Skip to content

sigopt/sigopt-python

Repository files navigation

OpenSSF Scorecard pre-commit.ci status integration tests

SigOpt Python API

This is the SigOpt Python API client. Use this to natively call SigOpt API endpoints to create experiments and report data.

For more help getting started with SigOpt and Python, check out the docs.

Take a look in examples for example usage.

Getting Started

Install the sigopt python modules with pip install sigopt.

Sign up for an account at https://sigopt.com. In order to use the API, you'll need your API token from the API tokens page.

To call the API, instantiate a connection with your token.

Authentication

Authenticate each connection with your API token directly (will override any token set via environment variable):

from sigopt import Connection
conn = Connection(client_token=api_token)

Authentication with Environment Variable

Insert your API token into the environment variable SIGOPT_API_TOKEN, and instantiate a connection:

from sigopt import Connection
conn = Connection()

Issuing Requests

Then, you can use the connection to issue API requests. An example creating an experiment and running the optimization loop:

from sigopt import Connection
from sigopt.examples import franke_function
conn = Connection(client_token=SIGOPT_API_TOKEN)

experiment = conn.experiments().create(
  name='Franke Optimization',
  parameters=[
    dict(name='x', type='double', bounds=dict(min=0.0, max=1.0)),
    dict(name='y', type='double', bounds=dict(min=0.0, max=1.0)),
  ],
  metrics=[dict(name='f', objective='maximize')],
)
print("Created experiment: https://sigopt.com/experiment/" + experiment.id);

# Evaluate your model with the suggested parameter assignments
# Franke function - http://www.sfu.ca/~ssurjano/franke2d.html
def evaluate_model(assignments):
  return franke_function(assignments['x'], assignments['y'])

# Run the Optimization Loop between 10x - 20x the number of parameters
for _ in range(20):
  suggestion = conn.experiments(experiment.id).suggestions().create()
  value = evaluate_model(suggestion.assignments)
  conn.experiments(experiment.id).observations().create(
    suggestion=suggestion.id,
    values=[dict(name='f', value=value)],
  )

API Token

Your API token does not have permission to view or modify information about individual user accounts, so it is safe to include when running SigOpt in production.

Endpoints

Endpoints are grouped by objects on the Connection. For example, endpoints that interact with experiments are under conn.experiments. ENDPOINT_GROUP(ID) operates on a single object, while ENDPOINT_GROUP() will operate on multiple objects.

POST, GET, PUT and DELETE translate to the method calls create, fetch, update and delete. To retrieve an experiment, call conn.experiments(ID).fetch(). To create an experiment call conn.experiments(ID).create(). Parameters are passed to the API as named arguments.

Just like in the resource urls, suggestions and observations are under experiments. Access these objects with conn.experiments(ID).suggestions and conn.experiments(ID).observations. The REST endpoint POST /v1/experiments/1/suggestions then translates to conn.experiments(ID).suggestions().create().

Testing

To run the included tests, just run

pip install -r requirements-dev.txt
make test

To lint, install requirements (included in the previous step) and run

make lint

Use vulture to check no use code/paramters

make vulture

Generate vulture allowlist

make vulture-allowlist

The vulture allowlist file .vulture_allowlist can be edit to add/remove allowed no use code/parameters.

Earlier versions

Earlier versions supported a hyperopt integration. Since hyperopt now seems to be a retired project, we are removing this integration. If you need that integration please select version 8.8.3

Acknowledgments

We would like to thank the following people for their contributions:

  • @aadamson for their contributions in supporting custom requests.Session objects #170