AlgorithmTimeLimitManager
AlgorithmTimeLimitManager(
additional_time_bucket: ITokenBucket,
time_loop_maximum: timedelta,
)
Bases: Object, IIsolatorLimitResultProvider
Provides an implementation of IIsolatorLimitResultProvider that tracks the algorithm manager's time loops and enforces a maximum amount of time that each time loop may take to execute. The isolator uses the result provided by is_within_limit to determine if it should terminate the algorithm for violation of the imposed limits.
Initializes a new instance of AlgorithmTimeLimitManager to manage the creation of IsolatorLimitResult instances as it pertains to the algorithm manager's time loop
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
additional_time_bucket
|
ITokenBucket
|
Provides a bucket of additional time that can be requested to be spent to give execution time for things such as training scheduled events |
required |
time_loop_maximum
|
timedelta
|
Specifies the maximum amount of time the algorithm is permitted to spend in a single time loop. This value can be overriden if certain actions are taken by the algorithm, such as invoking the training methods. |
required |
additional_time_bucket: ITokenBucket
Gets the additional time bucket which is responsible for tracking additional time requested for processing via long-running scheduled events. In LEAN, we use the LeakyBucket
is_within_limit() -> IsolatorLimitResult
Determines whether or not the algorithm time loop is considered within the limits
request_additional_time(minutes: int) -> None
Requests additional time to continue executing the current time step. At time of writing, this is intended to be used to provide training scheduled events additional time to allow complex training models time to execute while also preventing abuse by enforcing certain control parameters set via the job packet.
Each time this method is invoked, this time limit manager will increase the allowable execution time by the specified number of whole minutes
start_new_time_step() -> None
Invoked by the algorithm at the start of each time loop. This resets the current time step elapsed time.
try_request_additional_time(minutes: int) -> bool
Attempts to requests additional time to continue executing the current time step. At time of writing, this is intended to be used to provide training scheduled events additional time to allow complex training models time to execute while also preventing abuse by enforcing certain control parameters set via the job packet.
Each time this method is invoked, this time limit manager will increase the allowable execution time by the specified number of whole minutes