Portability	non-portable
Stability	experimental
Safe Haskell	None

TimeSeries.Routing

Contents

Types
Looping actions
Initial values
Pretty printer

Description

Data routing for time series analysis.

Synopsis

Types

data Config Source

Configuration values for detecting correlations from uncooperative time series data.

This data type shall relate to bootstrapping in future, but at the moment, nothing related.

Constructors

Config

Fields

sketchSize :: Int: Number of random elements in sketch.
sketchGroups :: Word8: Number of subsequences for grouping sketch.
corrCutoff :: Double: Cutoff value between of correlation, between 0 to 1.
swSize :: Word64: Size of sliding window.
bwSize :: Word64: Size of basic window.
randSeed :: Integer: Seed for random vectors.
impls :: Implementation: Implementation.
verbose :: Bool: To show trace output or not.

Instances

Eq Config
Show Config

type BasicWindows = IntMap (Seq Window)Source

Representing basic windows as IntMap of Seq of Windows.

Key of outer IntMap is ID for concurrent input stream, number of keys matches to number of concurrent input stream. Inner Seq is indexed basic window within sliding window, number of elements matches to nb, where `nb = sw bw`, sw/ is sliding window size and bw is basic window size.

data Summary Source

Summary of sliding window.

We don't need to maintain whole basic windows when sketch implementation is the only concern. To show correlation values with direct function, preserving the whole basic window contents. From "3.5 The Issues in Implementation section":

... We need to maintain

∑i=0,nb-1(sum(Xbwⁱ))
∑i=0,nb-1(sum((Xbwⁱ)²))

for a sliding window ...

Constructors

Summary

Fields

smrSum :: Double: Sum of elements in sliding window. In other words, ∑i=0,nb-1(sum(Xbwⁱ)).
smrSumSq :: Double: Sum of squared elements in sliding window. In other words, ∑i=0,nb-1(sum((Xbwⁱ)²).
smrSketch :: Window: Sketch vector, size of this window matches to sketch size.
smrPreSketches :: [Window]: Pre-sketches. "Pre" means that those sketch values before applying dot product with control vector.
smrBasicWindows :: Seq BWSummary: Summaries for basic windows, length of Seq is nb, where nb = sliding_window_size / basic_window_size.

Instances

Eq Summary
Show Summary

data BWSummary Source

Summary of basic window.

From "3.5 The Issues in Implementation section":

... We need to maintain

Xbwⁱ⋅R
sum(Xbwⁱ)
sum((Xbwⁱ)²)

for each basic window.

Constructors

BWSummary
Fields bwSketch :: Window Sketch of basic window, currently not in use. bwsSum :: Double Sum of elements in basic window. bwsSumSq :: Double Sum of squared elements in basic window.

Instances

Eq BWSummary
Show BWSummary

data SysState Source

State stored in analysis system.

Later this shall relate to memory data in hardware implementation, but at the moment its 100% Haskell data type and values.

Constructors

SysState
Fields sysBasicWindows :: BasicWindows sysRandomVectors :: IntMap RandomVector sysAppendCount :: Word64 sysCurrentTime :: Word64 sysNumberOfBasicWindows :: Word64 sysIsReadyToTell :: Bool sysSummaries :: IntMap Summary sysConfig :: Config

Instances

Eq SysState
Show SysState
MonadState SysState Loop

data CorrResult Source

Result of correlation analysis.

Inspired from output found in "statStream". See:

http://cs.nyu.edu/cs/faculty/shasha/papers/statstream/doc.html

Constructors

CorrResult
Fields crIndexA :: Int Index of time series input A. crIndexB :: Int Index of time series input B. crStartTime :: Word64 Start time of base window. crEndTime :: Word64 `crTimeStart` + (size of base window) crValue :: Double A value between -1 to 1.

Instances

Eq CorrResult
Show CorrResult
MonadWriter [Either String CorrResult] Loop

data Implementation Source

Implementations.

As for a prototype, used to analyse and compre resulting values for different implementations.

Constructors

Direct	Direct correlation computation.
Sketch	Sketch based computation.

Instances

Eq Implementation
Read Implementation
Show Implementation

newtype Loop a Source

Loop for analysing correlation of input data and updating analysis results and windowed data.

Constructors

Loop
Fields unLoop :: StateT SysState (Writer [Either String CorrResult]) a

Instances

Monad Loop
Functor Loop
MonadState SysState Loop
MonadWriter [Either String CorrResult] Loop

Looping actions

Looping with State and Writer.

loop Source

Arguments

:: Handle	Where to show results.
-> Config	Configuration values used for computation.
-> [[Double]]	Input values.
-> IO ()

The main loop of time series analysis.

Specifying number of concurrent time series data in this function. In hardware implementation, this may relates to fixed value, which possibly been configured at the time of code generation.

runLoop :: Loop a -> SysState -> ((a, SysState), [Either String CorrResult])Source

Unwrap Loop, run internal State and Writer.

step :: Config -> [Double] -> Loop ()Source

Single step to take with new input data.

Handle state management and if found any, report analysis result.

filterImpl Source

Arguments

:: Implementation	Which result to choose.
-> Double	Cutoff value.
-> (CorrResult, CorrResult)	(Direct correlation result, sketch correlation result).
-> [CorrResult]

Filter results with implementation and cutoff value.

corrPermute Source

Arguments

:: Word64	System current time. ... What is the time format used in hardware implementation?
-> Word64	Basic window size.
-> Word64	Sliding window size.
-> BasicWindows	Basic windows.
-> IntMap Summary	`Summary` for each input stream.
-> [(CorrResult, CorrResult)]	Direct result and sketch result.

Compute correlations.

slidingWindow :: Word64 -> BasicWindows -> IntMap Window Source

Struct sliding windows from basic windows.

This function is not needed in hardware implementation, should used for direct calculation only.

updateStates :: [Double] -> Loop ()Source

Adding new elements to windows, removing old elements.

consBW Source

Arguments

:: Seq Window	Singleton `Seq` of singleton `Window`.
-> Seq Window	Basic windows in system status.
-> Seq Window

Append first argument to second argument.

updateSummariesAndShift :: Loop ()Source

Update sum of elements in sliding window, and sum of squared elements in sliding window, and shift contents of basic windows.

rotateSeq :: a -> Seq a -> Seq aSource

Cons given element to given Seq and remove last element in Seq.

Initial values

initialSysState :: Config -> Int -> SysState Source

Initial state.

Other than Config, number of concurrent time series input data is passed.

initialSlidingWindows :: Size -> Int -> IntMap Window Source

initialBasicWindows :: Size -> Size -> Int -> BasicWindows Source

initialSummaries Source

Arguments

:: Config	Main configuration.
-> Int	Number of concurrent input streams.
-> IntMap Summary

initialBWSummary Source

Arguments

:: Size	Sketch size.
-> BWSummary

Initial basic window summary data.

randomVectors Source

Arguments

:: Integer	Random seed.
-> Size	Basic window size.
-> Size	Sliding window size.
-> Int	Sketch size.
-> IntMap RandomVector

Initial random vectors.

Pretty printer

Actually, not much pretty.

simpleCorrResult :: CorrResult -> String Source

Simplified string representation of CorrResult.

>>> simpleCorrResult (CorrResult 1 2 10 20 0.5)
"1, 2, 10, 20, 0.5"