User Guide

User Guide

The application is designed with easiness in mind. It, therefore, tries to use a straightforward approach to its usability.

How to make a new model?

In the menu on the top of the pages click on "Model Builder" to go to an overview to create a new model. This will show you all the models and their status.

Here the table has the following information:

• Slug --> A short, easy and unique way to find your model. This is basically the uniquely identifiable model name.
• Model --> The type of model that is used. e.g. Neural Network, Random Forests and Decision Tree.
• Property to Predict --> The type of property that is being predicted. e.g. Solubility
• Version --> If the model has a version it'll show the version here.
• Timestamp --> The time when the model was created.
• Status --> The status of the model.

A model can have the following status:

• NEW --> A model instance has been created in the data.
• UPLOADED --> A dataset has been uploaded for the model
• TRAINED --> The model has been trained and saved to the system.
• TESTED --> The model has been evaluated and stored its RMSE and MAE values in the database.
• SAVED --> The model has been published and is accessible throughout the rest of the system.

To create a new model you press the "Create" button on the top right of the page. Here you:

• Select the type of model that you want to use e.g. Random Forests
• Select the property to predict e.g. Solubility
• Optionally input a version. If not version is given, it'll use the Timestamp for the creation of the slug.

Once the model is created you can click on the pencil of that model in the overview. Here you need to follow the steps that are on the screen:

• First upload a file by clicking "Browse" in the dataset field and select the file that you want to import (As of 2023-01-25 it must be a .csv file).
• Optionally give a name. Original file name is used (or hashed variant) in any case for now. Thus this can be ignored.

If any of the required fields are missing, you'll receive an error message that will tell you the missing fields. If it uploaded the dataset successfully, it'll bring you to the next step.

This step is to train the dataset. You can tweak some parameters if you'd like, e.g. the random seed or train/test split. When done you can press the "Train" button which will train the network. This may take a few minutes, depending on the amount of data and efficiency of the model.

Once training has been completed, you'll see a message that says "No Performance Results known". Click on the button "Get Performance Results" to evaluate the model and retrieve the RMSE and MAE values.

Now your model has been trained and evaluated, and it's ready to be published. Click the "Publish" button and after that it'll show up in the rest of the system.

You have now created a new model that is usable within the system.

How to make an ensemble model?

How to make a prediction?