Build, train, and deploy models - detailed workflow

Building, training, deploying and iterating on your models is extremely easy on the Peltarion Platform by design. You can go from data to a trained model in just a couple of clicks! Let’s review the main steps of the workflow.

Create a Project → Upload your data → Configure your dataset → Build a model → Train a model → Evaluate your model → Deploy your model.

Step 1 - Create a Project

In the Project view, click the New Project button. Name the project, add a description and click Submit.


Step 2 - Upload your data

After creating the project you will automatically land in the Dataset view. Here you can upload your existing datasets, import data from your cloud storage, or use one of our ready-made datasets from our Data Library.

If this is your first time using the platform, we highly recommend that you use the MNIST - tutorial data dataset from the Data Library and follow this tutorial.


Step 3 - Configure your dataset

Once the dataset has loaded onto the platform it will visualize each of the features of the dataset. If needed, you can change the feature encoding, group features into feature sets, and/or distribute the examples of the dataset into subsets (other than the predefined one by the platform).


Once you’re done editing your features, make sure you save your dataset version by clicking on the Save version button in the upper right corner of the Dataset view. You won’t be able to progress to the next step if you don’t do this.

Step 4 - Build a model

After saving your dataset version (seriously, don’t forget to do this, it’s important), you can move on to building your model on the Modelling view. To get started, click the [.guilabel]#New experiment button and the Experiment wizard will appear.

Experiment view

The wizard will give you automatic recommendations of which of the prebuilt models that are included on the Peltarion Platform best fits your data and the problem you’re trying to solve. Follow the instructions of the wizard and your model will be ready to be trained.

Alternatively, you can skip the wizard and build a model from scratch simply by dragging & dropping individual building blocks on the canvas of the visual editor.

Modeling view

In this case, you will have to also configure the parameters of each of the blocks of the model yourself, so this is only something we would recommend trying if you already have a background in AI.

Step 5 - Train a model

After building / configuring your model, you’re ready to start training it. If needed, you can adjust the training settings by clicking on the Settings tab in the Inspector.

Run Settings

Click the Run button to start training your model.

Run button

Step 6 - Evaluate your model

While the model is training you can keep an eye on how well your it’s learning from the data in the Evaluation view.

Evaluation view

After the model is done training you can also check how it actually performs on individual samples, to gain even more insights about your model’s performance.

Test Deployment

For a complete list of the tools available to you in the Evaluation view, what they tell you and how to use them, make sure to read the many articles we have on this topic on the Knowledge Center.

From here you have two options:

  • If you’re unhappy with your model’s performance, you can go back to any of the previous steps and try different strategies to improve it. For example, try a different model from the list of suggested models in the Experiment wizard, train it and compare the performances of both models.

  • If you’re happy with your model’s performance, you could proceed to the last step and deploy it, to make it available to use in an application (see next step).

Step 7 - Deploy your model

The last step is to make your model available to use in an application of your choosing. This is done in the Deployment view.

Deployment view

What do you need to do to deploy your model? Just toggle the Enable switch. That’s it, it’s that easy! Your model is now available to the world.

Once the model is deployed it will be accessible through the Peltarion Deployment API for forward pass queries, but more on that in the next section.

And that’s it. These 7 steps cover the basic workflow of any project that you will ever develop on the Peltarion Platform.

Naturally, there are tons of extra configuration possibilities and tips & tricks that we have left out in this quick start guide, so make sure you have a look at the many articles about the platform in our Knowledge Center to find all the details you need.

Was this page helpful?