This is a Scalarm Pathfinder supervisor. For more information see: https://github.com/Scalarm/scalarm_experiment_supervisor

Dependencies

This program requires following libraries placed in its root directory to run:

• Newtonsoft.Json.dll - version >= 6.0.8
• RestSharp.dll - version >= 105.0.1
• Scalarm.dll - can be found here: https://github.com/Scalarm/scalarm_client_csharp
• SensitivityAnalysis.dll - a proprietary sensitivity analysis library by Daniel Bachniak; not available publicly at the moment

Building

Program builds in .Net4.0 compiler - both Mono and VisualStudio, but only Mono is officially supported and tested. The easiest way is to open project file in IDE, satisfy dependencies (listed above) and build the project.

Execution

Please put all dependencies DLLs in compiled program root directory and launch Program.exe with Mono/.Net runtime. At least Mono 4.2 runtime is required to run. The program will not run under Mono 3.x!

Execution parameters

• mono Program.exe will read config from config.json in current directory
• mono Program.exe -config <config_path> will read config from <config_path> file
• mono Program.exe -stdin will read config from standard input (e.g. cat config.json | Program.exe on *NIX)

Configuration

Example configurations

Method of Morris

{
  "method_type": "morris",
  "morris_samples_count": 2,
  "morris_levels_count": 3,
  "experiment_id": "56b1b4f9f83f4910c0000261",
  "user": "f784b56e-e02c-4ede-9d85-fb3ed5566cc5",
  "password":" <ciach>",
  "parameters": [
    {
      "id":"parameter1",
      "min":0,
      "max":1000
    },
    {
      "id":"parameter2",
      "min":-100,
      "max":100
    }
  ],
  "address": "149.156.11.4:10053"
}

Method of Sobol

{
  "method_type": "sobol",
  "sobol_base_inputs_count": 1000,
  "experiment_id": "56b1b4f9f83f4910c0000261",
  "user": "f784b56e-e02c-4ede-9d85-fb3ed5566cc5",
  "password":" <ciach>",
  "parameters": [
    {
      "id":"parameter1",
      "min":0,
      "max":1000
    },
    {
      "id":"parameter2",
      "min":-100,
      "max":100
    }
  ],
  "address": "149.156.11.4:10053"
}

Configuration keys

• method_type: string, name of SA method to use:
  • morris, when using, set additional keys in config:
    • morris_samples_count: integer
    • morris_levels_count: integer
  • sobol, when using, set additional keys in config:
    • sobol_base_inputs_count: integer
• address: string, address of Scalarm Experiment Manager (without protocol provided)
• credentials - either BasicAuth login/password or X509 proxy certificate used to authenticate in Scalarm
  • BasicAuth
    • user: string, username
    • password: string, password
  • X509 Proxy certificate
    • experiment_manager_proxy_path: string, path to file in PEM format X509 Proxy certificate
• entity to supervise - can be either Scalarm SimulationScenario or Experiment; of course these entities must exists and credentials should allow to view or/and modify it in provided Scalarm instance
  • SimulationScenario - if provided, ID of SimulationScenario will be used to instantiate an Experiment to supervise with sensitivity analysis algorithm
    • simulation_id: string, ID of Scalarm SimulationScenario
  • Experiment - used only if SimulationScenario is not provided (used by Pathfinder)
    • experiment_id: string, ID of Scalarm Experiment to use
• parameters: array, an array of SimulationScenario parameters specifications
  • parameter object
    • id: string
    • type: string, currently not used - all parameters should be float
    • min: float, lower limit of parameter value range to search
    • max: float, upper limit of parameter value range to search
    • start_value: float, currently not used
• fake_experiment (optional): boolean, if true - do not use real Scalarm server, randomly generate results instead and write SA results to stdout

Output

Depending on the configuration, results will be send to Scalarm or only printed on stout (when fake_experiment: true).

Example outputs

Method of Sobol

{  
  "sensitivity_analysis_method":"sobol",
  "moes":{  
    "moe_first":{  
      "parameter1":{  
        "sensitivity_indices":0.1553690685569363,
        "total_effect_indices":0.2037052342725223
      },
      "parameter2":{  
        "sensitivity_indices":0.77595737530684716,
        "total_effect_indices":0.85337015856275833
      }
    },
    "moe_random":{  
      "parameter1":{  
        "sensitivity_indices":0.0059681671107936606,
        "total_effect_indices":1.0526041838912277
      },
      "parameter2":{  
        "sensitivity_indices":0.008428390408741087,
        "total_effect_indices":0.9496497682209919
      }
    }
  }
}

Method of Morris

Technical documentation

Some aspects of internal implementation. Not for regular users :)

Getting Scalarm Results

After computing simulation by Scalarm:

IList<Scalarm.SimulationParams> scalarmResults = experiment.GetResults()

This is an array of Scalarm.SimulationParams. For example single SimulationParams (some JSON-like pseudocode):

{
  Input: { // Scalarm.ValuesMap
    "parameter1": 1000.0,
    "paraneter2": 100.0
  },
  Output: { // Scalarm.ValuesMap
    "moe": 100000.0
  }
}