Deploy & Test Machine Learning APIs on AWS: API Gateway, Lambda, and Jupyter

Key Takeaways

Once you have created the post method with lambda integration, it's a time to deploy the API. We always need an endpoint if we have to call any API. So, we are going to click on deploy API button. It will be asking you, do you want to create a new stage or if you are already having an existing stage, you can make use of it. What is stage? You can relate stage very simply with the environments in the industry. So, basically what happens? Like in industry, we have a concept of environments. We have development environment, staging environment, production environment, pilot prod. The concept of the environment is these are created with the different set of infrastructure for different type of users. For example, if a developer is developing something in the local machine. Now, developer has to provide some common endpoint to others in order to test their work. They can expose their work in the development environment. In a similar manner, production or live environment is for the end users or live users, you can say. In this case, I am creating a new stage, giving a name prod, and saying deploy. So, once it is deployed, I have got the invoke URL. Now, this invoke URL I will be using in order to test my API. As I told you, I will be testing it by three ways. The first way is using this console. The second way is using curl in the cloud shell, and the third one is I will be calling this API from my Jupiter notebook. Okay? So, first thing, let's do a round of testing from the console. How? Go to resources. In resources, you have to select a method. If you will scroll down, you can see you have different sections. Use this arrow, and you will be navigating to test section. Okay? So, here I have to pass the request body, the same which I had taken here. Let me take it. I'm going to paste this request body here. And now, let's click on test button. And let's see what is the result you're going to get. You are getting HTTP status 200, and this is your response body. Predictions zero, class setosa. Okay? The same thing we had seen while we were doing a round of testing using lambda. You can compare it. Come back here, and you can see. Now, the next round of testing we are going to do, that is using cloud shell. So, open cloud shell in new tab. And I'm going to use a curl command in order to do a round of testing. You have to ensure one thing while you are going to use the curl command, everybody. You have to provide the exact URL. Exact URL means this is a sample API invoke URL. You have to replace it with your invoke URL. So, now, from where you are going to get the URL? Let me guide you. Come back to the API Gateway. Either you can go to stages, that is one place. From stages, you can get the invoke URL. Let's take it. And you're going to replace this invoke URL up to prod. Okay? Don't forget to append the resource name, which is the entry point. Take this particular command, and paste it in the cloud shell. And now, let's see the response. Uh let me just quickly correct it. Let's see the error. What is the error it is giving to you? Because we always need to troubleshoot the error if something is wrong. Okay? So, what is wrong here? I can guide you. Let me just paste it again. Let me correct it. Actually, this is not correct. It should be like this. You also need to specify the method type. So, it is curl minus X post, and then you are going to provide all these things. Okay? Just a minute, everybody. Yes. It is already here. You can directly take it from here, test from the cloud shell or local machine. And don't forget to replace the invoke URL. Just take the invoke URL and replace it. As I told you, you have to ensure you have a resource name with you. Let's take this particular command, and execute it in your cloud shell. Okay? So, we had an issue with the syntax. Okay? We were pasting the wrong line of code. That's why we're getting the error. Now, you can see I'm getting the status code as 200, and then I'm got getting a response body. The same thing you had seen in the API console. What is last step, everyone? We want to test it using Python client. So, I am using a notebook, Jupiter notebook, in order to make a round of testing. Okay? So, first of all, don't forget to replace your URL. This is my URL. Let me take it. And then, next step all of you are aware about. You have to open the Jupiter notebook. So, I'm expecting that you all have a setup of client on your local machine. Okay? If you don't have, in my past video lectures, I have already explained you how can you set a Jupiter notebook in your local machine. I'm opening a command prompt, and I have to simply type a command Jupiter notebook. Like this. It's going to open a Jupiter notebook in my browser through my local machine, and then I'm going to invoke this API from Jupiter notebook. Okay? One thing which all of you have to ensure that you have configured the AWS credentials if you want to interact with AWS services. Okay? So, remember, your credentials should be having enough permissions to interact with AWS services. Click on file, new, and notebook. I'm already having the credentials configured on my local machine, so that's why I'm not going to do it again and again. Let me take the piece of code. This is a piece of code which I'm going to use. It's a very simple piece of code. I'm importing the request and JSON. This is a URL. This is a payload. This is a header. Content type application/json, and this is the response. I'm going to run this, and let's see if I'm going to get the response back. Yes, I'm getting the response back. Okay? If you are exploring other tools for API testing, you can also make use of Postman. So, this is something I wanted to show. How can you expose your model via APIs, and we have explored API Gateway as a part of this journey. So, we'll be helping you out with the further discussion on this particular video lecture. So, if I have to tell you a conclusion like what we have achieved exactly, so, we run the trained model on lambda container. We expose the HTTP endpoint via API Gateway. We allow the API Gateway to invoke the lambda function, and then we send the features for prediction using curl or Jupiter notebook. So, in short, we have deployed an end-to-end ML inference API, completely serverless and scalable.