README
¶
This sample workflow demos a long iterative math optimization process using particle swarm optimization (PSO).
The workflow first does some data structure initialization and then runs many iterations using a child workflow. The child workflow runs 10 iterations and then uses ContinueAsNew to avoid to store too long history in the Temporal database. In case of recovery the whole history has to be replayed to reconstruct the workflow state. So if history is too large the recover can take very long time.
Each particle is processed in parallel using worflow.Go and the math grunt work is done in the activites.
Since the data structure that maintains the optimization state has to be passed to the child workflow and the activities, a custom DataConverter has been implemented to take care of serialization/deserialization.
Also the query API is supported to get the current state of running workflow.
Steps to run this sample:
- You need a Temporal service running. See details in README.md
- Run the following command multiple times on different console window. This is to simulate running workers on multiple different machines.
go run pso/worker/main.go
- Run the following command to submit a start request for this PSO workflow.
go run pso/starter/main.go
- Query the call stack for the workflow with
go run pso/query/main.go -w <workflow_id from step 3> -r <run_id from step 3>
Documentation
¶
Index ¶
- Constants
- Variables
- func CalculateSwarmSize(dim, max_size int) int
- func EvalGriewank(vec []float64) float64
- func EvalRosenbrock(vec []float64) float64
- func EvalSphere(vec []float64) float64
- func NewJSONDataConverter() converter.DataConverter
- func PSOWorkflow(ctx workflow.Context, functionName string) (string, error)
- type ObjectiveFunction
- type Particle
- type ParticleResult
- type Position
- type Swarm
- type SwarmSettings
- type Vector
- type WorkflowResult
Constants ¶
const ( InitParticleActivityName = "initParticleActivityName" UpdateParticleActivityName = "updateParticleActivityName" )
*
- Sample activities used by file processing sample workflow.
const ContinueAsNewStr = "CONTINUEASNEW"
Variables ¶
var ActivityOptions = workflow.ActivityOptions{ ScheduleToStartTimeout: time.Second * 5, StartToCloseTimeout: time.Minute * 10, HeartbeatTimeout: time.Second * 2, RetryPolicy: &temporal.RetryPolicy{ InitialInterval: time.Second, BackoffCoefficient: 2.0, MaximumInterval: time.Minute, MaximumAttempts: 5, }, }
ActivityOptions can be reused
var Griewank = ObjectiveFunction{ Goal: 1e-5, Evaluate: EvalGriewank, // contains filtered or unexported fields }
var Rosenbrock = ObjectiveFunction{ Goal: 1e-5, Evaluate: EvalRosenbrock, // contains filtered or unexported fields }
var Sphere = ObjectiveFunction{ Goal: 1e-5, Evaluate: EvalSphere, // contains filtered or unexported fields }
Functions ¶
func CalculateSwarmSize ¶
func EvalGriewank ¶
func EvalRosenbrock ¶
func EvalSphere ¶
func NewJSONDataConverter ¶
func NewJSONDataConverter() converter.DataConverter
NewJSONDataConverter creates a json data converter
Types ¶
type ObjectiveFunction ¶
type ObjectiveFunction struct {
Goal float64 // optimization goal (error threshold)
Evaluate func(vec []float64) float64 // the objective function
// contains filtered or unexported fields
}
func FunctionFactory ¶
func FunctionFactory(functionName string) ObjectiveFunction
type Particle ¶
func InitParticleActivity ¶
func NewParticle ¶
func UpdateParticleActivity ¶
func (*Particle) UpdateFitness ¶
func (*Particle) UpdateLocation ¶
type ParticleResult ¶
type Position ¶
func NewPosition ¶
func RandomPosition ¶
func RandomPosition(function ObjectiveFunction, rng *rand.Rand) *Position
func (*Position) IsBetterThan ¶
type Swarm ¶
type Swarm struct {
Settings *SwarmSettings
Gbest *Position
Particles []*Particle
}
type SwarmSettings ¶
type SwarmSettings struct {
FunctionName string
// swarm size (number of particles)
Size int
// ... N steps (set to 0 for no output)
PrintEvery int
// Steps after issuing a ContinueAsNew, to reduce history size
ContinueAsNewEvery int
// maximum number of iterations
Steps int
// cognitive coefficient
C1 float64
// social coefficient
C2 float64
// max inertia weight value
InertiaMax float64
// min inertia weight value
InertiaMin float64
// whether to keep particle position within defined bounds (TRUE)
// or apply periodic boundary conditions (FALSE)
ClampPosition bool
Inertia float64 // current inertia weight value
// contains filtered or unexported fields
}
func PSODefaultSettings ¶
func PSODefaultSettings(functionName string) *SwarmSettings
type WorkflowResult ¶
type WorkflowResult struct {
Msg string // Uppercase the members otherwise serialization won't work!
Success bool
}
func PSOChildWorkflow ¶
PSOChildWorkflow workflow definition Returns true if the optimization has converged
Source Files
Directories