AIModelShare

Model Playground Class

Deployed models receive an individual Model Playground. Each Model Playground uses a personal REST API to provide a fully functional prediction dashboard that allows end-users to input text, tabular, or image data and receive live predictions.

Moreover, users can build on Model Playgrounds by:

  1. Creating ML model competitions

  2. Uploading Jupyter notebooks to share code

  3. Sharing model architectures, and

  4. Sharing data… with all shared artifacts automatically creating a data science user portfolio.

deploy

Launches a live prediction REST API for deploying ML models using model parameters and user credentials.

ModelPlayground.deploy(model_filepath, preprocessor_filepath, y_train, example_data=None, custom_libraries='FALSE', image='', reproducibility_env_filepath=None, memory=None, timeout=None)
Parameters:

  • model_filepath (string) – Absolute path to model file. .onnx is the only accepted model file extension. “example_model.onnx” filename for file in directory. “/User/xyz/model/example_model.onnx” absolute path to model file from local directory

  • preprocessor_filepath (string) – absolute path to preprocessor file. [REQUIRED] to be set by the user. Should be: “./preprocessor.zip”. searches for an exported zip preprocessor file in the current directory. File is generated using export_preprocessor function from the AI Modelshare library

  • y_train (pandas dataframe of one hot encoded y train data for classification, or list of values for regression) – training labels for classification models. [REQUIRED] for classification type models

  • custom_libraries (string) – “TRUE” if user wants to load custom Python libraries to their prediction runtime. “FALSE” if user wishes to use AI Model Share base libraries including latest versions of most common ML libs.

  • example_data (String) – Tabular data - pandas DataFrame in same structure expected by preprocessor function. Other data types - absolute path to folder containing example data (first five files with relevant file extensions will be accepted). REQUIRED for tabular data

  • reproducibility_env_filepath – [OPTIONAL] to be set by the user- absolute path to environment environment json file. Example: “./reproducibility.json”. File is generated using export_reproducibility_env function from the AI Modelshare library

  • memory (Int) – The amount of memory (in megabytes) to be allocated to the lambda function (default = 1024).

  • timeout (Int) – Number of seconds before the lambda times out (default = 30).

Returns:

Prints statements with generated live prediction API details also prints steps to update the model submissions by the user/team

Example

from aimodelshare import ModelPlayground

#Instantiate ModelPlayground() Class as placeholder
myplayground=ModelPlayground(model_type="image", classification=True, private=True)

# Create Model Playground (generates live rest api and web-app for your model/preprocessor)
myplayground.deploy(model_filepath = "model.onnx", preprocessor_filepath = "preprocessor.zip", y_train_labels, exampledata)

Example

# To instantiate a Model Playground that already exists:

myplayground=ModelPlayground(playground_url = "https://exampleapiurl.execute-api.us-east-1.amazonaws.com/prod/m")

#You can find your Model Playground url under the "Programmatic" sub tab of the "Predict" page

create_competition

Creates a model competition for a deployed prediction REST API

ModelPlayground.create_competition(data_directory, y_test, eval_metric_filepath=None, email_list=[], public=False)
Parameters:

  • y_test (list) – y values for test data used to generate metrics from predicted values from X test data submitted via the submit_model() function. [REQUIRED] to generate eval metrics in competition leaderboard

  • data_directory (string) – Path to folder storing training data and test data (excluding Y test data)

  • eval_metric_filepath (string) – Filepath pointing to a zipfile containing custom evaluation functions exported with export_eval_metric()

  • email_list (List of strings) – [OPTIONAL] list of comma separated emails for users who are allowed to submit models to competition.

  • public (Boolean) – [REQUIRED] True/false. Defaults to False. If True, competition is public and ANY AIModelShare user can submit models. Use with caution because one model and one preprocessor file will be be saved to your AWS S3 folder for each model submission.

Returns:

Information about how to submit models to competition

Example

# Create Competition
# Note -- Make competition public (allow any AI Model Share user to submit models)
# .... by excluding the email_list argument and including the 'public=True' argument
myplayground.create_competition(data_directory='my_competition_data_folder',
                        y_test = y_test_labels,
                        email_list=emaillist)
                        public=True)

update_runtime_model

Updates the prediction API behind the Model Playground with a new model from the leaderboard and verifies Model Playground performance metrics.

update_runtime_model(model_version=None)
Parameters:

model_version (integer) – model version number from competition leaderboard

Returns:

Success message when the model and preprocessor are updated successfully.

Example

myplayground.update_runtime_model(model_version=1)

delete_deployment

Delete all components of a Model Playground, including: AWS s3 bucket & contents, attached competitions, prediction REST API, and interactive Model Playground web dashboard.

ModelPlayground.delete_deployment(playground_url=None)
Parameters:

playground_url (string) – API URL that the user wishes to delete. WARNING: User must own an API in order to delete it.

Returns:

Success message when deployment is deleted.

Example

myplayground.delete_deployment()