Source code for composer.loggers.logger

# Copyright 2022 MosaicML Composer authors
# SPDX-License-Identifier: Apache-2.0

"""Base classes, functions, and variables for logger."""

from __future__ import annotations

import collections.abc
import operator
import pathlib
from functools import reduce
from typing import TYPE_CHECKING, Any, Optional, Sequence, Union

import numpy as np
import torch

from composer.utils import ensure_tuple, format_name_with_dist

if TYPE_CHECKING:
    from composer.core import State
    from composer.loggers.logger_destination import LoggerDestination

__all__ = ['LoggerDestination', 'Logger', 'format_log_data_value']


[docs]class Logger: """An interface to record training data. The :class:`.Trainer`, instances of :class:`.Callback`, and instances of :class:`~composer.core.algorithm.Algorithm` invoke the logger to record data such as the epoch, training loss, and custom metrics as provided by individual callbacks and algorithms. This class does not store any data itself; instead, it routes all data to the ``destinations``. Each destination (e.g. the :class:`~composer.loggers.file_logger.FileLogger`, :class:`~composer.loggers.in_memory_logger.InMemoryLogger`) is responsible for storing the data itself (e.g. writing it to a file or storing it in memory). Args: state (State): The training state. destinations (LoggerDestination | Sequence[LoggerDestination], optional): The logger destinations, to where logging data will be sent. (default: ``None``) Attributes: destinations (Sequence[LoggerDestination]): A sequence of :class:`~.LoggerDestination` to where logging calls will be sent. """ def __init__( self, state: State, destinations: Optional[Union[LoggerDestination, Sequence[LoggerDestination]]] = None, ): self.destinations = ensure_tuple(destinations) self._state = state def log_traces(self, traces: dict[str, Any]): for destination in self.destinations: destination.log_traces(traces) def log_hyperparameters(self, parameters: dict[str, Any]): for destination in self.destinations: destination.log_hyperparameters(parameters) def log_table( self, columns: list[str], rows: list[list[Any]], name: str = 'Table', step: Optional[int] = None, ) -> None: if step is None: step = self._state.timestamp.batch.value for destination in self.destinations: destination.log_table(columns, rows, name, step) def log_metrics(self, metrics: dict[str, float], step: Optional[int] = None) -> None: if step is None: step = self._state.timestamp.batch.value for destination in self.destinations: destination.log_metrics(metrics, step)
[docs] def log_images( self, images: Union[np.ndarray, torch.Tensor, Sequence[Union[np.ndarray, torch.Tensor]]], name: str = 'Images', channels_last: bool = False, step: Optional[int] = None, masks: Optional[dict[str, Union[np.ndarray, torch.Tensor, Sequence[Union[np.ndarray, torch.Tensor]]]]] = None, mask_class_labels: Optional[dict[int, str]] = None, use_table: bool = True, ): """Log images. Logs any tensors or arrays as images. Args: images (np.ndarray | torch.Tensor | Sequence[np.ndarray | torch.Tensor]): Dictionary mapping image(s)' names (str) to an image of array of images. name (str): The name of the image(s). (Default: ``'Images'``) channels_last (bool): Whether the channel dimension is first or last. (Default: ``False``) step (int, optional): The current step or batch of training at the time of logging. Defaults to None. If not specified the specific LoggerDestination implementation will choose a step (usually a running counter). masks (dict[str, np.ndarray | torch.Tensor | Sequence[np.ndarray | torch.Tensor]], optional): A dictionary mapping the mask name (e.g. predictions or ground truth) to a sequence of masks. mask_class_labels (dict[int, str], optional): Dictionary mapping label id to its name. Used for labelling each color in the mask. use_table (bool): Whether to make a table of the images or not. (default: ``True``). Only for use with WandB. """ if step is None: step = self._state.timestamp.batch.value for destination in self.destinations: destination.log_images( images, name, channels_last, step, masks, mask_class_labels, use_table, )
[docs] def upload_file( self, remote_file_name: str, file_path: Union[pathlib.Path, str], *, overwrite: bool = False, ): """Upload ``file_path`` as a file named ``remote_file_name``. Both ``file_path`` and ``remote_file_name`` can be specified as format strings. See :func:`~.composer.utils.file_helpers.format_name_with_dist` for more information. .. seealso:: :doc:`Uploading Files</trainer/file_uploading>` for notes for file uploading. Args: remote_file_name (str): A format string for the name of the file. file_path (str | pathlib.Path): A format string for the file path. overwrite (bool, optional): Whether to overwrite an existing file with the same ``remote_file_name``. (default: ``False``) """ file_path = format_name_with_dist(format_str=str(file_path), run_name=self._state.run_name) file_path = pathlib.Path(file_path) for destination in self.destinations: destination.upload_file( state=self._state, remote_file_name=format_name_with_dist(format_str=remote_file_name, run_name=self._state.run_name), file_path=file_path, overwrite=overwrite, )
[docs] def has_file_upload_destination(self) -> bool: """Determines if the logger has a destination which supports uploading files. Needed for checking if a model can be exported via this logger. Returns: bool: Whether any of the destinations support uploading files. """ for destination in self.destinations: if destination.can_upload_files(): return True return False
def format_log_data_value(data: Any) -> str: """Recursively formats a given log data value into a string. Args: data: Data to format. Returns: str: ``data`` as a string. """ if data is None: return 'None' if isinstance(data, str): return f"\"{data}\"" if isinstance(data, int): return str(data) if isinstance(data, float): return f'{data:.4f}' if isinstance(data, torch.Tensor): if data.shape == () or reduce(operator.mul, data.shape, 1) == 1: return format_log_data_value(data.cpu().item()) return 'Tensor of shape ' + str(data.shape) if isinstance(data, collections.abc.Mapping): output = ['{ '] for k, v in data.items(): assert isinstance(k, str) v = format_log_data_value(v) output.append(f"\"{k}\": {v}, ") output.append('}') return ''.join(output) if isinstance(data, collections.abc.Iterable): return '[' + ', '.join(format_log_data_value(v) for v in data) + ']' # Unknown format catch-all return str(data)