otp.agg.exp_w_average#

exp_w_average(decay, decay_value_type='lambda', running=False, bucket_interval=0, bucket_units=None, bucket_time='end', bucket_end_condition=None, boundary_tick_bucket='new', all_fields=False, group_by=None, end_condition_per_group=False, time_series_type='state_ts')#

EXP_W_AVERAGE aggregation.

For each bucket, computes the exponentially weighted average value of the specified numeric attribute. Weights of data points in a bucket decrease exponentially in the direction from the most recent tick to the most aged one, being equal to exp(-Lambda * N) for a fixed weight decay value Lambda, where N ranges over 0, 1, 2, … as ticks in reverse order of their arrival are treated. Once the weights are known, the average is found using the formula sum(weight*value)/sum(weight), where the sum is computed across all data points.

Parameters

decay (float) – Weight decay. If decay_value_type is set to lambda, decay provides the value of the Lambda variable in the aforementioned formula. Otherwise, if decay_value_type is set to half_life_index, decay specifies the necessary number of consecutive ticks, the first one of which would have twice less the weight of the last one. The Lambda value is then calculated using this number.
decay_value_type (Literal['lambda', 'half_life_index'], default=lambda) – The decay value can specified either directly or indirectly, controlled respectively by lambda and half_life_index values of this parameter.
running (bool, default=False) –
Aggregation will be calculated as sliding window. running and bucket_interval parameters determines when new buckets are created.
- running = True
  
  aggregation will be calculated in a sliding window.
  - bucket_interval = N (N > 0)
    
    Window size will be N. Output tick will be generated when tick “enter” window (arrival event) and when “exit” window (exit event)
  - bucket_interval = 0
    
    Left boundary of window will be bound to start time. For each tick aggregation will be calculated in [start_time; tick_t].
- running = False
  
  buckets partition the [query start time, query end time) interval into non-overlapping intervals of size bucket_interval (with the last interval possibly of a smaller size). If bucket_interval is set to 0 a single bucket for the entire interval is created.
  
  Note that in non-running mode OneTick unconditionally divides the whole time interval into specified number of buckets. It means that you will always get this specified number of ticks in the result, even if you have less ticks in the input data.
Default: False
bucket_interval (int or Operation or OnetickParameter or symbol parameter or datetime offset object, default=0) –
Determines the length of each bucket (units depends on bucket_units).

If Operation of bool type is passed, acts as bucket_end_condition.

Bucket interval can be set via datetime offset objects like otp.Second, otp.Minute, otp.Hour, otp.Day, otp.Month. In this case you could omit setting bucket_units parameter.

Bucket interval can also be set with integer OnetickParameter or symbol parameter.
bucket_units (Optional[Literal['seconds', 'ticks', 'days', 'months', 'flexible']], default=None) –
Set bucket interval units.

By default, if bucket_units and bucket_end_condition not specified, set to seconds. If bucket_end_condition specified, then bucket_units set to flexible.

If set to flexible then bucket_end_condition must be set.

Note that seconds bucket unit doesn’t take into account daylight-saving time of the timezone, so you may not get expected results when using, for example, 24 * 60 * 60 seconds as bucket interval. In such case use days bucket unit instead. See example in onetick.py.agg.sum().
bucket_time (Literal['start', 'end'], default=end) –
Control output timestamp.
- start
  
  the timestamp assigned to the bucket is the start time of the bucket.
- end
  
  the timestamp assigned to the bucket is the end time of the bucket.
bucket_end_condition (condition, default=None) –
An expression that is evaluated on every tick. If it evaluates to “True”, then a new bucket is created. This parameter is only used if bucket_units is set to “flexible”.

Also can be set via bucket_interval parameter by passing Operation object.
boundary_tick_bucket (Literal['new', 'previous'], default=new) –
Controls boundary tick ownership.
- previous
  
  A tick on which bucket_end_condition evaluates to “true” belongs to the bucket being closed.
- new
  
  tick belongs to the new bucket.
This parameter is only used if bucket_units is set to “flexible”
all_fields (Union[bool, str], default=False) –
- all_fields = True
  
  output ticks include all fields from the input ticks
  - running = True
  an output tick is created only when a tick enters the sliding window
  - running = False
  fields of first tick in bucket will be used
- all_fields = False and running = True
  
  output ticks are created when a tick enters or leaves the sliding window.
- all_fields = “when_ticks_exit_window” and running = True
  
  output ticks are generated only for exit events, but all attributes from the exiting tick are copied over to the output tick and the aggregation is added as another attribute.
group_by (list, str or expression, default=None) – When specified, each bucket is broken further into additional sub-buckets based on specified field values. If Operation is used then GROUP_{i} column is added. Where i is index in group_by list. For example, if Operation is the only element in group_by list then GROUP_0 field will be added.
end_condition_per_group (bool, default=False) –
Controls application of bucket_end_condition in groups.
- end_condition_per_group = True
  
  bucket_end_condition is applied only to the group defined by group_by
- end_condition_per_group = False
  
  bucket_end_condition applied across all groups
This parameter is only used if bucket_units is set to “flexible”.

When set to True, applies to all bucketing conditions. Useful, for example, if you need to specify group_by, and you want to group items first, and create buckets after that.
time_series_type (Literal['event_ts', 'state_ts'], default=state_ts) –
Controls initial value for each bucket
- event_ts
  
  only ticks from current bucket used for calculations
- state_ts
  - if there is a tick in bucket with timestamp = bucket start
    
    only ticks in bucket used for calculation max value
  - else
    
    latest tick from previous bucket included in current bucket

Examples

Basic example

>>> data = otp.Ticks({'A': [1.0, 2.0, 3.0, 3.0, 4.0]})
>>> data = data.exp_w_average('A', decay=2, bucket_interval=2, bucket_units='ticks')
>>> otp.run(data)
                     Time         A
0 2003-12-01 00:00:00.001  1.880797
1 2003-12-01 00:00:00.003  2.984124
2 2003-12-04 00:00:00.000  3.880797

You can switch to half_life_index as decay_value_type

>>> data = otp.Ticks({'A': [1.0, 2.0, 3.0, 3.0, 4.0]})
>>> data = data.exp_w_average(
...     'A', decay=2, decay_value_type='half_life_index', bucket_interval=2, bucket_units='ticks',
... )
>>> otp.run(data)
                     Time         A
0 2003-12-01 00:00:00.001  1.585786
1 2003-12-01 00:00:00.003  2.773459
2 2003-12-04 00:00:00.000  3.585786