Scheduler defines how often to run and make inferences, as well as what timerange to use to train the model.
Is specified in scheduler
section of a config for VictoriaMetrics Anomaly Detection.
Note: Starting from v1.11.0 scheduler section in config supports multiple schedulers via aliasing.
Also,vmanomaly
expects scheduler section to be namedschedulers
. Using old (flat) format withscheduler
key is deprecated and will be removed in future versions.
schedulers:
scheduler_periodic_1m:
# class: "scheduler.periodic.PeriodicScheduler"
infer_every: "1m"
fit_every: "2m"
fit_window: "3h"
scheduler_periodic_5m:
# class: "scheduler.periodic.PeriodicScheduler"
infer_every: "5m"
fit_every: "10m"
fit_window: "3h"
...
Old-style configs (< 1.11.0)
scheduler:
# class: "scheduler.periodic.PeriodicScheduler"
infer_every: "1m"
fit_every: "2m"
fit_window: "3h"
...
will be implicitly converted to
schedulers:
default_scheduler: # default scheduler alias, backward compatibility
# class: "scheduler.periodic.PeriodicScheduler"
infer_every: "1m"
fit_every: "2m"
fit_window: "3h"
...
Parameters#
class
: str, default="scheduler.periodic.PeriodicScheduler"
,
options={"scheduler.periodic.PeriodicScheduler"
, "scheduler.oneoff.OneoffScheduler"
, "scheduler.backtesting.BacktestingScheduler"
}
"scheduler.periodic.PeriodicScheduler"
: periodically runs the models on new data. Useful for consecutive re-trainings to counter data drift and model degradation over time."scheduler.oneoff.OneoffScheduler"
: runs the process once and exits. Useful for testing."scheduler.backtesting.BacktestingScheduler"
: imitates consecutive backtesting runs of OneoffScheduler. Runs the process once and exits. Use to get more granular control over testing on historical data.
Depending on selected class, different parameters should be used
Periodic scheduler#
Parameters#
For periodic scheduler parameters are defined as differences in times, expressed in difference units, e.g. days, hours, minutes, seconds.
Examples: "50s"
, "4m"
, "3h"
, "2d"
, "1w"
.
Time granularity | |
---|---|
s | seconds |
m | minutes |
h | hours |
d | days |
w | weeks |
Parameter | Type | Example | Description |
---|---|---|---|
fit_window | str | "14d" | What time range to use for training the models. Must be at least 1 second. |
infer_every | str | "1m" | How often a model will write its conclusions on newly added data. Must be at least 1 second. |
fit_every | str, Optional | "1h" | How often to completely retrain the models. If missing value of infer_every is used and retrain on every inference run. |
Periodic scheduler config example#
scheduler:
class: "scheduler.periodic.PeriodicScheduler"
fit_window: "14d"
infer_every: "1m"
fit_every: "1h"
This part of the config means that vmanomaly
will calculate the time window of the previous 14 days and use it to train a model. Every hour model will be retrained again on 14 days’ data, which will include + 1 hour of new data. The time window is strictly the same 14 days and doesn’t extend for the next retrains. Every minute vmanomaly
will produce model inferences for newly added data points by using the model that is kept in memory at that time.
Oneoff scheduler#
Parameters#
For Oneoff scheduler timeframes can be defined in Unix time in seconds or ISO 8601 string format. ISO format supported time zone offset formats are:
- Z (UTC)
- ±HH:MM
- ±HHMM
- ±HH
If a time zone is omitted, a timezone-naive datetime is used.
Defining fitting timeframe#
Format | Parameter | Type | Example | Description |
---|---|---|---|---|
ISO 8601 | fit_start_iso | str | "2022-04-01T00:00:00Z", "2022-04-01T00:00:00+01:00", "2022-04-01T00:00:00+0100", "2022-04-01T00:00:00+01" | Start datetime to use for training a model. ISO string or UNIX time in seconds. |
UNIX time | fit_start_s | float | 1648771200 | |
ISO 8601 | fit_end_iso | str | "2022-04-10T00:00:00Z", "2022-04-10T00:00:00+01:00", "2022-04-10T00:00:00+0100", "2022-04-10T00:00:00+01" | End datetime to use for training a model. Must be greater than fit_start_* . ISO string or UNIX time in seconds. |
UNIX time | fit_end_s | float | 1649548800 |
Defining inference timeframe#
Format | Parameter | Type | Example | Description |
---|---|---|---|---|
ISO 8601 | infer_start_iso | str | "2022-04-11T00:00:00Z", "2022-04-11T00:00:00+01:00", "2022-04-11T00:00:00+0100", "2022-04-11T00:00:00+01" | Start datetime to use for a model inference. ISO string or UNIX time in seconds. |
UNIX time | infer_start_s | float | 1649635200 | |
ISO 8601 | infer_end_iso | str | "2022-04-14T00:00:00Z", "2022-04-14T00:00:00+01:00", "2022-04-14T00:00:00+0100", "2022-04-14T00:00:00+01" | End datetime to use for a model inference. Must be greater than infer_start_* . ISO string or UNIX time in seconds. |
UNIX time | infer_end_s | float | 1649894400 |
ISO format scheduler config example#
scheduler:
class: "scheduler.oneoff.OneoffScheduler"
fit_start_iso: "2022-04-01T00:00:00Z"
fit_end_iso: "2022-04-10T00:00:00Z"
infer_start_iso: "2022-04-11T00:00:00Z"
infer_end_iso: "2022-04-14T00:00:00Z"
UNIX time format scheduler config example#
scheduler:
class: "scheduler.oneoff.OneoffScheduler"
fit_start_iso: 1648771200
fit_end_iso: 1649548800
infer_start_iso: 1649635200
infer_end_iso: 1649894400
Backtesting scheduler#
Parameters#
As for Oneoff scheduler, timeframes can be defined in Unix time in seconds or ISO 8601 string format. ISO format supported time zone offset formats are:
- Z (UTC)
- ±HH:MM
- ±HHMM
- ±HH
If a time zone is omitted, a timezone-naive datetime is used.
Defining overall timeframe#
This timeframe will be used for slicing on intervals (fit_window, infer_window == fit_every)
, starting from the latest available time point, which is to_*
and going back, until no full fit_window + infer_window
interval exists within the provided timeframe.
Format | Parameter | Type | Example | Description |
---|---|---|---|---|
ISO 8601 | from_iso | str | "2022-04-01T00:00:00Z", "2022-04-01T00:00:00+01:00", "2022-04-01T00:00:00+0100", "2022-04-01T00:00:00+01" | Start datetime to use for backtesting. |
UNIX time | from_s | float | 1648771200 | |
ISO 8601 | to_iso | str | "2022-04-10T00:00:00Z", "2022-04-10T00:00:00+01:00", "2022-04-10T00:00:00+0100", "2022-04-10T00:00:00+01" | End datetime to use for backtesting. Must be greater than from_start_* . |
UNIX time | to_s | float | 1649548800 |
Defining training timeframe#
The same explicit logic as in Periodic scheduler
Format | Parameter | Type | Example | Description |
---|---|---|---|---|
ISO 8601 | fit_window | str | "PT1M", "P1H" | What time range to use for training the models. Must be at least 1 second. |
Prometheus-compatible | "1m", "1h" |
Defining inference timeframe#
In BacktestingScheduler
, the inference window is implicitly defined as a period between 2 consecutive model fit_every
runs. The latest inference window starts from to_s
- fit_every
and ends on the latest available time point, which is to_s
. The previous periods for fit/infer are defined the same way, by shifting fit_every
seconds backwards until we get the last full fit period of fit_window
size, which start is >= from_s
.
Format | Parameter | Type | Example | Description |
---|---|---|---|---|
ISO 8601 | fit_every | str | "PT1M", "P1H" | What time range to use previously trained model to infer on new data until next retrain happens. |
Prometheus-compatible | "1m", "1h" |
ISO format scheduler config example#
scheduler:
class: "scheduler.backtesting.BacktestingScheduler"
from_start_iso: '2021-01-01T00:00:00Z'
to_end_iso: '2021-01-14T00:00:00Z'
fit_window: 'P14D'
fit_every: 'PT1H'
UNIX time format scheduler config example#
scheduler:
class: "scheduler.backtesting.BacktestingScheduler"
from_start_s: 167253120
to_end_s: 167443200
fit_window: '14d'
fit_every: '1h'