-
Notifications
You must be signed in to change notification settings - Fork 4
Library
Tracking module is used to track machine learning module during the process of their creation, training and evaluation. It allows users to store the most important information about the model (model name, dataset, parameters etc.) and later displays the information in MLOps App to provide insight.
An MLOps Project is a single machine learning project that consists of multiple experiments and models run as iterations
Function creates a project based on the unique title.
Arguments:
-
title: string
Title of the created project
-
description: string, optional
Description of the created project.
-
status: string, optional
Status of the created project
-
archived: bool, optional
Archived status of the created project
Returns:
-
project: dictionary
JSON data of the created project
Function retrieves an existing project from MLOps App
Arguments:
-
project_id:
Id od the desired project, that will be retrieved from MLOps app
Returns:
-
project: dictionary
JSON data of the project
MLOps experiment is a machine learning experiment that can contain many iterations
Function retrieves an experiment from MLOps App
Arguments:
-
experiment_id: string
Id of the experiment, that will be retrieved from MLOps app
-
project_id: string, optional
Id of the project, that the experiment comes from. By default value is the active project
Returns:
-
experiment: dictionary
JSON data of the experiment
Function creates a new experiment
Arguments:
-
name: string
Name of the created experiment
-
description: string, optional
Description of the created experiment
-
project_id: string, optional
Id of the project, that the experiment comes from
Returns:
-
experiment: dictionary
JSON data of the created experiment
Function creates new mlops dataset
Arguments
-
dataset_name: name of the created dataset
-
path_to_dataset: path to dataset files
-
dataset_description: short description of the dataset displayed in the app
-
tags: tags for dataset
-
version: version of the dataset
Returns:
- dataset: json data of created dataset
MLOps Iterations contain informations of a single machine learning model run
Function creates an instance of Iteration
Arguments:
-
iteration_name: string
name of the created iteration
-
project_id: string, optional
Id of the target project. By default value is the id of the active project
-
experiment_id: string, optional
Id of the target experiment. By default value is the id of the active experiment
-
send_email: bool, optional
If true email will be sent to email address specified in library settings. False by default.
Returns:
- iteration dictionary JSON data of the created iteration
Function logs the model name in the currently running iteration.
Arguments:
-
model_name: string
Name of the model that's being tracked
Function logs the path to model file
Arguments:
-
path_to_model: string
Path to the file containing the tracked model
Function logs a single metric along with it's value
Arguments:
-
metric_name: string
Name of the logged metric
-
value:
Value of the logged metric
Function logs multiple metrics at once
Arguments:
-
metrics: dictionary
Dictionary containing metric: value pairs that are going to be logged
Function logs a single parameter along with it's value
Arguments:
-
parameter_name:
Name of the logged parameter
-
value:
Value of the logged parameter
Function logs multiple parameters at once
Arguments:
-
parameters: dictionary
Dictionary containing parameter: value pairs that are going to be logged
Function logs an existing dataset with an iteration.
Arguments:
-
dataset_id: string
Id of an existing dataset in webapp
Function ends the iteration and sends the logged data to the MLOps App
Returns:
-
iteration: dictionary
JSON data of created iteration
Monitoring module provides the user with an interface to use the MLOps app to monitor running machine learning model performance and store model data within the mlops app. It's the second big part of the MLOps app designed to be used when the training process has been finished and an ml model is ready to be used to make predictions with real data.
Monitored models are mlops app objects that all of the monitoring module is based around. They represent a trained ml model that is ready to be used to make predictions. MLOps library provides the user with functionality to create models, download ml model information from the app and to make predictions using Pandas dataframes and model registered in the MLOps app
Function creates monitored model within MLOps app
Arguments:
-
model_name: string
Unique name of the created name
-
model_description: string, optional
Description of monitored model
-
iteration_dict: dictionary, optional
Dictionary containing valid iteration data with a path to model. It is recommended to use the result of the iteration.end_iteration method
Returns:
-
monitored_model: string
Json data of monitored model
Function for retrieving mlops monitored model from database
Arguments:
-
model_name: string
Unique name of the monitored model to be retrieved
Returns:
-
monitored_model: dictionary
Json data of monitored model
Function for setting active model from monitored models
Arguments:
-
model_name: string
Name of monitored model, that will be set as active
Returns:
-
result: string
Information about new active model setup
Function to invoke a prediction from monitored model. Function accepts a Pandas dataframe, where every record is taken as a separate prediction.
Arguments:
-
model_name: string
Name of monitored model that will be used in prediction
-
data: Pandas Dataframe
Pandas Dataframe containing data for prediction. Each row of data in the dataframe is used for a separate prediction
-
send_email: bool, optional
If true email will be sent to email address specified in library settings. False by default.
Returns:
-
prediction: list of dictionaries
List of dictionaries containing results for each executed prediction
Tracking module contains local settings that can specify active project and experiment
Function sets the active project to given project id of an existing MLOps project
Arguments:
-
project_id: string
Id of the project, that will be set as active
Returns:
-
result: string
Message informing about the new active project
Function sets the active experiment to given experiment id of an existing MLOps experiment
Arguments:
-
experiment_id: string
Id of the experiment, that will be set as active
Returns:
-
result: string
Message informing about the new active experiment