Building a TensorFlow Pipeline

Creating a TensorFlow training job

In this lesson, we will build a TensorFlow model from scratch.

Step 1: Create a high-compute environment

We will use the compute we created in the previous lesson. If we need high computation, we can use compute GPUs instead.

Step 2: Choosing the right environment

Depending on the library and CPU/GPU availability, we have two options:

1. Use a built-in environment, if available. We can find the list of environments in Azure Machine Learning studio.

We'll use the environment AzureML-tensorflow-2.4-ubuntu18.04-py37-cpu-inference:9, which contains the TensorFlow library and supports the CPU.

2. If the desired environment is unavailable, we can use a custom environment or the Docker environment.

Step 3: Define the hyperparameters

If we want to use any hyperparameters, we define them in the input of the job configuration. We then pass them as arguments to the model from the command configuration. We can use them in the code as desired. For example, we pass epochs as the parameter from the job command ...