学生对 Coursera Project Network 提供的 Using TensorFlow with Amazon Sagemaker 的评价和反馈

This course will serve as a guide on how to use AWS SageMaker. However, there is no technical challenge to it. In my opinion, since the course is marked as difficult the students should be capable of solving some problems on their own.

I would still encourage people to learn about this tool since it can be used to take models into production with a simple API.

The overall project is good.

But May be the coding style should have been better.

For example train.py could have been written in pycharm or even jupyter directly instead of %%writefile.

Writing codes line by line was not a good hands-on experience.

But anyways, it was a quick code-based example for TensorFlow on AWS with SageMaker.

By far the worst Coursera course I've ever done. The interface is barely useable (two windows in one tab), you have to sign up to various accounts including with a credit card for AWS, and the session times out in just over an hour for the two hour course. I want a refund! Avoid.

The instructor is great, he tries his best to explain all steps. However, a short project is not enough to grasp all materials in AWS Sagemaker. You may also need general AWS knowledge. There are details that you need to dig out further for better understanding.

Great project and awesome customization. I got to learn a lot and practice what I learned in this class. Thanks to Amit for teaching this class.

Very Interesting

good

I was glad to find that Tensorflow and Sagemaker were available on Coursera. When working through this project I ran into a couple of snags that weren't addressed in the content and had to go digging in the discussion forums. The first was the Resources limits on AWS regarding the instance type which I needed AWS support to engage on. This created some lag time to finishing the material. The way the course is designed now it assumes that you work through it in one session. Coming back to it creates some issues (such as the model needed to be retrained to be served). This serving portion also didn't work for me out of the box and I'm still trying to find a suitable solution on Inference Endpoints in SageMaker from saved model artifacts on S3.