Python models

To create a python model, following steps are advised:

Define model input data

Variables consumed by model must be defined in csv format

model.csv

true;x1;STRING;;	
true;x2;DOUBLE;;
true;x3;INTEGER;;
true;x4;MAP;;
false;x5;VECTOR;;
true;x6;BOOLEAN;;sample description

Each line contains information about one variable including

• field required indicator

• variable name

• variable type

• list of enum values

• optional description

Alternatively, python script to create file can be used

# Create .csv file with SCE parameters.
vars_def = """
true;x1;STRING;;	
true;x2;DOUBLE;;
true;x3;INTEGER;;
true;x4;MAP;;
false;x5;VECTOR;;
true;x6;BOOLEAN;;sample description
"""

model_file = open('./model.csv', 'wt')
model_file.write(vars_def)
model_file.close()

Serialize model

Model object must be saved into pickle file. Minimally the output structure of the model must be defined. Common machine learning libraries for instance sklearn , xgboostare supproted as well.

import pickle

model_object = {
    'SCORE_RAW': 100,
    'SCORE_CUT': 100,
    'SCORE_RANK': 100,
    'SCORE_NORMAL': 100
}
with open('./model.pkl', 'wb') as model_file:
    pickle.dump(model_object, model_file)
model_file.close()

Prepare scoring script

Usually some data processing is needed. Thus in python script such operations can be prepared. Scoring .py script is obligatory and model object must be always returned.

model.py

import pandas as pd
import random

# variables are automatically loaded
# Load map variables to Dataframe
df = pd.DataFrame(x4['table'])

# some data processing can be added here
# df  = transformations(df)

model_object = {
    'SCORE_RAW': 100,
    'SCORE_CUT': 100,
    'SCORE_RANK': 100,
    'SCORE_NORMAL': 100
}

# another processing transformations
# for example replace scores with random value
for key in model_object:
    model_object[key] = random.randint(1, 100)

# model object need to be retuned
return model_object

Upload model to Scoring One

Previously creaded 3 files: model.csv, model.plk, model.py must be packed together into one zip file to be able to send to Scoring One.

import zipfile

# Create .zip file to send it to SCE.
zf = zipfile.ZipFile(r"./model.zip", "w")
zf.write(r'./model.csv')
zf.write(r'./model.py')
zf.write(r'./model.pkl')
zf.close()

Then set up enviromental variables before sending request

api_environment_url_str = 'https://demo.scoring.one'
api_environment_key_str = '<your_key>'
scoring_code_name_str = 'model_python'

Finally use requests library to POST the data

import datetime
import requests

date_processing_str = datetime.datetime.now().strftime('%Y-%m-%d')
api_url_str = "{}/api/python/code/create".format(api_environment_url_str)
headers_dict = {'Authorization': 'Bearer {}'.format(api_environment_key_str)}
params_dict = {
    'dateFrom': date_processing_str,
    'dateTo': date_processing_str,
    'fileName': 'model.zip',
    'scoringCodeName': scoring_code_name_str,
    'charset': 'UTF-8'
}

fileobj = open(r'./model.zip', 'rb')
response = requests.post(api_url_str,
                         params = params_dict,
                         files={"file": ("./model.zip", fileobj)},
                         headers=headers_dict)
print(response.text)
fileobj.close()