Logml API#

logml.logger#

Logger class.

class Logger(n_epochs, n_batches, log_interval=1, name=None, *, styles='', sizes=6, average=None, silent=False, show_bar=True, show_time=True, bold_keys=False, name_style='')#

Bases: object

Logger class. Handles int, float, str and bool values.

Parameters:

  • n_epochs (int) – Number of epochs.

  • n_batches (Optional[int]) – Number of batches per epoch. Set to None if not available.

  • log_interval (int, optional) – Number of batches between each log. None for 4 update per second. By default 1.

  • name (Optional[str], optional) – Name of the logger. It will be display at the top of the logs. By default None (no name).

  • styles (Union[Dict, str], optional) – Default style of the values. Include color, bold, italic and more See https://rich.readthedocs.io/en/stable/style.html for more details. E.g. {‘loss’: ‘bold red’, ‘acc’: ‘italic #af00ff’} Use a single string to apply the same style to all values. These styles can also be set or overwritten in the log method. By default, use rich default style.

  • sizes (Union[Dict, int], optional) – Default sizes to display for each numerical values. Note that the dot is counted as a character. E.g. {‘loss’: 3, ‘acc’: 2}. Use a single int to apply the same size to all values. These numbers can also be set or overwritten in the log method. By default 6.

  • average (Optional[List[str]], optional) – Default list of the values to average over the epoch. This can also be overwritten in the log method. By default None.

  • silent (bool, optional) – Whether to print the logs or not. By default False.

  • show_bar (bool, optional) – Whether to show the progress bar or not. By default True.

  • show_time (bool, optional) – Whether to show the time or not. By default True.

  • bold_keys (bool, optional) – Whether to bold the key or not. By default False.

  • name_style (str, optional) – Style of the name. By default, use the rich default style.

Examples

logger = Logger(n_epochs=10, n_batches=100)
for _ in range(10):
    for _ in logger.tqdm(range(100)):
        logger.log(loss=0.5, acc=0.9)

Attribute#

stepint

Number of batches seen since the beginning.

current_epochint

Current epoch number (starting at 1).

current_batchint

Current batch number (starting at 1).

valsDict[str, VarType]

Last values called inside log (not averaged nor resized).

mean_valsDict[str, VarType]

Current mean values (only numerical values).

start()#

Set the start time of the training (already called at initialization).

Return type:

None

log(values=None, *, message='', styles=None, sizes=None, average=None)#

Log the values with style.

Parameters:

  • values (Dict[str, Any] | None) – Values to log. E.g. {‘loss’: 0.1, ‘acc’: 0.9}

  • message (str, optional) – Message to display at the end of the log. By default empty.

  • styles (Union[Dict, str, None], optional) –

    Style of the values. Include color, bold, italic and more See https://rich.readthedocs.io/en/stable/style.html for more details. E.g.:

    {'loss': 'bold red', 'acc': 'italic #af00ff'}

    Use a single string to apply the same style to all values. By default None (use the default style).

  • sizes (Union[Dict, int], optional) –

    Size of the values to display for each numerical values. E.g.:

    {'loss': 7, 'acc': 2}

    Use a single int to apply the same size to all values. By default None (use the default style).

  • average (Optional[List[str]], optional) – List of the values to average over the epoch. None to not average. By default None.

Return type:

None

tqdm(iterable, *, reset_means=True)#

Browse an iterable dataset and automatically update the epoch and batch.

No need to explicitly call new_epoch() or new_batch() as they are automatically called. If n_batches is not specified and the iterable implement a __len__ attribute, the number of batch is inferred and store in n_batches.

Parameters:

  • iterable (Iterable) – Iterable dataset.

  • reset_means (bool, optional) – Whether to reset the computed mean values at the start of the epoch. By default True.

Return type:

Any

new_epoch(*, reset_means=True)#

Declare a new epoch.

Parameters:

reset_means (bool, optional) – Whether to reset the computed mean values at the start of the epoch. By default True.

Return type:

None

start_epoch(*, reset_means=True)#

Declare a new epoch. Alias for new_epoch().

Return type:

None

new_batch()#

Declare a new batch.

Return type:

None

start_batch()#

Declare a new batch. Alias for new_batch().

Return type:

None

detach(*, skipline=True)#

Stop the live display.

Stop the live display while keeping the current display visible in the terminal. :rtype: None

Note

Calling detach() or stop() is necessary to print something at the end of an epoch. Otherwise, the print will be shown above the live display.

stop()#

Stop the live display.

Stop the live display while keeping the current display visible in the terminal. Alias for detach() with skipline=False. :rtype: None

Note

Calling stop() or detach() is necessary to print something at the end of an epoch. Otherwise, the print will be shown above the live display.

pause()#

Pause the current time.

Return type:

None

resume()#

Resume the current time if paused.

Return type:

None

get_current_time()#

Get the current time. Take into account the pause periods.

Return type:

float

reset()#

Reset the logger as at initialization.

Return type:

None

get_vals(*, average=None)#

Get the last values called with log, optionally averaged.

Parameters:

average (List[str], optional) – List of keys to return average. Support regex expressions. None for no averaged value. By default None.

Returns:

Last values called with log (optionally averaged).

Return type:

Dict[str, VarType]

logml.time_utils#

Utilities for time data.

sec_to_timestr(sec)#

Return a string corresponding to the time given in seconds.

Return type:

str

get_time_range(current_time, start_glob, start_epoch, current_epoch, current_batch, n_epochs, n_batches)#

Get delta and eta times for global and epoch.

Parameters:

  • current_time (float) – Current time in seconds.

  • start_glob (float) – Global starting time in seconds.

  • start_epoch (float) – Epoch starting time in seconds. None if not available.

  • current_epoch (int, optional) – Current epoch (starting at 1). None if not available.

  • current_batch (int, optional) – Current batch (starting at 1). None if not available.

  • n_epochs (int, optional) – Number of epochs. None if not available.

  • n_batches (Optional[int], optional) – Number of batches per epoch. None if not available. By default None.

Raises:

ValueError – If current_batch, current_epoch, n_epochs or n_batch are 0.

Return type:

Tuple[float, float, Optional[float], Optional[float]]

Returns:

  • delta_glob (float) – Delta time since global start in seconds.

  • delta_epoch (float) – Delta time since epoch start in seconds. None if unavailable.

  • eta_glob (Optional[float]) – Estimated time until global end in seconds. None if unavailable.

  • eta_epoch (Optional[float]) – Estimated time until epoch end in seconds. None if unavailable.