How to deploy your ML model with SmartPredict?

AI can solve problems related to real business challenges if only the trained machine learning model is put into production or actively used by customers. Thus, deploying an AI model is a crucial step not to be overlooked in the life cycle of a machine learning project.

With SmartPredict, we have brought ingenuity to how you put your model into production. We aim to make model deployment more accessible, faster, and effortless. You don't need to hire a whole team of software engineers or reinvent the technology infrastructure. Yes, whatever your background, whether you are a data scientist or a relatively non-technical person, you can do it successfully. The resulting advanced features make you more efficient and versatile.

In this blog post, you'll learn how to deploy a model with SmartPredict and discover how you can leverage this ingenuity.

Basically, how to deploy the ML model with SmartPredict?

Deploying an AI model is a simple drag-and-drop process with the SmartPredict platform. The main actions included are:

- Drag and drop built-in or your CUSTOM modules,

- Configure and link them to form a deployment flowchart.

- One-click: fully deploy the deployment flowchart as a web service.

- Test and update it if necessary.

That's it! There is no need to worry about software infrastructure and development operations; everything is done smoothly, and seamlessly and the most essential point is that you are in full control of it.

Let's dive into the 4 steps of deploying an ML model with SmartPredict.

Building the deployment flowchart to deploy trained ML model

When you have completed the training and evaluation of the model in the "Build" space: Go to the "the Deploy" space, where you build a flowchart that will be fully deployed as a web service to put the trained model into production.

Nothing is complicated about it as long as you know the processes by which your trained model will ingest data and provide the prediction. 

Each step will be presented and executed by a module. 

Drag and drop the process from the Module menu, or customize it in Python code with the Custom Module.

Don't worry. Here are the foremost commonly used modules that make up your deployment flowchart:

• The Web Service IN module: 

All deployment flowchart begins with this module.

It’s some kind of gateway that ingest data from other environments into the flowchart. It can ingest data from any data source, in the form of the JSON format and output it in dictionaries or data frame format.

• DataFrame Loader/Converter module:

It converts your dictionaries data into DataFrame.

• Saved data processing module: 

Be aware that a trained model can only provide predictions on the basis of data in the same format as that in which it was trained. So your deployment flowchart should have data processing too and you can reuse all the data processing processes during the model training with your saved data processing module.

• You can Customize your module:

You can also create your custom modules with Python code and add them to your deployment flowchart.

• Feature Selector Module:

It selects features to make the prediction.

• Saved trained Module:

You can retrieve your trained model as a module.

• Model Predictor Module:

It has as input the saved trained model module and the feature to deliver the prediction.

• Web Service Out Module: 

All deployment flowchart ends up with this module which returns the prediction value in the Web Service output.

Tips: You can know which input and output type the module required by holding on to the module's input and output. 

Deploying the whole flowchart as a Web Service in one click

Just click on the Rocket icon to fully deploy the whole pipeline in a minute, and an API is generated instantly. You can choose between 2 options mode: Server or Serverless.

Testing, debugging and updating the deployed pipeline at any time

Of course, you can test and debug your deployment flowchart when you have deployed it with the rocket icon. For this:

• First: Send data to the Web Service in the Predict Space by clicking the arrow button and getting a prediction.
• Second: If you get an error ( no prediction received), maybe there is an issue on your deployment flowchart, and you should edit it. So go back to the Deploy space, see logs and use Data/Object/Type Logger modules in any module output to track more of the module’s output information. Edit and click on the Rocket icon to update the deployment flowchart and make a test again.
• And proceed like this repeatedly as long as you receive a prediction.

Powering up app and software with the generated API

Well, when the web service returns a prediction to you by running a test in the Predict space, that means you can integrate it into production in another IT environment. So, get your API generated in the Monitor space and there's a piece of code to show how to use it.

And what else? Create a serverless AI application.

Now that you know the principle of deployment: drag and drop modules, the ability to customize functionality with Python, and deploying the entire pipeline in one click, you should know that you can create a serverless application with the SmartPredict Deploy space without going through the training of an ML model.

Conclusion

With SmartPredict, we continuously add new functionality that makes performing AI projects easier. 

We have seen in this blog post how easy model deployment is with its drag-and-drop workspace. It minimizes the tedious task of coding and all software engineering processes: we mainly drag and drop modules and put the whole pipeline into We Service in one click. 

To learn more about the platform, feel free to visit its website and youtube channel, and it's free to use it for an end-to-end AI project.