Series defines the data structure of the indicator data.
indicators use this series interface to manage these time-series data.
The interface allow strategy developers to switch similar indicators without checking the code.
Signal contributors or indicator developers were also able to be benefit from the existing interface functions, such as Add
, Mul
, Minus
, and Div
, without rebuilding the wheels.
The series interface in bbgo borrows the concept of series
type in pinescript that allow us to query data in time-based reverse order (data that created later will be the former object in series). Right now, based on the return type, we have two interfaces been defined in pkg/types/indicator.go:
type Series interface {
Last() float64 // newest element
Index(i int) float64 // i >= 0, query float64 data in reverse order using i as index
Length() int // length of data piped in array
}
and
type BoolSeries interface {
Last() bool // newest element
Index(i int) bool // i >= 0, query bool data in reverse order using i as index
Length() int // length of data piped in array
}
Series were used almost everywhere in indicators to return the calculated numeric results, but the use of BoolSeries is quite limited. At this moment, we only use BoolSeries to check if some condition is fulfilled at some timepoint. For example, in CrossOver
and CrossUnder
functions if Last()
returns true, then there might be a cross event happened on the curves at the moment.
The calculation could either be done during invoke time (lazy init, for example), or pre-calculated everytime when event happens(ex: kline close). If it's done during invoke time and the computation is CPU intensive, better to cache the result somewhere inside the struct. Also remember to always implement the Series interface on indicator's struct pointer, so that access to the indicator would always point to the same memory space.
We recommend developers to add the following line inside the indicator source:
var _ types.Series = &INDICATOR_TYPENAME{}
// Change INDICATOR_TYPENAME to the struct name that implements types.Series
and if any of the method in the interface not been implemented, this would generate compile time error messages.
Instead of simple Series interface, we have types.SeriesExtend
interface that enriches the functionality of types.Series
. An indicator struct could simply be extended to types.SeriesExtend
type by embedding anonymous struct types.SeriesBase
, and instanced by types.NewSeries()
function. The types.SeriesExtend
interface binds commonly used functions, such as Add
, Reverse
, Shift
, Covariance
and Entropy
, to the original types.Series
object. Please check pkg/types/seriesbase_imp.go for the extendable functions.
Example:
a := types.NewQueue(3) // types.Queue is a SeriesExtend container type that holds limit number of floats
a.Update(100.)
a.Update(200.)
a.Update(300.)
assert.Equal(t, a.Sum(3), 600.)
Avoid using floats.Slice
to hold unlimited number of floats, unless you clean up the memory regularly. Manipulate large array of numbers will give huge impact on the computation speed due to long malloc/dealloc time.