Skip to content

Latest commit

 

History

History
58 lines (46 loc) · 2.49 KB

README.md

File metadata and controls

58 lines (46 loc) · 2.49 KB
Raw

BigMLFlow Examples

The examples in this directory illustrate how you can use MLFlow to generate BigML models, log evaluation metrics and deploy the different Supervised Models available in the BigML platform.

  • bigml_train/train.py will create a Decision Tree in BigML and evaluate it. The data is extracted from the Diabetes dataset, that has been published in the https://static.bigml.com/csv/diabetes.csv URL. The accuracy, precision and recall will be logged and printed and the model will be registered for further deploy. Please, check the prerequisites section to ensure that the environment is correctly set up before running the script.
  • bigml_logistic/register.py will register a BigML model previously stored in the bigml_logistic/logistic_regression.json file, that can be deployed using MLFlow to produce classification predictions.
  • bigml_ensemble/register.py will register a BigML model previously stored in the bigml_ensemble/ensemble.json file and its inner models, that can be deployed using MLFlow to produce classification predictions.
  • bigml_linear/register.py will register a linear regression model previously stored in the bigml_linear/linear_regression.json file, that can be deployed using MLFlow to produce regression predictions.

Prerequisites

Installing this bigmlflow library should install both mlflow and bigml bindings, which are the only prerequisites.

NOTE: In order to run the bigml_train/train.py script, you need an active account in BigML and your credentials (username and API key) should be stored as environment variables https://bigml.readthedocs.io/en/latest/index.html?highlight=authentication#authentication to be able to connect and use the service. You can sign in BigML at https://bigml.com/account/register and reproduce these examples using a free account.

How to run the examples

Run them as regular Python scripts.

python bigml_train/train.py {conf}
python bigml_logistic/register.py
python bigml_linear/register.py
python bigml_ensemble/register.py

The {conf} optional argument for the bigml_train/train.py script accepts a JSON string describing the configuration arguments for the model creation. E.g:

python bigml_train/train.py "{\"node_threshold\": 3}"

will create a 3-node Decision Tree.