Product updates

Welcome to our product updates page, where we will share product updates and new features on a regular basis so that you can stay up to date on all tweaks and additions to the platform. Enjoy!

April 27, 2022 /


No more remembering a Peltarion specific set of login and password; you can now use your own Active Directory (AD) as identity provider to access the Peltarion platform!

How it works:

User of organizations with SSO enabled on their own email domain will be able to sign in seamlessly to the platform using their corporate credentials.

Setting it up:

  1. Setup Peltarion as a Registered app in Active Directory by following our knowledge center guide.
  2. Then reach out to with your Tenant ID and Client ID and our support staff will configure it on our side.

From here on, any user with the email domain registered for SSO will be able to sign up and sign in with SSO.

SSO sign in page

March 24, 2022 /

Download inspection results

You can now download your model’s predictions for the training, validation or test set. These are the data points that are available from the Predictions inspections part of the Evaluation module of the platform. This enables you to analyze the behavior of the model on the dataset used in any way that makes sense from your business perspective. 

The predictions are exported as a .csv file which makes it easy to use your favorite analysis tool (e.g., pandas, PowerBI, excel) to slice the data and look deeper into the model performance. It also gives you the flexibility to compare predictions from different models. Right now, this feature is enabled for regression and classification problems. The download is available from the download icon shown below

You have a model for garment price prediction where the scatter plot shows some problematic samples. You want more information. Then download the predictions to look closer at the results using your favorite analysis tool.
Here you can see that the model has problems in the T-shirt category (red dots). This insight provides valuable input on how to get a better model, or simply that you shouldn’t use the model when predicting prices for t-shirts.

January 14, 2022 /

Improved wizard models for tabular data

For tabular data, the modeling wizard currently suggests a model based on the input data. As part of a research project we have identified a better modeling scheme for this. The model proposed is a bit more complicated and requires some more computation time but should result in better model quality. In the picture below is an illustration of the changes made. To the left is the previous and to the right the new model the wizard created for a particular case.

December 15, 2021 /

Peltarion Prediction Server

Bring your platform built models straight to production on your own infrastructure with the Peltarion Prediction Server. This feature allows you to download, build and host your model as a Docker container with just a few steps:

  1. Export your model including the container definition from the deployment page
  2. Execute the Docker build command
  3. Execute the Docker run command

You can then already run inferences on the mode by calling the API and similarly you can deploy this container to any registry and start using it. More info here.

Happy containerizing!

/The Peltarion product team

Peltarion Prediction Server

December 2, 2021 /

Archive experiments

It is now possible to archive and unarchive experiments. When archiving an experiment it will not be available for modeling, evaluation, inspection or deployment. It is however not deleted and by unarchiving it, it is available like before. The reason for archiving experiments is to get a less cluttered experiment list and thereby improve the workflow when you have many experiments. 

Experiments can be archived by selecting the experiment you wish to archive, click on the Experiment options menu (the three dots on the top right corner) and select “Archive”. To restore the archived experiment, select “Unarchive”.

November 29, 2021 /

Active experiment adherance

From now on, you will be able to move seamlessly between evaluation and modelling view and stay within your selected experiment. In other words, if you are evaluating a model and then switch back to modelling view to do something there, the model you are currently working with is the one that will appear in this view.

November 25, 2021 /

Improved Web App results

Brand new result representation in our web app with better support for categorical and binary outputs including listing multiple results for multi-modal outputs.

Try it out yourself on the platform with most models. Just head to the deployment page, enable a deployment and open the web app.

Dramatic poetry in Hindi? The Peltarion Web app thinks so, powered with a multilingual text classification model.

November 9, 2021 /

Outlier handling added to our data cleaning capabilities

The newly added Outlier handling option is available for numerical data inputs and presents users with histogram views of their data, allowing them to identify potential outliers, see the number of outliers per feature, as well as manage outliers by setting valid value ranges (min and max). Values outside of the set range will be removed from the dataset version.

The feature can easily be found via the Data cleaning tab on the Datasets view. Choose dataset features to manage, select appropriate range and click on the Apply changes button to remove your outliers. Read more about outliers and how you can work with them on the Peltarion platform via this link.

October 15, 2021 /

We’ve updated our model download feature

Deploy and serve your platform-built models off platform with our model download option that is now available in tf.savedModel format. The format includes all the pre- and post-processing done by the platform, and is compatible with TensorFlow 2.5.x. 

If you are interested in working with your models off platform, you will be happy to hear that we have created a guide that describes how to download your model as well as provides utility functions to make it easier to work with your models outside the Peltarion platform. You can see how to deploy a SavedModel in these selected frameworks and platforms: ML Flow, TF Serving, UIPath.

You can find the guide here.

October 11, 2021 /

More analytics and monitoring added to the platform

Following the afterlife of your deployed models is a way of making sure that your model is continuously performing well and is also a way of checking that you have made the right decisions when shaping your model, if you need to tune it or potentially replace it.

From now on, whenever you want to monitor API calls, predictions and errors of your live platform deployments, you can do so by visiting the Deployments section of the platform and selecting "Monitoring" in the upper left corner.

September 28, 2021 /

Share your AI projects easily via our Deployment web app

We are very happy to announce that our previous Test deployment link that has been available on the deployment page of the platform has evolved. The new Deployment web app is a feature that allows you to test your deployment from a GUI that you can share with people around you. As soon as you make your deployment public, you are able to share your model through a URL with your coworkers, stakeholders, community, you name it. 

Also, it's responsive. Meaning that you can use it on your phone and get your predictions on the go.

We built an author similarity model for you to test. Check it out: Take me to the app.

If you want to know how the Deployment web app works, have a look here.

If you want to redo our author similarity model, this is the dataset we used.

July 20, 2021 /

Multi-label classification at your fingertips

Could the cats in your dataset look both grumpy and mysterious at the same time? Yes. From now on, you can solve your multi-label classification problems on the Peltarion platform. We have added the IoU (intersection over union) metric to the evaluation view, that will let you evaluate which of your experiments perform the best. We have also added inspection support so that you can inspect how good your model is at predicting a certain label. For instance, is your model good at predicting grumpy cats while struggling with predicting mystery? The Inspection view will tell.

Test the feature through this tutorial

July 20, 2021 /

Introducing our Tune experiment feature

Iterating on (or tuning) experiments is a big part of succeeding in your projects. And with that in mind, we are happy to present you with our Tune experiment feature, a feature that aims to simplify the process of iterating on your projects. By activating the Tune experiment feature, the platform suggests changing the experiment’s settings or the model, and then generates and launches the new experiments from that.

June 21, 2021 /

Our data-cleanser tool is now available on platform

On May 6, we shared the news of our data-cleanser tool with you, a tool that was made with the purpose of allowing users to prepare and cleanse .csv files for an easier data upload onto the platform. At that point, the tool was available off-platform.

Today, we are happy to announce that our data-cleanser tool is available on platform. This means that you can clean and prepare your datasets directly on the platform after they have been uploaded, and:

  • remove the risk of your models seeing missing values
    (ordinarily, missing values could lead to your models crashing)
  • choose different cleaning-strategies between dataset versions. For example, if your missing values are highly concentrated to a specific feature within your dataset, it might be valuable to remove that feature instead of all rows with missing values.

And this is just a start. Going forward, we aim at delivering more options for data-cleaning on platform, such as:

  • Detecting and ignoring outliers
  • Replacing missing values (instead of removing them)
June 28, 2021 /

Introducing Peltarion's new modeling engine - Okra

With the power of our new modeling engine, we will be able to bring you loads of simplified processes and handy features going forward. As of today, Okra displays its value by bringing you a simplified Wizard, the possibility to autofix your models and a tighter iteration cycle.

June 28, 2021 /

Our Datasets view has changed for the better

We love ease of use. And as a result, we have made an update to our dataset view so that the platform is even more intuitive and can cater to different needs.

The new view is perfect for the user that swiftly wants to get to a trained model, since this view takes care of tedious tasks such as versioning the dataset and creating subset splits. When using tutorial datasets, this view will also provide you with a link to the tutorial you are using so that you can easily move back and forth.

May 6, 2021 /

Data-cleanser tool upodates

Last week, we launched a tool for automatically preparing and cleansing your .csv files for an easier data upload onto the platform. Users will no longer need to do their removal of rows/handling of null values off platform.

May 6, 2021 /

Multiple output feature selection for Text similarity

When creating a deployment for Text similarity models, users are now able to select multiple features from their datasets as output features. The below video shows you an example of a model providing similar questions as well as the corresponding answers to an input text.

May 6, 2021 /

Bubble-connector for Text similarity

Following the launch of Text similarity capabilities on the platform, we are happy to announce that we have built a Bubble connector for the feature. With Text similarity, you can build models that compare and find texts that are similar in context and meaning, without them sharing a single common word.

March 24, 2021 /

Text similarity is now available on the platform

From now on, you can build models that find and compare texts that are similar in context and meaning. This means that for a given set of texts, the model can give you a quantifiable measure of how alike they are and give you the best matches in return, and in 100 languages.

To test the feature, go to this tutorial.

January 29, 2021 /

Image similarity is now available on the platform

From now on, you can use the platform for building models that allow you to compare images with each other and give you a score for how visually similar they are.

You can use these models to drive recommender systems that surface similarly looking products, build monitoring tools that track where your images are being used online, and much more.

November 3, 2020 /

Introducing Multilingual BERT on the platform

You can now use the Peltarion platform on texts in over 100 languages.

By added Multilingual BERT to the platform, you are now able to build models that can automatically understand and sort info in documents in the language of your choice. Try our tutorial or read more about Multilingual BERT and what it means for your work.