Bulk Calculations in Python Models

By default, Python Models have a single 'calculation' function that is run on every 'sample' of data.

However, ML models in particular work much better working on bulk data. For this reason, the framework also includes the option of using a bulk calculation function that works on Pandas dataframes rather than individual data dictionaries.

This example uses an AI model.

import os
import ardifunctions
import pandas as pd
import numpy as np
import json
 
defaults = {}
 
dfunc = ardifunctions.DynamicFunction("http://localhost/s/default")
dfunc.query = "'Machine' ASSET AIPOINTS"
dfunc.Defaults(defaults)
dfunc.ParseParameters()
 
import ardiai
import tensorflow as tf
 
#Load the AI Models
mymodel = ardiai.AIData(os.path.dirname(__file__) + "/aidata.json")
mymodel.LoadConfig(os.path.dirname(__file__) + "/model.json")
 
tflow = tf.keras.models.load_model(os.path.dirname(__file__) + '/model.keras')
 
#Register all the variable names, so we know what to load from the UI
varnames = []
for c in dtatop.icolumns:    
    varnames.append(c.name)
 
dfunc.variables = varnames
 
def SingleSearch(factors,results):
    if factors['Running'] == 0:
        return {"AI Output 1": 0,"AI Output 2": 0}
 
    try:
       inp = dtatop.DictToInput(factors)   
       outputs = topmodel.predict(np.array([inp]),verbose=False)    
       modeldata = dtatop.OutputDict(outputs[0])
    except:
       return {"AI Output 1": 0,"AI Output 2": 0}
 
    return modeldata
 
def MultiSearch(frame,context):
 
    inputs,indexes = mymodel.FrameToInputs(frame,training=False)
 
    outputs = tflow.predict(inputs,verbose=False)
    for v in indexes:
        if indexes[v] is None:
            context.Write(v,{"AI Output 1": 0,"AI Output 2": 0})
        else:
            resp = dtatop.OutputDict(outputs[indexes[v]])
            context.Write(v,resp)            
 
dfunc.SingleCalc = SingleSearch
dfunc.BulkCalc = MultiSearch
 
dfunc.Execute()