You're reading from Machine Learning Engineering with MLflow Manage the end-to-end machine learning life cycle with MLflow

Product type Paperback

Published in Aug 2021

Publisher Packt

ISBN-13 9781800560796

Length 248 pages

Edition 1st Edition

Tools

Maven

Concepts

Machine Learning

Author (1):

Lauchande

View More author details

Table of Contents (18) Chapters

Preface

1. Section 1: Problem Framing and Introductions

2. Chapter 1: Introducing MLflow FREE CHAPTER

3. Chapter 2: Your Machine Learning Project

4. Section 2: Model Development and Experimentation

5. Chapter 3: Your Data Science Workbench

6. Chapter 4: Experiment Management in MLflow

7. Chapter 5: Managing Models with MLflow

8. Section 3: Machine Learning in Production

9. Chapter 6: Introducing ML Systems Architecture

10. Chapter 7: Data and Feature Management

11. Chapter 8: Training Models with MLflow

12. Chapter 9: Deployment and Inference with MLflow

13. Section 4: Advanced Topics

14. Chapter 10: Scaling Up Your Machine Learning Workflow

15. Chapter 11: Performance Monitoring

16. Chapter 12: Advanced Topics with MLflow

17. Other Books You May Enjoy

Setting up a batch inference job

The code required for this section is in the pystock-inference-api folder. The MLflow infrastructure is provided in the Docker image accompanying the code as shown in the following figure:

Figure 9.1 – Layout of a batch scoring deployment

If you have direct access to the artifacts, you can do the following. The code is available under the pystock-inference-batch directory. In order to set up a batch inference job, we will follow these steps:

Import the dependencies of your batch job; among the relevant dependencies we include pandas, mlflow, and xgboost:
```
import pandas as pd
import mlflow
import xgboost as xgb
import mlflow.xgboost
import mlflow.pyfunc
```
We will next load start_run by calling mlflow.start_run and load the data from the input.csv scoring input file:
```
if __name__ == "__main__":
    with mlflow.start_run(run_name="batch_scoring") as run:
    ...
```