Frequently Asked Questions

Answers to common questions about licensing, workflows, troubleshooting, and advanced data transformations.

Getting Started

  • Do I need deep ML expertise to use Tvaritam?
    No, Tvaritam abstracts complexity while still allowing advanced users fine-grained control when needed.
  • Which models are supported?
    The current version supports Linear regression, decision tree, random forest, gradient boosting and neural networks for both regression and classification problems.
  • Do I need GPU?
    No, GPU is not essential. However, its presence can significantly augment training speed, particularly of Neural Network, but is not essential.
  • Why software is not using my GPU?
    There are two potential reasons: first, missing or incompatible driver- Tvaritam needs CUDA 12.1 or later to utilize GPU; second, you may be using models which does not require GPU acceleration, such as Trees and their ensembles.
  • Can I use models developed with a trial license after upgrading to a paid version?
    Yes, models created during the trial version remain accessible after upgrading to a paid version.
  • Can I move my license to another machine?
    No, licenses are tied to specific devices for security reasons and cannot be transferred directly. However, if your device is damaged or you plan to switch devices, please email us at support@tvaritam.ai using the email address associated with your purchase, and we will assist with the transfer after verification.

Special Data Transformations

  • How can I impute values for the target (output) feature, similar to input features?
    Imputation of the target variable is generally not recommended. However, in rare cases where it is necessary, it should be guided by domain knowledge. To perform this, temporarily change the feature type to 'Input' in 'Feature Categorization', apply the desired imputation in 'Individual Missing Features', and then revert the feature type back to 'Output'.
  • How can I derive the day of the week from a datetime feature?
    Use 'Convert Datetime' to transform the date into the number of days relative to a reference date that falls on a Sunday. Then, use 'Feature Engineering' to compute the remainder when divided by 7. The resulting feature will represent the day of the week.
  • How can I apply binning based on custom ranges?
    Custom binning is not directly supported. However, it can be implemented using 'Feature Engineering'. Compute the quotient of the feature with respect to the upper limits of each desired bin. Then, set the 'Category' of the newly created features to 'Categorical' in 'Feature Categorization'. Next, use 'Transform Categorical Features' to convert all non-zero levels to 1. Finally, sum the resulting features using 'Feature Engineering' to obtain the binned feature.
  • How can I check the cardinality of continuous features, if needed?
    Temporarily change the 'Category' to 'Categorical' in 'Feature Categorization', then navigate to the 'Analytics Base Table' under 'Exploratory Data Analysis' to view the 'unique' count. Be sure to revert the 'Category' afterward.
  • How can I remove rows with specific values?
    Use the 'Drop the following feature levels' option under 'Transform Categorical Features'. For continuous features, temporarily change their 'Category' to 'Categorical' in 'Feature Categorization', perform the operation, and then revert them back to 'Continuous'.
  • How can I group multiple feature levels into fewer levels?
    Use the 'Replace Feature Level' option in 'Transform Categorical Features' to merge similar levels, and then apply the appropriate encoding method under 'Encoding Categorical'.

Troubleshooting & Errors

  • Why is the loss value NaN during training?
    A NaN loss value typically indicates that the loss has diverged to infinity. The most common cause is an excessively high learning rate, which leads to instability during training. Another possible cause is the presence of extremely large values in the target feature.
  • Why do I get Error code 1Z0002 while loading a model?
    Error code 1Z0002 indicates that your license does not include the encryption key required to decrypt the model. This may occur if you are attempting to load a model created by someone outside your organization.
  • How do I stop training, when UI is unresponsive?
    The response time or UI is increased to ensure computational capacity is effectively utilized for training process. Click on 'Stop Training'; the command will be passed on the next log update and process will terminate after the second log update.
  • How to de-select optional hyperparameters?
    A value equivalent to non-selection of hyperparameter can be entered, such as decay rate of 1 or regularization parameter 0.
  • Why are training and other losses column missing?
    Missing training, CV, or test loss column signifies that loss value for all trained models were infinite (NaN).

Usage & Limitations

  • Is it possible to save the project at an intermediate stage?
    No, this feature is currently not available.
  • Can I run multiple instances of the software simultaneously?
    Yes, you can run multiple instances simultaneously. However, before doing so, ensure that sufficient computational resources are available (e.g., via Task Manager).
  • Can I use preprocessed downloaded data for training?
    Although this is feasible, it is highly discouraged. The preprocessing steps applied to the downloaded data are not included in the saved model; therefore, they would need to be manually repeated for each prediction.

Security & Governance

  • Is my training data accessed or used by Tvaritam?
    No, Tvaritam does not access or use your training data or models in any form.
  • Do I need to take any action to enable model encryption?
    No, models are automatically encrypted before being saved.

Still Have Questions?

Our team is happy to help with specific use cases or challenges.

Request Support