Experiment tracking is very important steps when it comes to model deployment in production. When we make model ready for deployment we compare the performances of different recorded experiments and check which one is the optimal one. That is where it becomes very much important to understand how to do experiment tracking for different models trained under different machine learning frameworks. In this post I am explaining step by step approach to do experiment tracking for TensorFlow Based image classification model.
There are two ways to record experiments for Tensorflow models.
- using mlflow.tensorflow.autolog method
- manually record what you want to log
After going through this post you will have fair understanding on both the methods and then decide which one to use for your use case.
Below are the main steps covered in the complete explanation:
- 00:00 Introduction
- 04:40 TensorFlow Model Training for MNIST image classification
- 09:53 Experiment Tracking using autolog method
- 17:55 Experiment Training using custom method
- 22:32 Load model from MLFlow
- 23:49 Do Prediction
You know video explanation is always good than reading a blog post by yourself. Hence I have created a live demo explaining every step in detail, which I will suggest you to go through and execute together by pausing the video. This will give better understanding of each steps and see if you getting any error.
So, if you are new to tensorflow modelling, you can also learn how to train a basic image classification model using tensorflow and further do the experiment tracking for the same model.
Just go through the video, and let me know if you face any error while executing. I would be happy to respond to your queries.
That’s it for this article, see you in the next one !
Data Science Duniya 