Models Module
The Models module is DashAI's environment for training, evaluating, comparing, and deploying machine learning models. Everything is organized around Sessions — a session groups one or more models trained on the same dataset and task, keeping all results, predictions, and explainability tools in one place.
Key Concepts
- Task — The type of machine learning problem you want to solve. Each task determines which models are available, what column types are valid as inputs and outputs, and which metrics are used to evaluate results.
- Session — A working environment tied to a specific dataset and task. A session can contain multiple models trained under the same conditions, making it easy to compare approaches side by side.
- Model — A specific algorithm added to a session, configured with its own hyperparameters. Multiple models of the same type can coexist in the same session with different configurations.
- Run — Each time a model is trained, it produces a run with its own metrics, predictions, and explainability results.
Available Tasks
When you open the Models module, the main area shows the available task types with a description of each. Use the search bar to filter tasks by name.
| Task | Description |
|---|---|
| Tabular Classification | Predict a categorical label from structured tabular data (rows and columns). |
| Text Classification | Assign predefined categories or labels to text documents based on their content. Useful for sentiment analysis, spam filtering, topic categorization. |
| Regression | Predict a continuous numerical value from structured tabular data. |
| Translation | Convert text from one language to another while preserving meaning and context (NLP task). |
Each task enforces specific requirements on column types for input and output — these are validated automatically when you configure the session.
Sessions
Sessions are listed in the left sidebar under Sessions, grouped by task type. Each session entry shows its name and the dataset it was created from.
Click New Session (the + button in the sidebar) to start creating a new session.
Click any existing session to reopen it and continue working.
The sidebar also shows the count of sessions per task type, making it easy to navigate when working with multiple experiments.
Right Panel — Available Models
When a session is open, the right panel shows the models available for the current task. Each model has a unique icon and name. Hovering over a model card shows a popup with a description of the algorithm, its strengths, and typical use cases.
The panel includes a search bar to filter models by name.
Available models vary by task. For Tabular Classification, for example, the panel includes models such as Support Vector Machine (SVM), Decision Tree, K-Nearest Neighbors (KNN), Logistic Regression, Random Forest, Gradient Boosting, and Dummy Classifier. Other tasks have their own corresponding model catalogs.
Section Structure
This section is divided into the following pages:
- Train a Model — How to create a session, configure input/output columns, define data splits, add models, set hyperparameters, and run training.
- Predictions — How to generate predictions using trained models, both from a full dataset and from manually entered data.
- Explainability — How to use global and local explainers to understand model behavior.
- Model Comparison — How to compare metrics across models using tables and charts.