Distributed Analytics Service (DAS)

Distributed Analytics Service Overview

A-Stack Distributed Analytics Service (DAS) allows TQL developers to seamlessly integrate a number of python-based regression and classification algorithms from into their application.

 

DAS Functional Components

Component Name

Description

Examples

Component Name

Description

Examples

Pre-defined set of A-Stack Activities

A set of simple activities which collects commands from DASScript and executes in a python runtime as part of Exec Activity

 

DAS Configuraton Model

Configuration parameters for DAS

 

# scope(fid: dasact): $vopak_dasconfig = DASConfig: DASContextID: "SteamHist-1" ModelOutputHandler: facetID: "steamws" ActionName: "HandleModelOutput_Action" InputDataSource: TQLFacetID: steamws ColumnNames: "steamConsumedValue,deltaTemp,Valve_open_Temp" IndexNames: "deltaTemp" FindQuery: Find(as: "$none:$none:$none", Only: "steamConsumedValue, deltaTemp, Valve_open_Temp"): SteamConsHistory: SteamHistId(ne: "") DASScript: @: 'resources/SteamPredict/spaces/DASPolyline.das.cdm'

DAS Python SDK

Offers basic functions that can called from DASConfig’s DAS Script.

Functions:

  1. broadcastmsg

  2. savemodel

  3. readmodel

Currently Supports following libraries

import numpy as np
from numpy import nan
import datetime
from sklearn.ensemble import RandomForestRegressor
from sklearn.preprocessing import PolynomialFeatures
from sklearn.linear_model import LinearRegression
from sklearn.preprocessing import MinMaxScaler
from sklearn.pipeline import make_pipeline
import statsmodels.api as sm
from sklearn.metrics import r2_score
from sklearn.metrics import mean_squared_error
from sklearn.metrics import mean_absolute_error
import statsmodels.api as sm
from sklearn.model_selection import train_test_split
from collections import namedtuple
from sklearn import linear_model

 

 

DAS Features

DAS Features description

Feature Category

Feature Description

Example

Feature Category

Feature Description

Example

Feeding Training or Testing (To-be-Predicted) Data

InputDataSource - DAS Provides InputDataSource parameter as part of the DASConfig Parameter

 

InputDataSource: TQLFacetID: steamws ColumnNames: "steamConsumedValue,deltaTemp,Valve_open_Temp" IndexNames: "deltaTemp" FindQuery: Find(as: "$none:$none:$none", Only: "steamConsumedValue, deltaTemp, Valve_open_Temp"): SteamConsCurrent: SteamId(ne: "")

Training Data Preparation

  1. Data_Shape - DAS Provides Data_Shape Section as part of the DASConfig’s DASScript where TQL developers can add their Data Shaping code.

  2. Data_Split_TrainAndValidate - DAS Provides Data_Split_TrainAndValidate Section as part of the DASConfig’s DASScript where TQL developers can add further steps their Data Shaping code.

 

Providing Algorithm Details

PredictiveModels - DAS Provides PredictiveModels Section as part of the DASConfig’s DASScript where TQL developers can add specific algorithm steps

 

Handle Model Output

 

 

Saving Trained Model for Prediction re-use

 

 

Reading Saved model

 

 

Real-time viewing of algorithm execution logs

All broadcastmsg statements from DAS Script are published to DASLog websocket endpoint

 

 

Getting Started with DAS

Future Features

  1. Seamless stream ingestion into DAS

  2. Seamless stream egres to Python from DAS

  3. Scalable Data Preprocessing

  4. Analytics Model versioning

  5. Analytics library exposing functional interface to BML (python sdk)

  6. Host and run external models

Â