lightning/pytorch_lightning/metrics/metric.py

# Copyright The PyTorch Lightning team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from abc import ABC, abstractmethod
from typing import Any, Optional

import torch
import torch.distributed

from pytorch_lightning.metrics.converters import (
    tensor_metric, numpy_metric, tensor_collection_metric)
from pytorch_lightning.utilities.apply_func import apply_to_collection
from pytorch_lightning.utilities.device_dtype_mixin import DeviceDtypeModuleMixin


class Metric(DeviceDtypeModuleMixin, torch.nn.Module, ABC):
    """
    Abstract base class for metric implementation.

    Should be used to implement metrics that
    1. Return multiple Outputs
    2. Handle their own DDP sync
    """

    def __init__(self, name: str):
        """
        Args:
            name: the metric's name

        """
        super().__init__()
        self.name = name

    @abstractmethod
    def forward(self, *args, **kwargs) -> torch.Tensor:
        """
        Implements the actual metric computation.

        Returns:
            metric value

        """
        raise NotImplementedError


class TensorMetric(Metric):
    """
    Base class for metric implementation operating directly on tensors.
    All inputs and outputs will be casted to tensors if necessary.
    Already handles DDP sync and input/output conversions.
    """

    def __init__(self, name: str,
                 reduce_group: Optional[Any] = None,
                 reduce_op: Optional[Any] = None):
        """

        Args:
            name: the metric's name
            reduce_group: the process group for DDP reduces (only needed for DDP training).
                Defaults to all processes (world)
            reduce_op: the operation to perform during reduction within DDP (only needed for DDP training).
                Defaults to sum.
        """
        super().__init__(name)
        self._orig_call = tensor_metric(group=reduce_group,
                                        reduce_op=reduce_op)(super().__call__)

    def __call__(self, *args, **kwargs) -> torch.Tensor:
        def _to_device_dtype(x: torch.Tensor) -> torch.Tensor:
            return x.to(device=self.device, dtype=self.dtype, non_blocking=True)

        return apply_to_collection(self._orig_call(*args, **kwargs), torch.Tensor,
                                   _to_device_dtype)


class TensorCollectionMetric(Metric):
    """
    Base class for metric implementation operating directly on tensors.
    All inputs will be casted to tensors if necessary. Outputs won't be casted.
    Already handles DDP sync and input conversions.

    This class differs from :class:`TensorMetric`, as it assumes all outputs to
    be collections of tensors and does not explicitly convert them. This is
    necessary, since some collections (like for ROC, Precision-Recall Curve etc.)
    cannot be converted to tensors at the highest level.
    All numpy arrays and numbers occuring in these outputs will still be converted.

    Use this class as a baseclass, whenever you want to ensure inputs are
    tensors and outputs cannot be converted to tensors automatically

    """

    def __init__(self, name: str,
                 reduce_group: Optional[Any] = None,
                 reduce_op: Optional[Any] = None):
        """

        Args:
            name: the metric's name
            reduce_group: the process group for DDP reduces (only needed for DDP training).
                Defaults to all processes (world)
            reduce_op: the operation to perform during reduction within DDP (only needed for DDP training).
                Defaults to sum.
        """
        super().__init__(name)
        self._orig_call = tensor_collection_metric(group=reduce_group,
                                                   reduce_op=reduce_op)(super().__call__)

    def __call__(self, *args, **kwargs) -> torch.Tensor:
        def _to_device_dtype(x: torch.Tensor) -> torch.Tensor:
            return x.to(device=self.device, dtype=self.dtype, non_blocking=True)

        return apply_to_collection(self._orig_call(*args, **kwargs), torch.Tensor,
                                   _to_device_dtype)


class NumpyMetric(Metric):
    """
    Base class for metric implementation operating on numpy arrays.
    All inputs will be casted to numpy if necessary and all outputs will
    be casted to tensors if necessary.
    Already handles DDP sync and input/output conversions.
    """

    def __init__(self, name: str,
                 reduce_group: Optional[Any] = None,
                 reduce_op: Optional[Any] = None):
        """

        Args:
            name: the metric's name
            reduce_group: the process group for DDP reduces (only needed for DDP training).
                Defaults to all processes (world)
            reduce_op: the operation to perform during reduction within DDP (only needed for DDP training).
                Defaults to sum.
        """
        super().__init__(name)
        self._orig_call = numpy_metric(group=reduce_group,
                                       reduce_op=reduce_op)(super().__call__)

    def __call__(self, *args, **kwargs) -> torch.Tensor:
        def _to_device_dtype(x: torch.Tensor) -> torch.Tensor:
            return x.to(device=self.device, dtype=self.dtype, non_blocking=True)

        return apply_to_collection(self._orig_call(*args, **kwargs), torch.Tensor,
                                   _to_device_dtype)