Train a Model
This page walks you through creating a session, adding models, configuring hyperparameters, and running training in DashAI.
Prerequisites
- At least one dataset uploaded to DashAI.
- The dataset must have columns compatible with the task you want to run (e.g., a categorical output column for classification tasks).
Step 1 — Select a Task
In the Models module landing page, click on the task that matches your problem. Each task card shows a description to help you choose the right one.
If you have an existing session, you can also click it directly in the left sidebar to skip this step and resume working.
Step 2 — Create a Session
After selecting a task, a two-step session creation flow begins.
2a — Select Dataset and Name the Session
- Use the Select a dataset dropdown to choose from your available datasets. Once selected, a summary card shows the dataset name, creation date, row count, and column count.
- Enter a name in the Session Name field. A default name is pre-filled
based on the task type (e.g.,
Session_Tabular Classification_1), but you can change it to something more descriptive.
Click NEXT to continue.
2b — Prepare the Dataset
This step has two parts: defining columns and defining the data split.
Input and Output Columns
DashAI validates whether the selected dataset is compatible with the chosen task. A banner at the top of the step confirms compatibility and shows the required column types for inputs and outputs.
- Input Columns — Select the features the model will use to learn. Each column
is shown as a tag with its type badge (Float, Integer, Categorical). Click inside
the field and select columns from the dropdown. Remove any column by clicking the
×on its tag. - Output Columns — Select the target column the model will predict. The output must match the type required by the task (e.g., Categorical for classification tasks).
Each task enforces specific type requirements. For Tabular Classification, inputs can be Float, Integer, or Categorical, while the output must be Categorical with cardinality 1. If the compatibility banner shows a warning, review your column types in the Dataset Explorer before creating the session.
Data Split
Define how DashAI divides the dataset into training, validation, and test subsets. Three options are available:
| Option | Description |
|---|---|
| Use predefined splits | Uses train/validation/test splits already defined in the dataset file. Only available if the dataset was uploaded with pre-split structure. |
| Random split by proportion | Randomly assigns rows to each subset according to the proportions you specify. Default is Train: 0.6, Validation: 0.2, Test: 0.2. |
| Manual split by row indices | Manually specify the start and end row indices for each subset. |
When using Random split, three additional options are available:
| Option | Description |
|---|---|
| Shuffle | Randomly shuffles the rows before splitting. Recommended to avoid ordering bias in the data. Enabled by default. |
| Stratify | Ensures each split preserves the same class proportions as the full dataset. Useful for imbalanced datasets. |
| Seed | A fixed random seed for reproducibility. Default is 42. Set a specific value to ensure the same split is produced every time. |
Click CREATE SESSION to finish setup. The session opens immediately.
Step 3 — Add Models
Once the session is open, the main area shows the Model Comparison panel and the right panel lists the Available Models for your task.
To add a model:
- Click on a model in the right panel. A description popup appears on hover — read it to understand what the algorithm does before adding it.
- Clicking the model opens the Add Model to Session modal.
Configure the Model
The modal has one step: Configure Model.
- Model Name — A unique name for this model instance within the session.
Pre-filled with a default (e.g.,
SVC_1). Change it to something meaningful if you plan to add multiple instances of the same model type. - Model — Shows the algorithm type (read-only, set by your selection).
Hyperparameters
Below the name, each configurable hyperparameter is listed with its current value and a ? help icon. Parameters vary by model — for example:
- An SVM (SVC) has parameters like
C,coef0,degree,gamma,kernel,max_iter,shrinking,tolerance. - A Decision Tree has
max_depth,min_samples_split,criterion, and others. - A KNN has
n_neighbors,weights,metric, and others.
Hyperparameter Optimization
Each numeric hyperparameter has a toggle: Optimize hyperparameter [name]. When enabled, DashAI will automatically search for the best value for that parameter during training instead of using the fixed value you entered. You can enable optimization for any combination of parameters.
Click ADD MODEL to add it to the session. Click CANCEL to discard.
You can add as many models as needed — including multiple instances of the same algorithm with different configurations. Each appears as an independent row in the comparison table.
Step 4 — Train Models
Once models are added, they appear in two places:
Comparison Table (top)
A summary table showing all models in the session with columns:
- Model Name — the name you assigned.
- Model — the algorithm type.
- Actions — three buttons per row: ▶ run, 👁 view details, 🗑 delete.
Model Cards (below the table)
Each model has an expandable card showing its name, algorithm, current status badge, and action buttons:
| Button | Description |
|---|---|
| EDIT | Reopens the configuration modal to change the model name or hyperparameters. |
| TRAIN | Starts training this model. Changes to RE-TRAIN after the first run. |
| Status badge | Shows the current state: Not Started, Finalizado (completed), or Error. |
| 🗑 | Deletes the model from the session. |
To train a single model: click TRAIN on its card.
To train all models at once: click RUN ALL in the top right of the comparison table header. This queues all untrained models for sequential execution.
Training runs as a background job — you can monitor progress in the Job Queue at the bottom right of the screen.
Step 5 — Review Results
After training completes, each model card shows a Finalizado badge and expands to show four tabs:
LIVE METRICS
Shows the model's performance metrics organized into three sub-tabs: TRAINING, VALIDATION, and TEST.
A Metrics dropdown lets you select which metric to display. Available metrics depend on the task — for classification they include Accuracy, F1, Precision, Recall, ROCAUC, LogLoss, HammingDistance, CohenKappa. For regression they include RMSE, MAE, and others.
Metrics are only available after training is complete. If a model shows "No metrics available for this view", it has not been trained yet.
EXPLAINABILITY
Shows global and local explainers attached to this model. See the Explainability page for details.
PREDICTIONS
Shows dataset predictions and manual predictions generated from this model. See the Predictions page for details.
HYPERPARAMETERS
Shows the exact hyperparameter values used in the last training run. Useful for verifying which values were selected when hyperparameter optimization was enabled.
Tips
- Add a Dummy Classifier (or equivalent baseline model for your task) alongside your main models. It gives you a performance baseline to compare against — any model that doesn't beat the dummy needs more tuning.
- Enable Stratify when working with imbalanced datasets to ensure each split has a representative sample of all classes.
- Use a fixed Seed when comparing multiple models to ensure they all train and evaluate on exactly the same data split.
- You can add multiple instances of the same model type with different hyperparameter configurations to explore how parameters affect performance.
Troubleshooting
| Symptom | Likely cause | Solution |
|---|---|---|
| Compatibility banner shows a warning | Column types don't match task requirements | Check column types in the Dataset Explorer and re-upload if needed |
| NEXT button is not active in session setup | Required fields are empty | Ensure a dataset is selected and a session name is entered |
| Model shows Error badge after training | Invalid hyperparameter values or data issue | Click EDIT to review parameters, or check the Job Queue for error details |
| No metrics available after training | Model trained on incompatible data | Review input/output column selection and retrain |
| RUN ALL is not visible | No models have been added yet | Add at least one model before using RUN ALL |