TaskTimer

class lightautoml.utils.timer.TaskTimer(pipe_timer, key=None, score=1.0, overhead=1, mode=1, default_tuner_time_rate=0.7)

Bases: Timer

Timer is used to control time over single ML task run.

It decides how much time is ok to spend on tuner and if we have enough time to calc more folds.

Parameters:

• pipe_timer (PipelineTimer) – Global automl timer.

• key (Optional[str]) – String name that will be associated with this task.

• score (float) – Time score for current task. For ex. if you want to give more of total time to task set it > 1.

• overhead (Optional[float]) – See overhead of PipelineTimer.

• mode (int) – See mode for PipelineTimer.

• default_tuner_time_rate (float) – If no timing history for the moment of estimating tuning time, timer will use this rate of time_left.

property in_progress

Check if the task is running.

start()

Starts counting down.

Returns:

self.

set_control_point()

Set control point.

Updates the countdown and time left parameters.

write_run_info()

Collect timer history.

get_run_results()

Get timer history.

Return type:

Optional[ndarray]

Returns:

None if there is no history, or array with history of runs.

get_run_scores()

Get timer scores.

Return type:

Optional[ndarray]

Returns:

None if there is no scores, or array with scores of runs.

estimate_folds_time(n_folds=1)

Estimate time for n_folds.

Parameters:

n_folds (int) – Number of folds.

Return type:

Optional[float]

Returns:

Estimated time needed to run all n_folds.

estimate_tuner_time(n_folds=1)

Estimates time that is ok to spend on tuner.

Parameters:

n_folds (int) – Number of folds.

Return type:

float

Returns:

How much time timer will be able spend on tuner.

time_limit_exceeded()

Estimate time limit and send results to parent timer.

Return type:

bool

Returns:

True if time limit exceeded.

split_timer(n_parts)

Split the timer into equal-sized tasks.

Parameters:

n_parts (int) – Number of tasks.

Return type:

List[TaskTimer]

Returns:

Timers for each tasks.