dist: bool | None = None,
+        sync_dist_group: Any | None = None,
+        add_dataloader_idx: bool | None = None,
+        batch_size: int | None = None,
+        rank_zero_only: bool | None = None,
+    ) -> None:
+        """Log a dictionary of values at once.
+
+        Example::
+
+            values = {'loss': loss, 'acc': acc, ..., 'metric_n': metric_n}
+            self.log_dict(values)
+
+        Args:
+            dictionary: key value pairs.
+                Keys must be identical across all processes if using DDP or any other distributed strategy.
+                The values can be a ``float``, ``Tensor``, ``Metric``, or ``MetricCollection``.
+            prog_bar: if ``True`` logs to the progress base.
+            logger: if ``True`` logs to the logger.
+            on_step: if ``True`` logs at this step.
+                ``None`` auto-logs for training_step but not validation/test_step.
+                The default value is determined by the hook.
+                See :ref:`extensions/logging:Automatic Logging` for details.
+            on_epoch: if ``True`` logs epoch accumulated metrics.
+                ``None`` auto-logs for val/test step but not ``training_step``.
+                The default value is determined by the hook.
+                See :ref:`extensions/logging:Automatic Logging` for details.
+            reduce_fx: reduction function over step values for end of epoch. :meth:`torch.mean` by default.
+            enable_graph: if ``True``, will not auto-detach the graph
+            sync_dist: if ``True``, reduces the metric across GPUs/TPUs. Use with care as this may lead to a significant
+                communication overhead.
+            sync_dist_group: the ddp group to sync across.
+            add_dataloader_idx: if ``True``, appends the index of the current dataloader to
+                the name (when using multiple). If ``False``, user needs to give unique names for
+                each dataloader to not mix values.
+            batch_size: Current batch size. This will be directly inferred from the loaded batch,
+                but some data structures might need to explicitly provide it.
+            rank_zero_only: Tells Lightning if you are calling ``self.log`` from every process (default) or only from
+                rank 0. If ``True``, you won't be able to use this metric as a monitor in callbacks
+                (e.g., early stopping). Warning: Improper use can lead to deadlocks! See
+                :ref:`Advanced Logging <visualize/logging_advanced:rank_zero_only>` for more details.
+
+        """
+
+        _, fn_kwargs = self._make_prefix_and_kwargs_dict(
+            _LogContextKwargs(
+                prog_bar=prog_bar,
+                logger=logger,
+                on_step=on_step,
+                on_epoch=on_epoch,
+                reduce_fx=reduce_fx,
+                enable_graph=enable_graph,
+                sync_dist=sync_dist,
+                sync_dist_group=sync_dist_group,
+                add_dataloader_idx=add_dataloader_idx,
+                batch_size=batch_size,
+                rank_zero_only=rank_zero_only,
+            )
+        )
+        # NOTE: Prefix will be handled by the individual log calls.
+        return super().log_dict(dictionary, **fn_kwargs)

nshtrainer-1.0.0b47/src/nshtrainer/metrics/_config.py

@@ -1,42 +0,0 @@
-from __future__ import annotations
-
-import builtins
-from typing import Any, Literal
-
-import nshconfig as C
-
-
-class MetricConfig(C.Config):
-    name: str
-    """The name of the primary metric."""
-
-    mode: Literal["min", "max"]
-    """
-    The mode of the primary metric:
-    - "min" for metrics that should be minimized (e.g., loss)
-    - "max" for metrics that should be maximized (e.g., accuracy)
-    """
-
-    @property
-    def validation_monitor(self) -> str:
-        return f"val/{self.name}"
-
-    def __post_init__(self):
-        for split in ("train", "val", "test", "predict"):
-            if self.name.startswith(f"{split}/"):
-                raise ValueError(
-                    f"Primary metric name should not start with '{split}/'. "
-                    f"Just use '{self.name[len(split) + 1:]}' instead. "
-                    "The split name is automatically added depending on the context."
-                )
-
-    @classmethod
-    def loss(cls, mode: Literal["min", "max"] = "min"):
-        return cls(name="loss", mode=mode)
-
-    @property
-    def best(self):
-        return builtins.min if self.mode == "min" else builtins.max
-
-    def is_better(self, a: Any, b: Any):
-        return self.best(a, b) == a