Skip to content

Latest commit

 

History

History
110 lines (66 loc) · 5.02 KB

observable.rst

File metadata and controls

110 lines (66 loc) · 5.02 KB
Raw

Events, functions and observables

Events

Events in the simultator are conceptually close to events in C# or boost::signal in C++. Number of listeners can be subscribed for event notification using += operator and unsubscribed with -= operator. A listener should be a callable object that accepts a single argument - the source of the event (it allows not to store an event source in the listener and also use a single listener for several event sources at the same time).

Event subscription

Since event subscription is often done at binding time and unadvising should be done at dispose time event.subscribe helper function was introduced to manage event subscription declaratively:

class ProgressPrinter(object):

  def __init__(self):
    event.subscribe(event.Every(ops.constant(10)), # event source to subscribe to
                    _(self)._wakeUp,               # event listener
                    self)

  def _wakeUp(self, _):
    print '.'   # prints . every 10 units of time

Event helpers

event.After and event.Every are used to represent a single and a repeating action

event.Array is used when several event sources are to be represented as a single event. For example,

event.Array([BestPrice(Asks(orderbook)), BestPrice(Bids(orderbook))])

fires an event once bid or ask of orderbook has changed.

Filtered events

Sometimes an event source (for example, price of an asset) has hundreds subscribers who check is the event source is less or greater than some value. In order to avoid excessive calls event.LessThan and event.GreaterThan classes are introduced. Event sources detect objects of these types and store them sorted in order to lookup relevant listeners very quickly. For example,

observable.MidPrice(orderbook) += event.LessThan(100, listener)

will call listener only if mid-price of the asset is less than 100.

Functions and observables

Function
Object with overloaded __call__ operator that accepts no arguments and returns a value of type T. Further we will refer to the type of a function as () -> T or IFunction.
Observable
Function notifying listeners about its value changes. Only piecewise constant functions should be considered as observables. An arbitrary function can transformed into observable by means of observable.OnEveryDt(dt, func).

Conceptually observable might be defined as

class IObservable[T] (IFunction[T], IEvent): ...

Basic observables

  • Constant[T] / None[T] represent typed constant value or None value

  • Arithmetic operations (Sum, Sub, Product, Div, Mod). If both arguments of the operation are observable the operation fires events about its value change and thus becomes an event itself. Functions overload operations +, -, *, -, % so they construct respective objects. For example,

    (observable.AskPrice(orderbook) + observable.BidPrice(orderbook)) / 2

    creates an observable that notifies about every change of the mid-price.

  • Logic operations (Equal[T], NotEqual[T], Less[T], Greater[T], GreaterEqual[T], LessEqual[T]). If both arguments of the operation are observable the operation fires events about its value change and thus becomes an event itself. Functions overload operations ==, !=, >, <, <= so they construct respective objects. These operations inherit from IFunction[bool]/IObservable[bool] and have overloaded indexing operator that constructs Condition[T] object and it allows to write if-then-else expressions like:

    # randomly chooses either Side.Buy or Side.Sell
(math.random.uniform(0.,1.) < 0.5)[ ops.constant(Side.Buy), ops.constant(Side.Sell) ]

  • Functions from standard math module: Atan, Pow etc

  • Random distributions from random module: uniform, lognormvariate, expovariate etc.

  • Derivative of a differentiable function

  • Lagged: returns function values with some lag

Max(x,y) ::= if x > y then x else y
Min(x,y) ::= if x < y then x else y
Sqr(x) ::= x*x
  • Quotes: downloads external historical data

There are also statistics related functions, functions of orderbooks and of traders.

TODO: functions and observables with identical parameters should share their state.

TODO: true topological sort should be done for event propagation in observables.