Machine Learning Lab
Ivanti Neurons Machine Learning Lab provides you with a centralized environment to create, train, and deploy machine learning models. Currently, these models can be used to power Neurons for ITSM Ticket Classification, but more applications will follow. The models are created from within the Ivanti Neurons Platform and remain there even when deployed.
With Machine Learning Lab, you no longer need in-depth knowledge about machine learning models.
You can find the Ivanti Neurons Machine Learning Lab at Software > Machine Learning Lab.
For a successful implementation of a machine learning model, you will have to prepare the following:
- Training data: a set of examples where the combination of input and output are known to be correct.
- Stopwords (optional): a list of words that the model must ignore. This can be used to improve model performance.
Please note that a default set of stopwords from NLTK is applied automatically.
For more information about preparing your dataset and training a machine learning model, see Getting Started with Ticket Classification.
Currently, Machine Learning Lab supports only machine learning models for Neurons for ITSM Ticket Categorization, and the English language.
The Machine Learning Models table shows a list of existing machine learning models and their properties.
Click the Model Name for more detailed information about the model.
Below is a list of possible actions related to machine learning models.
For a step-by-step guide to configure Ticket Classification, see Getting Started with Ticket Classification.
Add a Machine Learning Model
To add a new model, you first have to export and prepare training data from the system you will deploy the model to. Optionally you can add a file with stopwords, to fine tune the model.
For more information about preparing this data, see Getting Started with Ticket Classification, step 3. Cleaning training data.
- Click Add Model.
The Add Model pane opens. - Specify a Name for the model and select the desired Type and Language.
Currently, the fields Type and Language have only one option and cannot be changed. - Select a CSV file containing Training Data.
The CSV file must not contain a period in the Service column. - Optionally, Select a CSV file containing Stopwords.
- Save your changes. You can also immediately start training your model by selecting Save and Train.
The Add Model pane closes.
You can now click 
to select your next step.
Edit a Machine Learning Model
Depending on the Status of the model, the option to Edit may not be available.
- Click on a model's name in the Machine Learning Models table or click and select Edit.
The edit pane for the model opens. - On the Model data tab of the pane you will find information about the configuration of the model. You can edit this data if needed.
- Save your changes. You can also immediately start training your model by selecting Save and Train.
The pane closes.
Train a Machine Learning Model
Depending on the Status of the model, the option to Train may not be available.
- Select Save and Train from the model's edit pane or, from the model list view, click and select Train.
- Machine Learning Labs loads the training data and stopwords you provided, and starts training.
After a model has been trained, you can find its accuracy on the model's Details tab (click and select View).
Accuracy is expressed in two numbers:
- Training Accuracy: the percentage of correct classifications that was achieved during training.
- Test Accuracy: the percentage of correct inferences on samples from the training data, that were not used in training.
These percentages can help you decide what the best model is for your environment.
You can have multiple trained models in your Machine Learning Lab.
If Machine Learning Lab encounters a problem when it attempts to train the model, the status changes to Training failed. If this occurs, click , select View and go to the Details tab of the model for more information.
When a new class is added or an existing class is deprecated in your Neurons for ITSM environment, you must train a new model with an updated set of training data. See also When to train a new model.
Deploy a Machine Learning Model
To deploy a model means that data from newly created incidents starts being processed by the model.
The model itself remains in the Machine Learning Lab.
To deploy a model to your Neurons for ITSM environment, click and select Deploy. You can deploy only models that have the status Trained.
If another model was already deployed when you deploy a model, the status of the earlier model is set back to Trained. The new model will then be deployed. During this process, newly created incidents are not classified.
After deployment has finished, classification will (re)commence for incidents that are created from that moment onwards.
After a model has been deployed, you can find the following information on the model's Details tab (click and select View):
- Number of API calls: The number of times the model was called to predict the class for an incident since deployment.
- Average Top-Class Probability: When the model is called to predict the class of an incident, the class with the highest probability according to the model is proposed to Neurons for ITSM.
This value shows the average probability of the proposed top-classes for incidents since the model was deployed.
Currently, the only way to 'un-deploy' a model is to delete it.
If Machine Learning Lab encounters a problem when it attempts to deploy the model, the status changes to Deployment failed. If this occurs, click , select View and go to the Details tab of the model for more information.
When a new class is added or an existing class is deprecated in your Neurons for ITSM environment, you must train a new model with an updated set of training data. See also When to train a new model.
Delete a Machine Learning Model
From the model list view, click on the line of the model you want to delete, and select Delete.
If the model was deployed, newly created incidents will no longer be classified.