Skip to main content

Python Client

Promote uses Python's Built-in Modules or Standard Library.

Important

We recommend that you build Python models using version 3.9 to ensure compatibility with Promote. If you upgrade from a previous installation of Promote (for example, 2020.4.0), you might need to update your models and their dependencies to work with this Python version. If you have questions about Python versions, contact Support.

Note

Visit our GitHub repository for technical information about how to deploy models with the Python client.

There, you can also find specialized instructions for how to deploy models using the H2O Python module.

Structure Your Directory

To correctly transfer all model dependencies to Promote, you have to structure your directory in a particular way. Create a folder that contains these files:

promote.sh

(optional)

This file executes before Promote builds your model. Include low-level system packages here.

main.py

(required)

This file contains your primary model deployment script.

helpers/

(optional)

Include helper functions in this directory.

objects/

(optional)

Include data fields needed for model execution (for example, pickle files) here.

requirements.txt

(required)

This file contains a list of the packages you need to run your model. You must include the promote package here.

Set Authentication Credentials

You must include your Promote username, API key, and instance URL to deploy your model:

# Replace "USERNAME" with your Promote username. 
# Replace "APIKEY" with your API key. 
# Replace "PROMOTE_URL" with the URL of your instance.

p = promote.Promote(USERNAME, APIKEY, PROMOTE_URL)

Functions and Methods

Promote.metadata

Use the Promote.metadatamethod to assign related metadata to the model so the Promote platform can view it.

Avertissement

The Python client assigns metadata individually, with a maximum of six metadata values per model and with no more than 50 characters for each value.

Usage

promote.metadata["name"] = value

Example

promote.metadata["voting"] = ENSEMBLE.voting

Promote.deploy

Use the Promote.deploymethod to capture a model you've created as a function and any related files in the model directory, then send that information to the Promote platform.

Usage

promote.deploy(name, model, testdata, confirm=False, dry_run=False, verbose=0)

Arguments

name

(required)

Data type: string

Name the model you want to deploy to Promote.

model

(required)

Data type: string

Name the promoteModel() function you want to capture.

testdata

(required)

Data type: object

Provide a data object that the model can use to make predictions.

confirm

(optional)

Data type: boolean

Specify whether you want a confirmation prompt when you deploy your model.

Default: False

dry_run

(optional)

Data type: boolean

Specify whether you want Promote to determine if it can pickle the model. (This option doesn't deploy the model.)

Default: False

verbose

(optional)

Data type: integer

Specify how much information you want to log during deployment. Enter a value from 0 to 4, with 4 being the most information logged.

Default: 0

Example

promote.deploy("SimpleMarkovChain", MarkovChain, TESTDATA, confirm=True, dry_run=False, verbose=4)

Promote.predict

Use the Promote.predictmethod to send data to a deployed model and request a prediction.

Usage

promote.predict(model, data)

Arguments

model

(required)

Data type: string

Name the model you want to query.

data

(required)

Data type: dictionary or dataframe

Provide the data that the model needs to make a prediction.

Example

# Providing test data

TESTDATA = {'step': 344, 'TransMatrix': [[0.69, 0.24, 0, 0, 0, 0.06],
                                         [0, 0.49, 0.05, 0.46, 0, 0],
                                         [0.5, 0, 0, 0.5, 0, 0],
                                         [0.76, 0.24, 0, 0, 0, 0],
                                         [0, 0, 0, 0, 0, 0],
                                         [1, 0, 0, 0, 0, 0]], 'InitState': [[1, 0, 0, 0, 0, 0]]}

# After the model is online, sending data and receiving predictions

promote.predict("SimpleMarkovChain", TESTDATA)

@promote.validate_json

Use the @promote.validate_jsondecorator to validate data passed into your model. If the data fails the checks for validation, it doesn't pass to your model.

Note

This function requires the schema package to validate JSON data.

Usage

@promote.validate_json(Schema({'name': And(DATATYPE)}))

Example

import promote
from schema import Schema # https://pypi.python.org/pypi/Schema

@promote.validate_json(Schema({'step': int, 'TransMatrix': [[float, int]], 'InitState': [[float, int]]}))