Difficulty: Beginner
Estimated Time: 10 minutes

In this first scenario, you'll be introduced to FastScore components and example models to walk through the deployment process.

FastScore is comprised of docker containers customized to execute analytics and is agnostic to the language, compute environment, and data sources. See architecture here. FastScore executes each model or piece of code in an individual FastScore Engine that can be scaled on demand. FastScore Manage communicates to the underlying storage location (example: Git) of models and other assets required to deploy into a FastScore Engine.

There are 6 ways to interact with FastScore:

  1. Command Line Interface
  2. FastScore SDKs
  3. FastScore Dashboard
  4. FastScore API
  5. FastScore Composer
  6. FastScore Deploy (built using FastScore SDK)

In this scenario, we will provide one model in Python and the associated assets needed to score data using that model through the CLI.

The model in this scenario is a Gradient Boosting Machine model. The model will consume data about cars and predict the risk factor of each car on a range of -3 to 3, -3 being very risky and 3 being the least risky.

Let's get started!

Please refer to our documentation for more detailed information.

We just scored a gradient boosting machine model written in python in batch mode and then quickly switched it to streaming.

We also went over all of the critical components needed to score a model in FastScore. The scoring happened in a FastScore engine customized for that model which can be saved as an image and referenced using any orchestration tool on any infrastructure that supports docker.

To set up FastScore on your own, please see our Getting Started Guide!

To see a deeper explanation of the model set up for FastScore, please see our GBM tutorial!

Don’t stop now! The next scenario will only take about 10 minutes to complete.

Introduction to FastScore

Step 1 of 5

Let's Look at the Scoring Models

Models are created in a variety of creation environments like R Studio or Jupyter Notebooks. Those models most likely were trained and created in a separate environment than they will run in production. FastScore enables data science teams to test, deploy, and monitor their models in a way that is portable and scalable. This scenario will introduce all of the components needed to deploy a model using FastScore after it has been created.

Let's start with the python model we will be using in this example.

This model is stored in a MySQL database which is the default backing store configured for FastScore. It could just as easily be stored in a code repository that FastScore Manage connects to and exposes the model to be used in a deployment configuration.

This model is specifically a scoring model. It uses the weights created during the training phase to predict a score off of the attributions in the incoming data.

Let's take a look at our model:

fastscore model show gbm_python

There are a few things to notice here:

  1. The model references data schemas at the top using smart comments called "gbm_input" and "gbm_output", which we will discuss in the next step
  2. The begin function performs preparation work for the execution of the model which will load any custom libraries and/or prepare the model coefficients (or weights) to be used
  3. The action function uses a specified method to calculate a score